ILMA - Volume V - Neurological Disorders and Imaging Physics - App

Neurological Disorders and
Imaging Physics, Volume 5

Applications in dyslexia, epilepsy and Parkinson’s
Neurological Disorders and
Imaging Physics, Volume 5
Edited by
Ayman El-Baz
University of Louisville, Louisville, KY, USA
Jasjit S Suri
AtheroPoint, Roseville, CA, USA
IOP Publishing, Bristol, UK

ª IOP Publishing Ltd 2020
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system
or transmitted in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, without the prior permission of the publisher, or as expressly permitted by law or
under terms agreed with the appropriate rights organization. Multiple copying is permitted in
accordance with the terms of licences issued by the Copyright Licensing Agency, the Copyright
Clearance Centre and other reproduction rights organizations.
Permission to make use of IOP Publishing content other than as set out above may be sought
at permissions@ioppublishing.org.
Ayman El-Baz and Jasjit S Suri have asserted their right to be identified as the authors of this work
in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
ISBN 978-0-7503-2723-7 (ebook)

ISBN 978-0-7503-2721-3 (print)
ISBN 978-0-7503-2724-4 (myPrint)
ISBN 978-0-7503-2722-0 (mobi)
DOI 10.1088/978-0-7503-2723-7
Version: 20200701
IOP ebooks
British Library Cataloguing-in-Publication Data: A catalogue record for this book is available
from the British Library.
Published by IOP Publishing, wholly owned by The Institute of Physics, London
IOP Publishing, Temple Circus, Temple Way, Bristol, BS1 6HG, UK
US Office: IOP Publishing, Inc., 190 North Independence Mall West, Suite 601, Philadelphia,
PA 19106, USA
With love and affection to my mother and father, whose loving spirit sustains me still
Ayman El-Baz
To my late loving parents, immediate family, and children
Jasjit S Suri
Contents
Preface xiv
Acknowledgements xv
Editor biographies xvi
List of contributors xvii
1 Majority vote of machine learning methods for 1-1

real-time epileptic seizure prediction applied on
EEG pediatric data
Medina Bandic, Jasmin Kevric, Dino Keco and Samed Jukic
1.1 Introduction 1-1
1.1.1 Epilepsy and epileptic seizure 1-2
1.1.2 EEG signals related to epileptic seizures 1-3
1.2 Literature review 1-3
1.3 Methodology 1-7
1.3.1 Data and data collection 1-7
1.3.2 EEG signal processing techniques 1-7
1.3.3 Meta heuristic methods 1-10
1.4 Experimental setup 1-12
1.4.1 Data setup 1-12
1.4.2 Data preparation with de-noising and feature extraction 1-13
1.4.3 Training the data—Matlab GUI 1-15
1.4.4 Data testing, processing and seizure prediction 1-16
1.5 Results 1-17
1.6 Discussion 1-22
1.7 Conclusion 1-28
References 1-28
2 Delineation of epileptogenic zone 2-1

João Paulo Sant Ana Santos de Souza, Brunno Machado Campos,
Felipe Franco da Graça, Ricardo Brioschi, Larissa Núbia Nunes Vilany,
Wu Shin-Ting and Enrico Ghizoni
2.2 Seizure semiology 2-2
2.3 Scalp electroencephalography (EEG) and long-term 2-4
video-EEG (VEEG) monitoring
vii
Neurological Disorders and Imaging Physics, Volume 5
2.3.1 Introduction 2-4

2.3.2 Terminology 2-4
2.3.3 Applications to epilepsy 2-6
2.3.4 Long-term video-EEG monitoring 2-8
2.4 PET-CT 2-8
2.4.3 Conclusion 2-12
2.5 Magneto-electroencephalography (MEG) 2-12
2.5.2 Applications and limitations 2-12
2.6 Single photon emission computed tomography (SPECT) 2-13
2.7 Functional MRI (fMRI) 2-17
2.7.2 Application to epilepsy 2-18
2.8 Subdural grids (SBG) 2-19
2.8.3 Complications 2-21
2.8.4 Conclusions 2-22
2.9 Stereoelectroencephalography (SEEG) 2-22
2.9.2 SEEG versus SBG 2-23
2.9.3 Patterns of explorations 2-24
2.9.4 Technique description 2-26
2.9.5 Surgical outcomes 2-28
2.9.6 Conclusions 2-29
2.10 Illustrative case 2-29
2.11 Conclusion 2-31
References 2-31
3 Dyslexia: behavioral and biological correlates and treatment 3-1

Kathleen H Nielsen, Todd L Richards and Virginia W Berninger
3.2 Definitional issues 3-2
viii
3.3 Phenotypes (behavioral markers) and genetic bases of dyslexia 3-3

3.3.1 Phenotypes 3-3
3.3.2 Genetic findings 3-3
3.4 Assessment of dyslexia 3-4
3.4.1 Family history and neurodevelopmental history 3-4
3.4.2 Educational history 3-4
3.4.3 Language 3-4
3.4.4 Motor 3-5
3.5 Phenotypic markers 3-5
3.5.1 Verbal comprehension 3-5
3.5.2 Assessing word level reading and spelling skills 3-5
3.5.3 Assessing verbal working memory components 3-5
most often impaired in dyslexia [5]
3.6 Brain imaging behavioral and brain correlates 3-6
3.6.1 Brain imaging methods 3-6
3.6.2 Example brain imaging tasks linked to behavioral 3-7
phenotypes outside scanner
3.6.3 Brain imaging before and after treatment 3-9
3.7 Future research directions 3-9
3.8 Assessment–instruction links for treatment 3-9
3.8.1 Educational and clinical applications 3-9
3.8.2 Biological and environmental risk factors 3-9
References 3-14
4 Cholesterol and oxidized cholesterol derivatives: 4-1

potential biomarkers of Parkinson’s disease?
Thomas Nury, Gerard Lizard and Anne Vejux
4.2 Parkinson’s disease 4-3
4.3 Cholesterol and Parkinson’s disease 4-5
4.4 Oxysterols and the brain 4-7
4.5 Oxysterols in Parkinson’s disease 4-9
4.6 Potential involvement of oxysterols in mechanisms 4-11
implicated in Parkinson’s disease
4.7 Conclusion 4-13
References 4-14
ix
5 Computer assisted diagnosis of gait dynamics 5-1

neurodegenerative diseases using a machine learning approach
J Prasanna, S Thomas George and M S P Subathra
5.2 Methodology 5-3
5.2.1 Gait dataset 5-3
5.2.2 Feature extraction 5-3
5.2.3 Artificial neural network (ANN) 5-7
5.3 Results and discussion 5-8
5.4 Conclusion 5-11
References 5-11
6 Rett syndrome 6-1

Ola H Skjeldal
6.1 Background 6-1
6.2 Clinical features 6-3
6.2.1 Congenital Rett syndrome 6-4
6.2.2 Early onset seizure variant (Hanefeld variant) 6-4
6.2.3 Preserved speech variants 6-5
6.2.4 Male variants 6-5
6.3 Important clinical symptoms and signs in Rett syndrome 6-5
6.3.1 Seizures 6-5
6.3.2 Respiratory and cardial disturbances 6-6
6.3.3 Sleep 6-6
6.3.4 Gastrointestinal disturbances 6-6
6.3.5 Communication 6-6
6.4 Genetic basis of Rett syndrome 6-7
6.5 Conclusions 6-7
References 6-9
7 Knowledge about epilepsy in university health students 7-1

Natalia Cristina de Oliveira Vargas e Silva and Priscila de Souza
7.2 Cultural differences in knowledge of epilepsy 7-3
7.3 Assessment of knowledge 7-5
7.4 Future perspectives 7-5
References 7-7
x
8 Current methods and new trends in signal processing 8-1

and pattern recognition for the automatic assessment
of motor impairments: the case of Parkinson’s disease
Juan Rafael Orozco-Arroyave, Juan Camilo Vásquez-Correa and Elmar Nöth
8.2 Clinical assessment of the disease 8-2
8.2.1 General symptoms and common tools for neurological 8-2
evaluation of PD
8.2.2 Clinical evaluation of dysarthria 8-4
8.2.3 The modified Frenchay dysarthria assessment (m-FDA) 8-5
8.3 Automated analysis of the disease 8-6
8.3.1 Motor signals for the study of PD 8-6
8.4 Revealing features for the automatic diagnosis and monitoring 8-12
of Parkinson’s disease
8.4.1 Speech 8-13
8.4.2 Gait 8-18
8.4.3 Handwriting 8-21
8.5 Existing data 8-22
8.5.1 Speech 8-23
8.5.2 Gait 8-24
8.5.3 Handwriting 8-25
8.5.4 The extended multimodal PC-GITA corpus 8-25
8.6 Automatic analysis of the extracted features 8-28
8.6.1 Performance metrics 8-29
8.7 Experiments and results 8-31
8.7.1 Experiments with speech signals 8-32
8.7.2 Experiments with gait signals 8-37
8.7.3 Experiments with handwriting signals 8-40
8.7.4 Fusion of modalities 8-42
8.8 Discussion 8-45
8.9 Future trends in PD analysis 8-46
References 8-49
9 ‘They labelled me ignorant’: the role of neuroscience 9-1

to support students with a profile of dyslexia
Stephen Camilleri, Deborah Chetcuti and Ruth Falzon
9.2 Method 9-3
xi
9.2.1 The participants 9-4

9.2.2 Data collection and analysis 9-4
9.3 Bridging the gap: linking neuroscience and educational research 9-5
9.3.1 The dyslexia debate: is dyslexia a real learning difficulty 9-5
or is it a myth?
9.3.2 Paying attention to early diagnosis, support and intervention 9-8
9.3.3 ‘They labelled me ignorant’—Examinations and dyslexia 9-12
9.4 Final reflections 9-15
References 9-16
10 Advances in epilepsy: from gender to genetics 10-1

Suchitra Joshi and Denia Ramirez-Montealegre
10.2 Minimally invasive techniques used to achieve seizure control 10-2
10.3 Rhythms in seizure frequency 10-3
10.3.1 Hormonal regulation of seizures 10-3
10.3.2 Catamenial epilepsy 10-4
10.3.3 Progesterone 10-4
10.3.4 Estrogen 10-8
10.4 Genetic epilepsies 10-10
10.4.1 GABAAR associated epilepsies 10-11
10.4.2 Glutamate receptors associated epilepsies 10-14
10.4.3 Voltage-gated sodium channels associated epilepsies 10-15
10.4.4 Voltage-gated potassium channels associated epilepsies 10-17
10.4.5 Voltage-gated calcium channels associated epilepsies 10-18
10.4.6 Nicotinic acetyl-choline receptor associated epilepsies 10-20
10.5 Conclusions 10-21
References 10-21
11 Neuroimaging in Parkinson’s disease 11-1

Pearl S Rodrigues, Pravin P Kale and Lokesh K Bhatt
11.2 Neuroimaging biomarkers in Parkinson’s disease 11-2
11.3 Molecular imaging of Parkinson’s disease 11-2
11.3.1 Dopamine 11-3
11.3.2 Serotonin 11-3
11.3.3 Cholinergic dysfunction 11-4
11.3.4 Noradrenergic function 11-4
xii
11.4 Imaging midbrain structural changes in PD 11-4

11.4.1 Magnetic resonance imaging techniques 11-4
11.4.2 Transcranial sonography (TCS) 11-5
11.5 Positron emission tomography (PET) 11-6
11.5.1 Presynaptic and postsynaptic dopaminergic imaging 11-6
11.5.2 Assessment of cerebral glucose metabolism 11-6
11.6 Single photon emission computed tomography in 11-7
Parkinsonian disorders (SPECT)
11.7 Diffusion tensor magnetic resonance imaging (DTI) 11-8
11.8 Proton magnetic resonance spectroscopy (1H-MRS) 11-8
References 11-9
xiii
Preface
This book covers the state-of-the-art automation of non-invasive approaches for the
early detection of neurological disorders, focusing on dyslexia, epilepsy and
Parkinson’s. Dyslexia is a brain disorder that is associated with a disability in
reading, which affects both the behavior and the learning abilities of children.
Epilepsy is a brain disorder in which clusters of nerve cells, or neurons, in the brain
sometimes signal abnormally. This can cause epileptic seizures, episodes that can
result in physical injuries. Parkinson’s is a neurological disorder that affects
dopamine-producing neurons in the substantia nigra brain region. This develops
symptoms that progress over time, e.g. tremor, slowness of movement, limb rigidity,
and balance problems.
Recent advances in neuroimaging techniques have enabled imaging of different
brain structures and correlating the results to clinical findings, which can help in the
early diagnosis of these neurological disorders. This will be covered in the book.
Among the topics discussed in the book are: real-time epilepsy prediction applied on
EEG pediatric data; delineation of epileptogenic zone; behavioral and biological
correlates and treatment of dyslexia; potential biomarkers of Parkinson’s disease;
Rett syndrome; and automatic assessment of motor impairments for Parkinson’s
disease.
In summary, the main aim of this book is to help advance scientific research
within the broad field of automated, non-invasive and early diagnosis of dyslexia,
epilepsy and Parkinson’s neurological disorders. The book focuses on major trends
and challenges in this area, and presents work aimed to identify new techniques and
their use in biomedical analysis.
Ayman El-Baz
Jasjit S Suri
xiv
Acknowledgements
The completion of this book could not have been possible without the participation
and assistance of so many people whose names may not all be enumerated. Their
contributions are sincerely appreciated and gratefully acknowledged. However, the
editors would like to express their deep appreciation and indebtedness particularly to
Dr Ali H Mahmoud and Islam Abdelmaksoud for their endless support.
Ayman El-Baz
Jasjit S Suri
xv
Editor biographies
Ayman El-Baz
Ayman El-Baz is a Distinguished Professor, University Scholar, and
Chair of the Bioengineering Department at the University of
Louisville, Kentucky. Dr El-Baz earned his BSc and MSc degrees in
electrical engineering in 1997 and 2001, respectively. He earned his
PhD in electrical engineering from the University of Louisville in
2006. In 2009, Dr El-Baz was named a Coulter Fellow for his
contributions to the field of biomedical translational research.
Dr El-Baz has 17 years of hands-on experience in the fields of bio-imaging modeling
and non-invasive computer-assisted diagnosis systems. He has authored or
coauthored more than 500 technical articles (151 journals, 29 books, 80 book
chapters, 255 refereed-conference papers, 173 abstracts, and 35 US patents and
Disclosures).
Jasjit S Suri
Jasjit S Suri is an innovator, scientist, visionary, industrialist and an
internationally known world leader in biomedical engineering.
Dr Suri has spent over 25 years in the field of biomedical
engineering/devices and its management. He received his PhD from
the University of Washington, Seattle and his Business
Management Sciences degree from Weatherhead, Case Western
Reserve University, Cleveland, Ohio. Dr Suri was crowned with
President’s Gold medal in 1980 and made Fellow of the American Institute of
Medical and Biological Engineering for his outstanding contributions. In 2018, he
was awarded the Marquis Life Time Achievement Award for his outstanding
contributions and dedication to medical imaging and its management.
xvi
List of contributors
Medina Bandic
Francuske Revolucije bb, Ilidza, Sarajevo, Bosnia and Herzegovina
Jasmin Kevric
Dino Keco
Samed Jukic
João Paulo Sant Ana Santos de Souza
Neuroimaging Laboratory (LNI), Department of Neurology, University of Campinas,
São Paulo, Brazil
Brunno Machado Campos
São Paulo, Brazil
Felipe Franco da Graça
São Paulo, Brazil
Ricardo Brioschi
São Paulo, Brazil
Larissa Núbia Nunes Vilany
São Paulo, Brazil
Wu Shin-Ting
Department of Computer Engineering and Industrial Automation, School of
Electrical and Computer Engineering, University of Campinas, São Paulo, Brazil
Enrico Ghizoni
Neuroimaging Laboratory (LNI), Department of Neurology, University of
Campinas,
São Paulo, Brazil
Kathleen H Nielsen
University of Washington, Seattle, WA, USA
Todd L Richards
Virginia W Berninger
xvii
Thomas Nury
Université de Bourgogne Franche-Comté/Inserm, Team ‘Biochemistry of the
Peroxisome, Inflammation and Lipid Metabolism’ EA 7270, Dijon, France
Gerard Lizard
Anne Vejux
Prasanna J
Karunya Institute of Technology and Sciences (KITS), Coimbatore, India
S Thomas George
Karunya University, Coimbatore, India
M S P Subathra
Karunya University, Coimbatore, India
Ola H Skjelda
Gillberg Neuropsychiatry Centre, Institute of Neuroscience, Sahlgrenska Academy,
University of Gothenburg, Gothenburg, Sweden
Natalia Cristina de Oliveira Vargas e Silva
Adventist University of São Paulo, São Paulo, Brazil
Priscila de Souza
Health Ability - care, education and health promotion, São Paulo, Brazil
Juan Rafael Orozco-Arroyave
GITA Lab, Faculty of Engineering, University of Antioquia, Medellín, Colombia &
Pattern Recognition Lab, University of Erlangen, Germany
Juan Camilo Vásquez-Correa
GITA Lab, Faculty of Engineering, University of Antioquia, Medellín, Colombia &
Elmar Nöth
Stephen Camilleri
Directorate for Learning and Assessment Programmes, University of Malta, Malta
Deborah Chetcuti
Ruth Falzon
xviii
Suchitra Joshi
Department of Neurology, University of Virginia, Charlottesville VA, USA
Denia Ramirez-Montealegre
University of Tennessee, Knoxville TN, USA
Pearl Rodrigues
SVKM’s Dr Bhanuben Nanavati College of Pharmacy, Vile Parle (W), Mumbai,
India
Pravin P Kale
India
Lokesh K Bhatt
India
xix
IOP Publishing

Ayman El-Baz and Jasjit S Suri
Chapter 1
Majority vote of machine learning methods
for real-time epileptic seizure prediction
applied on EEG pediatric data
Medina Bandic, Jasmin Kevric, Dino Keco and Samed Jukic
This research represents the practical application of machine learning in healthcare

and biomedicine. More specifically, it will be shown how analysis and processing of
EEG signals together with machine learning can help epileptic seizure prediction in
real time. The objective was to try to reach better, improved results by trying
different techniques and algorithms for pre-processing and processing of data and
measuring performance. EEG data is taken from Children’s Hospital Boston. After
removing the ‘noise’ using MSPCA and classification to appropriate subsets, wavelet
packet decomposition (WPD) proved to be the best for feature extraction and
random forest for classification, with satisfying training time and high level of
accuracy in predicting the epileptic seizures. The limitation is that only a few patients
were chosen as a sample for analysis, but that limitation can be overcome in future
research. The obtained results can be used in either further studies about epilepsy
and preventing epileptic seizures, or as direct application in healthcare in the form of
biomedical devices or software for seizure prediction. In that way, it is the first step
of solving one of the very common problems in healthcare and helping epileptic
patients live their lives more easily and better.
1.1 Introduction
Electroencephalogram, or EEG, is a way to monitor and record electrical brain
activity by using electrodes—metal disks which are attached to the scalp during the
process. Output of the process is recorded electrical waves and impulses which can
be analysed in order to find some patterns and detect potential problems (Greco et al
2008). The purpose of this work is to do experiments which will show how EEG
signal analysis can help in epileptic seizure prediction by training the data examined
from the Children’s Hospital Boston’s EEG signal database. The challenge while
doi:10.1088/978-0-7503-2723-7ch1 1-1 ª IOP Publishing Ltd 2020

doing the research was to deal with digital signal processing as well as studying the
healthcare and biomedical aspect of the problem. If a solution is found, it can be
very useful and have various applications which are mentioned later.
1.1.1 Epilepsy and epileptic seizure

Epilepsy is a well known neurological disorder which affects millions of people
worldwide. It is recognizable by epileptic seizures—vigorous shaking of the whole
body. Its cause is not known in many cases. A person can have it from birth as well
as get it in later stages of life as a consequence of some brain injury, disease or stroke
during a process named epileptogenesis. Even though it cannot be exactly cured yet,
it can be treated, usually by appropriate medications and only when really necessary
by surgical methods (Begg et al 2008).
However, epileptic seizures cause a lot of difficulties in the life of an epilepsy
patient. There is no specific cause of a seizure, which make it unpredictable, and no
rule as to how long the seizure will last. There are some conditions that can trigger it,
such as stress, lack of sleep, but also light flickering or some other external stimuli.
When it happens, neurons are firing in a non-regular way. There are two types: focal
(starts in one hemisphere of the brain) and generalized (starts in both hemispheres).
1.1.1.1 Epilepsy seizure prediction

A medical application designed to control seizures comprises two systems
(Winterhalder et al 2003):
• a seizure prediction algorithm raising an alarm when it senses an upcoming
seizure;
• a system acting to control a seizure.
A system controlling seizures may be an effective anticonvulsive agent in the

epileptic focus or may act as an electrical stimulation of the vagus nerve.
Moreover, one simple alert of upcoming seizure may be enough for a patient to
avoid dangerous situations or actions, like climbing stairs or playing sports.
The main aim of a seizure prediction method is to predict an upcoming epileptic
seizure. If a flawless prediction method were to exist it would postulate the precise
time when seizure onset was going to happen, which is not predictable by present
seizure prediction methods. Accordingly, a period called seizure occurrence period
(SOP) defined as a time interval where the seizure possibly might occur, can be taken
into consideration. Furthermore, a minimum time window between the alarm and
the beginning of SOP should exist to make an intervention or a behavioral
adjustment possible. This window of time is referred to as the seizure prediction
horizon (SPH), figure 1.1 (Winterhalder et al 2003).
These two periods should be considered, with the goal that a seizure must happen
between the seizure event period, and not between the seizure expectation horizon.
The vulnerability of the prediction has been reflected by the fact that the exact time
of seizure onset differs within SOP. The SPH ensures that an alarm is not given too
early, prompting an appropriate reaction to an upcoming seizure. Alarm signals
1-2
Figure 1.1. SPH and SOP.
with no seizure during SOP are classified as false predictions (Winterhalder et al

2003). Seizure prediction horizons that last long enough give a patient enough time
to prepare herself or himself for an arising seizure, avoiding dangerous situations or
actions. In addition to warning the patient, an intervention system in the form of an
implanted ‘brain pacemaker’ can be imagined, which would activate a mini pump to
deliver anticonvulsive drugs into the epileptic focus or trigger electrical stimulations.
The strength of negative effects caused by false predictions depends on the chosen
intervention system. When simply warned, the patient prepares during the SPH and
waits for a seizure during the SOP, which will not arise in the case of a false
prediction, making the patient unreasonably nervous. After too many false alarms, a
patient may not take further alarms seriously and risks being unprepared for
seizures. However, patients will be under huge psychological stress if they take all
alarms seriously. If too many interventions based on false predictions are carried
out, like triggering an electrical stimulation or the administration of anticonvulsive
drugs, additional side effects may arise (Winterhalder et al 2003). For the practical
application of a prediction method, the information on the distribution of false
predictions over the day may be important. A missed seizure while a patient is asleep
may be less dangerous than a seizure during the day, since patients are less exposed
to accidents or injuries. Therefore, seizure-prediction methods could be adjusted if
information on the distribution of false predictions is available or if they occurred
only during specific states of vigilance (Schelter et al 2006).
1.1.2 EEG signals related to epileptic seizures

Patients are sometimes admitted to hospitals in order to be recorded constantly for a
longer period, so that they can be monitored during the seizures. Recording during
the seizure can give significantly more information than recordings between seizures.
Speaking of which, there are different types of signals according to the stages of
seizure: ictal signals are those recorded during the seizure; preictal signals are
recorded right before ictal ones, and mean that a seizure is coming, and finally,
interictal signals are recorded in between the seizures (figure 1.2) (Goldberger et al
2000).
1.2 Literature review

The reference work that was crucial to this research was ‘Implementation of novel
signal decomposition and data mining techniques in EEG signal analysis’ and it
represents experiments done by Jasmin Kevrić and Abdulhamit Subasi, and presents
1-3
Figure 1.2. Interictal, preictal and ictal signals.
solutions on this same topic. It is a research in which some certain signal

decomposition and data mining techniques for classification of different types of
EEG signals are combined.
The first experiment was about the effect of multiscale principal component
analysis (MSPCA), de-noising method for EEG signals for epileptic seizure
detection (interictal versus ictal). The second experiment is done with the addition
of the preictal class. A Freiburg EEG dataset is used and three time-scale signal
decomposition methods (DWT, EMD, and WPD). The last experiment deals with
EEG signals that were recorded using a Mindwave Mobile Headset.
The final results show that if MSPCA is applied on signals that belong to one
particular class, we get really satisfying results. The drawback is that only part of the
database is used, the data from only one randomly chosen group of patients. More
specific results would probably be obtained from all of them, even though that
would require much more time. Hadoop or Apache Spark might also be good
solutions for faster data processing, so perhaps it would be good to try some other
tools and programs as well as some different algorithms and techniques in order to
see if there could be some improvements in efficiency. (The ones used in this one are:
1-4
artificial neural networks with the back-propagation algorithm; support vector

machines; nearest neighbour and some decision tree algorithms and classifiers.)
There are a lot of research papers that describe and discuss different techniques
for solving similar problems to this one. Descriptions of some of them with their
findings and results relevant to the research are given in the following paragraphs.
‘A MapReduce-based rotation forest classifier for epileptic seizure prediction’
(Jukić and Subasi 2017) describes analysis of EEG signals for seizure detection by
combining rotation forest classifier, multiscale principal component analysis for
removing the ‘noise’ from data, WPD for feature extraction and, most importantly,
map-reduce-based algorithm for prediction. Results showed that performance of the
model was satisfying: training time was reduced and accuracy to a high level. This
paper also set up the base for this research as well.
‘Application of machine learning to epileptic seizure onset detection and treat-
ment’ (Shoeb 2009) is a very similar work to the previously mentioned one. Data
from 23 patients from CHB-MIT database is used and processed by different
algorithms for seizure detection. In addition to this, Shoeb also explained how this
can be implemented in an embedded device which could deliver some therapy in the
predicted time, in order to prevent or at least weaken the seizure. He also describes
how this whole system can be used for medical diagnostics and therapeutic purposes.
‘A fuzzy rule-based system for epileptic seizure detection in intracranial EEG’
(Aarabi et al 2009) represents a seizure detection system based on fuzzy rule. It is
used as a method for detection of epileptic seizures. Different kinds of features were
extracted from IEEG segments and integrated using the fuzzy rule-based system.
302.7 h of intracranial EEG recordings from 21 patients having 78 seizures were
used for performance evaluation of the model. This system can be a good seizure
detection tool with high sensitivity and low false detection rate.
‘Application of higher order spectra to identify epileptic EEG’ (Chua et al 2011)
describes making an automatic system that can detect seizures as well. The research
is about performance of GMM—Gaussian mixture model and SVM classifier using
the features from higher order spectra and from the power spectrum. HOS features
achieved 93.11% accuracy versus 88.78% with the power spectrum features for a
GMM classifier. The SVM classifier improved to 92.56% with features based on the
bispectrum. It is shown that GMM classifier with HOS based features has better
performance.
‘Analysis of EEG signals under flash stimulation for migraine and epileptic
patients’ (Akben et al 2011) is the study about multichannel EEG in migraine and
epileptic patients and the existence of magnitude increasing under visual stimulation
is investigated. AR (autoregressive model) parametric method is used for EEG
signals analysis. Different EEG power spectra shapes were examined for medical
information. The results of this research represent proof of the power of AR method
in resolution of frequencies.
‘Real-time epileptic seizure prediction using AR models and support vector
machines’ (Chisci et al 2010) is research about predicting epileptic seizures from
online analysis of EEG data. Combination of AM of the EEG time series and least-
squares parameter estimator for EEG feature extraction with SVM for classification
1-5
between preictal/ictal and interictal EEG stages is implemented. The Freiburg

dataset is used and the research resulted in correct prediction of all seizures and a
low false alarm rate.
‘Extracting and selecting distinctive EEG features for efficient epileptic seizure
prediction’ (Wang and Lyu 2014) contains feature representations for the EEG
signal for achieving good epileptic seizure prediction results. Different ML
approaches have been used for classifiers to divide the extracted features into
preictal and interictal classes. 98.8% sensitivity results are gained on the 19 patients
included in the experiment.
‘Principal component analysis-enhanced cosine radial basis function neural
network for robust epilepsy and seizure detection’ (Ghosh-Dastidar et al 2008)
presents principal component analysis combined with mixed-band wavelet-chaos
methodology for accurate classification of EEGs into healthy, ictal, and interictal
EEGs. PCA is employed for feature extraction and enhancement. The accuracy of
the classifier is evaluated by parametric and sensitivity analysis. This methodology
resulted in high EEG classification accuracy—96.6%, and showed itself to be robust
to changes in training data with 1.4% standard deviation. When only normal and
interictal EEGs are considered, the classification accuracy is 99.3%.
‘Epileptic seizure classification using neural networks with 14 features’ (Costa
et al 2008) is about classification of epileptic EEG data into four stages (interictal,
preictal, ictal and postictal) and applying neural networks architectures on it. There
were fourteen features extracted: accumulated energy, level, Lyapunov exponents,
correlation dimension, five variants of energy STE and five variants of energy LTE.
In terms of accuracy, the following neural networks are compared: radial basis
function; feed-forward backpropagation; layer-recurrent; Elman; feed-forward
input time-delay backpropagation and distributed time delay. The performance
should be checked by both sensitivity and specificity, if it is intended to be used for
seizure prediction.
‘Epileptic seizure prediction using hybrid feature selection over multiple intra-
cranial EEG electrode contacts: a report of four patients’ (D’Alessandro et al 2003)
is about a new model for choosing EEG features and locations of electrodes for
seizure prediction focused on precursors that occur within ten minutes of electro-
graphic seizure onset. The algorithm is trained on preseizure records and then used
on others for validation. Records of four patients are used for demonstrating the
results. An average probability of prediction obtained by the research is 62.5%.
‘Fuzzy techniques for classification of epilepsy risk level from EEG signals’
(Harikumar and Narayanan 2003) is about creating a fuzzy classification model for
epilepsy risk level analysis. Fuzzy techniques are used for classification of risk levels
for each in all the EEG-signal channels. For improving classification result, a new
optimization technique was used and results improved from 6.25 to 11.9 quality
value.
‘Seizure clustering during epilepsy monitoring’ (Haut et al 2002) is about
identification of risk factors connected with seizure clustering during epilepsy
monitoring. Data from patients who are at the stage of presurgical epilepsy
monitoring at Montefiore Medical Center or Yale–New Haven Hospital is taken
1-6
for the research. The risk factors include history of clustering at home, MTS on
MRI, and possibly more than one seizure focus. In this study, clustered seizures were
as good in localization as were non-clustered ones.
1.3 Methodology
1.3.1 Data and data collection
Data represents EEG signal recordings from Children’s Hospital Boston from
different-aged pediatric patients suffering from epilepsy. There are 23 different
subjects who were observed for several days (around 844 h to be more specific). Out
of those, there are 5 males with ages 3 to 22 and 17 females with ages 1.5–19 years
old. The 21st and 1st subjects are actually the same person, just recorded at different
ages. The 24th subject was added later and there is no more information about the
person (Goldberger et al 2000).
Recorded data is stored in 24 folders in .edf-format files. There are from 4 to 42
consecutively recorded files per subject. The biggest eventual gaps in recording are
up to 10 s, caused by some unpredicted, usually hardware, issues. Each file
represents a 1 h long recording, except for those ones for patient 10—2 h per file,
and patients 4, 6, 7, 9 and 23—4 h.
The hospital’s electroencephalogram sampled 256 signal samples per second. As
already mentioned, the International 10–20 system is used for naming EEG
electrodes’ locations. Some other signals are recorded alongside EEG ones for
some patients, like ECG for the 4th subject and VNS or vagal nerve stimulus for
patient 9 (Goldberger et al 2000).
In the end, it is important to outline that there are files with and without seizures
(129 out of 664 are with seizures). Some files have more than one seizure, so in total
there are 198 recorded seizures. More details about number of seizures per patient
are shown in table 1.1.
Based on information in this table as well as information like number of files,
hours of recording, consecutive files and consecutive seizure files and so on, five
subjects were chosen for analysis: chb03, chb05, chb08, chb16 and chb18.
1.3.2 EEG signal processing techniques

There are many methods developed and used for signal decomposition and analysis.
In general, they can be divided in two groups—time-scale (wavelet analysis) and
time-frequency ones. Time-scale methods (waveform segmented by scale) have been
proved to be a better choice in most cases (Kevrić and Subasi 2015) and are used in
this research as well. Wavelet methods that are used as a part of this experiment are
discrete wavelet transformation and wavelet packet decomposition. Also, multiscale
principal component analysis is used for removing the noise from the data. They are
all briefly described in the next paragraphs.
1.3.2.1 Discrete wavelet transformation (DWT)

Discrete wavelet transformation is a very effective technique for signal analysis. It
works in such a way that it transforms signal x(t) into functions—wavelet
1-7
Table 1.1. CHB-MIT subjects—details.
Number of files
Subject Number of files with seizures Extra information
CHB01 42 7
CHB02 36 3
CHB03 38 7
CHB04 42 3 1 with 2 seizures, ECG signals
CHB05 39 5
CHB06 18 7 1 with 2; 1 with 3 seizures
CHB07 19 3
CHB08 20 5
CHB09 19 3 1 with 2 seizures, VSN signals
CHB10 25 7
CHB11 35 3
CHB12 24 13 4 with 2; 1 with 3; 1 with 4; 2 with 5;
2 with 6 seizures
CHB14 26 7 1 with 2 seizures
CHB17 21 3
CHB18 36 6
CHB19 30 3
CHB20 29 6 2 with 2 seizures
CHB21 33 4
CHB22 31 3
CHB23 9 3 1 with 2; 1 with 4 seizures; same patient
like CHB01
coefficients, by the process of scaling the ‘mother’ wavelet function. A base

waveform can be decomposed to a combination of so called ‘approximation signal’
and ‘detail’. Approximation signal is the smooth part of the base signal and
represents the main features. Details are ‘fine’ parts with very quick oscillations.
Decompositions are done in levels or scales, where the approximation signal is
further split. The formula of the process is as follows:
xj + 1(t ) = xj (t ) + yj (t ) (1.1)
where j represents the scale level, xj(t) is approximation signal and yj(t) represents
the details. So approximation signal on the level j + 1 is sum of approximation and
details of the level j. Approximations are low-pass, and details are of a high
resolution (Ghorbanian et al 2012). In order to know which one is which, the signal
has to pass through low-pass and high-pass filters.
1-8
Firstly, base signal is converted to discrete time signal x[k] and the scale j is set to
1. The signal then goes through the already mentioned high-pass and low-pass filters.
The result of the first one is detail and the result of the second one is approximation:
[i ] = whigh[i ] = ∑x[k ]h[2i − k ]k (1.2)
[i ] = who[i ] = ∑x[k ]l [2i − k ]k (1.3)
Signals are down-sampled by 2 in each step. x[k] is set to Aj and the process repeats
until we get to the maximum j value we chose in the beginning as our preference. It
can be seen that only the approximation signal is decomposed.
1.3.2.2 Wavelet packet transform (decomposition)—WPT (WPD)

WPD is an algorithm pretty similar to DWT except for the fact that the detail part is
decomposed as well. WPD provides even better frequency resolution than DWT. In
the end the number of wavelet coefficients is 2j for j number of scale levels
(compared to j + 1 for the DWT).
1.3.2.3 Multiscale principal component analysis—MSPCA

DWT and WPD can be efficiently applied in de-noising of signals. Multiscale
principal component analysis is a very efficient de-noising technique which is based
on those wavelet transformations. The reason for this is that the noise is always in
the detailed part of the transform, and that makes it easier to filter it. Basically, the
process generally speaking consists of three steps (Bakshi 1998):
1. DWT or WPD for extracting the detail part necessary for the process;
2. Making adjustments of details (the coefficients) based on certain criteria for
removing noisy parts;
3. The recovered waveform with those changes is the de-noised output signal.
In order to understand what MSPCA is, it should be explained what basic principal
component analysis (PCA) is. PCA is actually a statistical technique applied on
multivariate data existing in multidimensional space. Each space represents a
‘variable’ and the goal is to reduce the quantity of those variables in order to
simplify the understanding of the dataset. Removed variables are those that are of
less importance. During the PCA process the variables are transformed in new
uncorrelated variables. They are called principal components. If signal data is
represented as a matrix with n × m dimensions, where n is number of variables and m
the number of signals, then: X = TPT (3.8), where T is matrix of PC scores, and P is
matrix of PC loadings (Jolliffe 2002). A crucial part in this technique is choosing the
essential principal components, and there are many proposed methods for this.
When PCA is combined with previously described wavelet transformation, we get
MSPCA. W is the matrix which is a result of wavelet transform and which contains
the previously described approximation and details. The process is to perform PCA
on all components of W and then choose a certain number of principal components.
The last part is filtering wavelet coefficients, so they meet certain threshold criteria.
1-9
In the end, new reduced matrix should be recovered, and PCA executed on that
result, which forms the de-noised output. This method is used in this research and it
will also be described later.
1.3.3 Meta heuristic methods

Speaking of machine learning algorithms, all of those ‘standard’ ones which can be
applied in data mining in general, can be applied on data extracted from signals as
well, and used for classification of data. Most of them were tried in this research as
experiments, in order to see which one performs the best on the data, and results of
the analysis will be stated later on in the chapter. In the following paragraphs a short
summary of well-known algorithms is stated.
1.3.3.1 Naive Bayes

Naive Bayes is a very simple model of creating a classifier by assigning some certain
and specific classes of characteristics to a problem. Depending on those character-
istics and how they match possible classes of results we get our output. The main
characteristic of Naïve Bayes algorithms is independence of each feature instance.
Naive Bayes algorithms are based on Bayesian theorem. That is actually
calculating the probability of some event to happen by taking into consideration
probabilities of event(s) that already happened. The formula is as follows:
(y∣X ) = (X ∣y )(y )/ P(X ) (y − output, X − feature vector) (1.4)
In order to create a Naive Bayes classifier out of this probabilistic model, the model
is combined with ‘the decision rule’. For Naive Bayes that rule in most cases means
either that the possibility of posterior loss is minimized, or that it represents so called
‘maximum a posteriori’ or MAP probability estimation (maximum likelihood).
Based on these differences there are different types, such as Gaussian Naive Bayes,
Multinomial Naive Bayes and Bernoulli Naive Bayes. This algorithm is very useful
in practice and has many applications in spite of the fact that it is inaccurate very
often.
1.3.3.2 Decision tree and decision tree classifiers

Decision tree is an algorithm that is based on a tree-like graph or model, and is
therefore very easy to understand. It is often applied in decision analysis. This
algorithm is highly effective and can be performed on high-dimensional data as well.
Speaking of ‘the tree’, there are two important segments: nodes and branches. Each
node in the tree represents some sort of test and branches represent outcomes of each
test (Berry and Browne 2006).
There are three types of nodes:
• root node represents the starting point, training data which is about to be
divided based on some condition;
• root node is split into internal nodes in the next level. They can be either
decisional nodes or chance nodes and based on that are usually marked as
rectangles or circles in decision tree diagrams;
1-10
• when the internal nodes cannot be divided anymore, they are called leaf nodes
and they are usually marked with a triangle on the graph.
There are many classifications of decision trees. They can be binary—where nodes
are always divided into two new nodes—and non-binary or multi-branch—where
nodes are divided into more than two new nodes. They can be classification trees—
where a predefined class of data is predicted, or regression tree—where some specific
value is predicted as outcome.
Random forest works in the way that some more groups of trees are grown and the
most frequent outcome from all of them is chosen to be the final output. Groups of
trees are grown by creating independent random factor, which is combined with
input pattern for creating classifier. It is based on CART (so both classification and
regression). This algorithm is characterized by high accuracy and speed, and
resistance to noise (Breiman 2001).
Rotation forest is built by first applying PCA on one part of the features. Let us
say there are k random subsets of features. Classifiers are then made by rotations of
feature axes. Decision trees are usually used as the base of classification, because
they are very sensitive to rotation. The results of this algorithm are very powerful
and accurate as well as individual and very diverse classifiers. This is perhaps the
best example of classifier ensembles, based in general on grouping more classifiers
which are trained on different features or data subsets (Rodríguez-Bermúdez and
García-Laencina 2012).
1.3.3.3 k-Nearest neighbor

Nearest neighbour is a very commonly used algorithm, mostly because of its
simplicity. This is one kind of supervised learning where training data is divided
into labeled classes and can be multidimensional. When we want to classify a new
sample of testing data in one of those categories the only thing that should be done is
finding the most frequent category among the k number of nearest training samples.
Constant ‘k’ is user-defined (Jukić et al 2018).
There are two big disadvantages of the algorithm. The first one is the noise and
‘dummy’ features which can directly affect the output. The second is the skewed
boundary between classes. In cases like this, what mostly happens is that the class
with more frequent data samples has a much higher chance to be chosen as a result,
because it is a higher possibility that it dominates among the k number of neighbors.
In order to solve this problem categories can be ‘weighted’ in the way that the
distance between the new sample and each of the k-neighbors is measured.
1.3.3.4 Support vector machine (SVM)

Support vector machine is one of the supervised learning models and it is used for
regression analysis as well as classification. One of the biggest advantages of this
algorithm is that it results in global minima and cannot be stuck in local minima,
which is really important for the training process (Vapnik 1995). Support vector
classifier is defined in the following way:
1-11
P = {(xi , )∣xi ∈ Rk , yi ∈ { −1, 1}} i = 1n (1.5)

In this formula x represents features with k number of attributes and y is output. ‘k’
also represents the number of dimensions. After the training set was given, where
data is marked in the way that it belongs to one of two subsets, SVM builds a model
in order to fit some new examples in one of those subsets (Begg et al 2008). To be
more specific, desired output of this classifier is to make a hyperplane which would
split up the data. The number of those hyperplanes can be infinite. The one that is
chosen is the one which separates the data the best and whose distance from the
nearest data point is the biggest. The hyperplane margin can be extended for better
results.
1.3.3.5 Artificial neural networks (ANNs)

An artificial neural network is a data processing algorithm, that basically imitates
biological nervous systems. The main part is the information processing system
made of a huge number of processing units—neurons, which can together solve
problems. ANNs learn by example like people. Researchers from many scientific
areas design ANNs to solve a variety of problems. Modern computers are better
than people in things like numeric computation and related symbol manipulation.
However, people effortlessly solve complex perceptual problems faster than the
fastest computers. That is why ANN models use some ‘organizational’ principles
used only in the human brain (Stergiou and Siganos 1996).
An artificial neuron is like an engineering approach of a biological neuron—a
device with many inputs and one output. ANN consists of a large number of simple
processing units, interconnected with each other and layered also. In the case of a
biological neuron, information comes into the neuron through the dendrite (input),
after what it processes and passes on through the axon (output). In the case of ANN,
the information comes into the neuron through weighted inputs. In the neuron the
weighted inputs are summed up and processed and then passed through output.
1.4 Experimental setup

1.4.1 Data setup
In order to perform experiments for this research, a proper environment should be
set up first. Regarding the data, it is taken from the previously described CHB-MIT
Scalp EEG database from Children’s Hospital Boston. The database is accessible
online, and information can be retrieved from it in many ways and formats. For this
research, five patients were chosen after analyzing information obtained about data
from all the patients: subjects chb03, chb05, chb08, chb16 and chb18. The first thing
to do was to separate the data in folders with ictal and interictal signals, where an
approximately similar number of files were used for training and testing (four folders
per patient).
Data about subject-patient, chb05—a 7 year-old girl, consisted of 39 .edf files.
Each file represents about one hour of recording. Five of those files contain seizures
1-12
(one seizure per file), which last around 100 s per seizure in average. Files are divided
in folders as follows:
• Interictal train: chb05_23, 24, …, 39 (17 files in total).
• Ictal train: chb05_06; 13; 22 (3 files in total).
• Interictal test: chb05_01,…, 05; 07, …, 12; 14; 15; 18, …, 21 (17 files in total).
• Ictal test: chb05_16; 17 (2 files in total).
Data about the second subject-patient, chb18—18 year-old girl, consisted of 36 .edf
files. Each file represents about one hour of recording. Six of those files contain
seizures (one seizure per file), which individually vary from 30 to 70 s in duration per
seizure. Files are divided in folders as follows:
• Interictal train: chb18_01, …, 15 (15 files in total).
• Ictal train: chb18_29, …, 32 (4 files in total).
• Interictal test: chb18_16, …, 28; 33; 34 (15 files in total).
• Ictal test: chb18_35; 36 (2 files in total).
1.4.2 Data preparation with de-noising and feature extraction

The overall procedure for both patients is the same. The first thing to do was to load
23 channels of signals. Almost all of the patients in the database have those 23
channels in files and chb05 is one of them. However, for patient chb18 they are
distributed a bit differently. That is mostly because of the fact that during the time
there were two versions of channel distribution for this patient. Data sampling rate
for both patients is 256 Hz s−1.
For the interictal training part, data from all the interictal training signals,
previously selected, is loaded in the signal matrix in samples of random 10
min. After the 23-signal data is loaded, it is divided in 2048 data chunks.
Signal window size (fast Fourier parameter) of 2048 samples is chosen because
it is usually the most optimum choice. Those chunks represent 8 s of data
(2048 chunks/256 Hz s−1 = 8 s). For each chunk we invert the signal, and out
of all those segments generate a new one. The matrix for the new signal is
created for every 8 min (later on we will describe why exactly we used this
data frame).
The next step is de-noising the data by the previously described MSPCA. In
order to perform MSPCA for this experiment, the first thing to do was to perform
wavelet transform—WPD method in this case. The scale level chosen for wavelet
transformation is 5 (this means the resulting number of wavelet coefficients is
25 = 32) and sym4 is chosen for the wavelet filter. The number of principal
components or npc is set automatically by the selected Kaiser’s rule. All of these
parameters are taken as an input for MSCPA. MSPCA is applied on the 8 min
data segments, previously created. As an output, de-noised signal as well as
percentage of performance quality and number of retained principal components
are obtained.
Regarding training ictal data, the procedure is similar compared with the one
with training interictal data. The main difference is combining the 48 min long
1-13
preictal (preseizure) signals with ictal (seizure) signals (from ictal train folder),
and that is how the input data for the training is obtained. This means that
signals which are predecessors of selected ictal signals have to be loaded as well.
The starting point of the preictal part is calculated by using the following
formula:
preictalStart = fileLength − (length48min − seizureStart) (1.6)
The next step is segmenting data in 2048 parts like before, where each part represents
8 s of data, and recovering new signal from these chunks. We repeat this for each
ictal file separately. In the case of subject chb05 there are three of them (so we repeat
the process three times) and for chb18—two of them (twice). Everything is loaded in
one all-inclusive signal in the end. The matrix is created in the same manner like for
interictal signals for every 8 min and MSPCA performed on it.
The reason the 8 min interval is chosen is because a 48 min long preictal period
could be segmented on six of those time-frames. The first five out of those six time-
frames—40 min—are used for seizure prediction. That process will be explained
later on in detail. The result of this part of the experiment is de-noised preictal signal
data later used for making a classifier.
It is important to mention that WPD final coefficients were also used for feature
extraction, not just for MSPCA. Feature extraction for both interictal and ictal
train datasets happened after performing MSPCA on the data matrix. This part is
crucial, considering that those features actually represent characteristics of
EEG signals and their behaviour. In these examples the WPD algorithm generates
95 features. To be more specific, for each wavelet coefficient the following is
calculated:
• Mean of absolute values:
1 M
μ= ∑ j=1∣yi ∣ (1.7)
M
• Average power:
1 M
λ=
M
∑ j =1
y2 (1.8)
• Standard deviation:
1 M
σ= ∑ j=1(yj − μ)2 (1.9)
M
• Ratio of absolute mean values:
1-14
M
∑J = 1∣yj ∣
X= M (1.10)
∑ j = 1zj
• Skewness:
3
1 M (yj − μ)
ϕ= ∑ (1.11)
M j =1 σ3
• Kurtosis:
4
1 M (yj − μ)
φ= ∑ (1.12)
M j =1 σ4
where M is the number of samples; y, z are the vectors of the de-segmented data.
1.4.3 Training the data—Matlab GUI

As result from the previous part, de-noised signal matrices for interictal and ictal
training data as well as extracted features are obtained. The final part of preparing
the data for training is to combine those two matrices into one. In that combined
matrix class a column is added so that interictal and ictal data can be differentiated.
In this case class 1 represent signals with, and class 0 represent signals without
seizures. Data should be stored in .mat and .csv file formats.
The prepared data is loaded in software called Matlab GUI. Matlab GUI
implements the Java library from WEKA. WEKA by itself is software which
provides all the varieties of machine learning algorithms for training and testing the
data. It contains tools for classification, regression, visualization, clustering and so
on. It can be applied through Java code (the software is in Java in general) or
directly on the dataset. WEKA is a very simple program very often used by
beginners in data science.
However, a disadvantage is the inability to save the classifier once it is made. That
can be done through Matlab code, so therefore that is why we need the Matlab GUI.
Matlab GUI is software created as a project at International Burch University in
order to overcome some problems that were present with some other software and
tools. It is used offline and has the capability to implement all the main machine
learning algorithms.
As can be seen in figure 1.3, the Matlab GUI looks very simple. The main purpose
of this tool is to create and store classifiers. Data can be loaded in .csv and .mat
format. It is possible to use cross-validation (a method where loaded data is
segmented on train and test part) for training and testing or to load the test data
separately. Machine learning algorithms which can be chosen are: SVM, kNN,
Random Forest, Naive Bayes, NBTTree, RandomTree, SimpleCart, ANN,
RotationForest, C4, RBFNetwork and Simple Logistic.
1-15
Figure 1.3. Matlab GUI.
1.4.4 Data testing, processing and seizure prediction

This is the part where we use signal files from interictal and ictal test folders. Those
signals in combination with the previously obtained classifier are taken as an input
for processing. The overall process is done in iterations. The first thing to do is to
load interictal test data. In the first iteration the first ictal signal (out of two) is
loaded. Data is again split into intervals of 8 s and stored in a new matrix. Just like
before, 8 min data chunks were pre-processed separately—firstly they were de-noised
by WPD and MSPCA and then the features extracted.
The next part is also the main part of this project and describes how to get to
results and seizure prediction, which is the whole point. For this part those
previously extracted features and class, based on which seizure and non-seizure
signals can be differentiated, are required. 1 is for ictal files and 0 for interictal files.
The process works in the way that for each 8 min data part, ‘1s’ or detected seizure
parts are summed up.
Afterwards, this certain threshold is chosen—the value which would filter whether
the predicted sum value calculated in the previous step is enough to become the
possible seizure prediction. In the case of patient chb05, that threshold value was set
to 600 first, but after getting many false positive alarms decreased to 400, which
showed good performance. For patient chb18 the most appropriate threshold value
was 450 in the end.
As previously described, a 48 min long period is chosen as the preictal signal.
That can be divided into six 8 min long segments. The overall system would not
make any sense if the patient is warned in the last 8 min time-frame right before the
seizure. The point is that in the worst case the patient has 8 min to prepare for a
coming seizure. Because of this, the first five segments are chosen and the number of
predicted seizures summed up for each. If the number is higher than the chosen
threshold, we count it as a positive prediction. In the case where three positive
predictions out of those five are positive, the ‘alarm’ is raised.
1-16
ALARM SEIZURE
1 1 0 1 0 0
8 16 24 32 40 48min
Figure 1.4. Example of predicted seizure.
1 0 0 1 0 0
Figure 1.5. No seizure detected.
Figure 1.4 shows an example of positive prediction. Each rectangle represents an

8 min signal chunk, previously processed. ‘1s’ represent positive seizure prediction. If
three out of five chunks predict seizure, an alarm is raised. One or two predictions
would not be enough and would lead to many false positive predictions. On the
other hand, for four or five predictions we would have too many false negative
predictions, and even those predicted ones could be perhaps predicted earlier.
Therefore, three is an optimal value for raising an alarm. In this example the patient
is warned 16 min upfront. In the worst case, prediction would happen 8 min before
the seizure (figure 1.5).
Basically, that prediction represents the end of the experiment. In this case, the
whole procedure is repeated for both patients. Results that are obtained can be
graphically represented and further analyzed (next chapters). The last thing to do is
to calculate and compare the overall performance. Training and testing time is also
measured and compared. Performance evaluation and possible further improve-
ments of the system are discussed later on.
1.5 Results
Unlike any other known seizure prediction method, our algorithm provides real-
time prediction. We mentioned before that our proposed method is patient oriented,
which means that each patient has a different machine learning trained model and
the result of every patient is a different case.
For this experiment, based on gender, age, number of seizures and seizure length
for each patient, we pick the first five patients.
Experiments done in this research represent models which do real-time seizure
predictions for patients suffering from epilepsy. In order to get from raw data at the
beginning to results and predictions, different methods had to be tried out, from de-
noising methods, through training algorithms, feature extraction and in the end the
testing. In each step the best choice is taken after which the process is moved
forward. In the previous section all the used methods and algorithms are described.
1-17
Table 1.2. Time of seizure prediction (alarm) before it happened in minutes for five pediatric patients.
Patient P5 P 18 P3 P8 P 16
Seizure number 1 2 1 2 1 2 3 1 2 1 2 3 4
Predicted time (min) 120 20 030 300 200 50 45 65 25 70 70 30 3 25
Table 1.3. Comparison table for seizure prediction on CHB-MIT database.
Author Accuracy (%)
Khan et al (2012) 91.8

Nasehi and Pourghassem (2013) 98
Shoeb (2009) 96
Fergus et al (2015) 98
Qidwai et al (2014) 90
Alickovic et al (2017) 99.7
This work 100
In this section all the performance results are stated. Each patient is unique and
therefore execution results are different.
Table 1.2 presents when proposed the method made an alarm before the seizure
happened. As you can see, in all cases it is at least 25 min beforethe seizure except for
the third seizure of patient 16, which gives enough time for the patient to take some
steps which can prevent the seizure or give time to call a medical center, or any other
process which can make the seizure easier. For patient 5, for the second seizure that
happened twice, this was probably because signals where the first alarm was
predicted were also the preictal signals from some other seizure which is not in
the test.
Table 1.3 represents the accuracy which is achieved using a training dataset for
pediatric patients. After we have finished with the training algorithm and get these
accuracies, we save machine learning models and use them for testing. We can see
from the table that for each patient we got a different accuracy result on the training
dataset. It is important to state that accuracy in a training dataset is not the key
point in accuracy for a testing dataset, and it will be proven in the following
figures where you can see that for some patients we achieved great results in the
training dataset, but when we applied that model in test, we actually did not get that
great a result. For some patients we did not get such high accuracy in the training
dataset, but when we applied that model in the testing phase, we got really good
results. It should be pointed out that accuracy is measured by cross-validation and
specificity/sensitivity method (confusion matrix).
As can be seen in the table, random forest classifier showed the best performance.
It classified almost all the data correctly in each try. Other classifiers except for
Naive Bayes are pretty accurate as well, but this will be further discussed in the next
1-18
section. Based on previous results, random forest classifier is chosen to be used as the
classifier for testing the data.
Patient chb05 has around 39 h of recording data. After the data was separated in
interictal/ictal classes the first thing to do was to pre-process and de-noise the ictal
and preictal dataset.
The next part was classification of the data. Matlab GUI—software based on
WEKA libraries, was used for this purpose and therefore the machine learning
algorithms provided by that software.
The next step is testing part or actual prediction of seizures. Testing data is divided
into 8 min chunks and each part tested individually. A constant threshold, is chosen
as a boundary, and if the number of positive predictions for a data chunk is higher
than that, and that repeats 3 out of 5 consecutive times, an alarm is raised for seizure.
All of this is really nicely shown in figures 1.6 and 1.7 where seizure predictions can be
seen for the two seizures chosen for testing data, separately.
Each bar on the graph represents an 8 min data chunk, and its predictions. Red
bars represent raised alarms. The yellow line is the threshold (so it is 400), the pink
one is preictal period and the green one at the end represents a seizure. There is
around 17 h of interictal and 1 h of ictal data. In these results seizure is predicted,
however, it is predicted perhaps too early (around 3.5 h before) and there are
perhaps too many raised alarms, but pros and cons as well as some possible solutions
will be discussed in the next section.
Figure 1.7 represents again the prediction for patient chb05, except for the fact
that this time different ictal signal was used. The threshold is again 400, and the ratio
Figure 1.6. Seizure prediction for seizure 1 (CHB05).
1-19
Figure 1.7. Seizure prediction for seizure 2 (CHB05).
of interictal and ictal signals is 17:1. This model had better performance considering
that there is a prediction about thirty minutes before the seizure. However, there are
still alarms which are raised much earlier, and that should not be the case. This all
indicates that there should be modifications for escaping these disadvantages.
Patient chb18 has around 36 h of recording data. After the data was separated in
interictal/ictal classes the first thing to do was to pre-process and de-noise the ictal
and preictal dataset.
The next part was classification of the data. Matlab GUI was again used for this
purpose and therefore the machine learning algorithms provided by that software.
Some of them were tried out and table 1.3 contains the results.
The next step is the testing part or actual prediction of seizures. Testing data is
divided into 8-min chunks and each part tested individually. A threshold is chosen as
a boundary, and if the number of positive predictions for a data chunk is higher than
that, and that repeats 3 out of 5 consecutive times, an alarm is raised for seizure. All
of this is really nicely shown in the next figures where seizure predictions for the two
seizures chosen for testing data are shown, separately. We also describe the
importance of choosing the right threshold value as the experiment had to be
repeated to obtain better results, just because of that value.
The graph in figure 1.8 is similar to the ones from before. Each bar on the graph
represent 8 min data chunk, and its predictions. Red bars represent raised alarms.
The yellow line is threshold (so it is 400). There is around 15 h of interictal and 1 h of
ictal data. In these results we can see that there is a false alarm right at the beginning.
That is not a desirable result so we moved the threshold value to 450 (figure 1.9).
In this example with higher threshold we can see improved results. Pink and green
lines are added where the pink one is preictal period and the green one represents the
seizure. In these results we can see that the seizure is predicted, however, it is
1-20
Figure 1.8. Seizure prediction for seizure 1 (CHB18) where threshold = 400.
1-21
predicted perhaps too early (around 5 h before) with five raised alarms, probably
because in a real-life situation there used to be a seizure in that period of time as well.
Pros and cons of the stated prediction as well as some possible solutions will be
discussed in the next section.
Figure 1.10 represents again the prediction for patient chb18, except for the fact
that this time a different ictal signal was used. The threshold is in this case 400, and
the ratio of interictal and ictal signals is 15:1. Just like before, there is one immediate
false alarm which is why the threshold is changed to 450, as seen in figure 1.11.
This model had better performance considering that there is no false alarm at the
beginning. There are also some after-seizure alarms, which means that there would be a
seizure following those alarms if a signal had been loaded afterwards. However, there
are still five alarms which are raised much earlier (3.5 h), and that should not be the case.
This all indicates that there should be some modifications of the model for escaping
these disadvantages, even though this one results with correct predictions as well.
The result of this experiment applied on patients 3, 8 and 16 is presented in a
research paper (Jukić et al 2018).
1.6 Discussion
This section represents the discussion about all the findings and results of the
research. First of all, performance of MSCPA and de-noising is discussed. After this,
discussion continues with the chosen machine learning algorithms used for building
1-22
the classifier, comparison of their performance and which one is chosen for making
predictions and why. Training and testing execution time and accuracy are discussed
as well, but the most important part of all is the performance of the actual seizure
prediction system and its comparison to some other seizure prediction systems from
referenced previous works and literature.
In the beginning it should be pointed out that data is taken from different
patients, and regarding health any kind of health records, each person’s data is
unique and should be independently considered. That is why there are different
prediction models and results for each patient.
Our proposed method is completely real-time, because datasets that we used for
training and testing are completely separated. So for the training phase we used 15 h
from interictal and two or three preictal and ictal phases, so we used all of those
signals to generate the training dataset. On the other hand, for testing purposes, we
used completely unknown signals, we processed them in the same way as we
processed the training signals, and sent to the trained model for evaluation. We sent
a chunk of 8 min signals and made a decision after every chunk. If three chunks out
of five were categorized as a seizure, then we defined the alarm state, and the patient
received a notification about that.
In our experiment, we used a threshold which is half of the total value for the
preictal signal, and compared our predicted value with the threshold. It gives good
accuracy, so we did not suggest changing that way, but it will be better if we use a
dynamic threshold, which will be half of the mid-value for preictal predicted values.
If we check our results, you can see that we predict all seizures, so we can say that
for these three patients we have predicted with 100% accuracy.
1-23
Table 1.4. Accuracy which is achieved for the training dataset for five patients.
Algorithm P5 P 18 P3 P5 P 16
Naive Bayes 74% 66% 70% 68% 65%

SVM 86% 80% 89% 87% 78%
kNN 93% 84%
Random forest 98% 99%
Random tree 92% 83%
ANN 93% 87%
Rotation forest 94% 90% 95% 92% 87%
For all patients that were included in the experiment we obtained excellent results,
however, patient 16’s third seizure was not provided with enough preictal signal,
which produced a very late seizure prediction, but again we predicted the seizure
before it started. Overall, for all patients, we managed to predict a seizure 30 min
before its start, which is enough time for a patient to take some action in order to
make the seizure easier or to take some medicine which to stop the seizure.
All patients have different prediction values for interictal signals. Patient 16 has
two chunks which are predicted as preictal period, but because of the majority
system implemented in our experiment, we did not mark that as an alarm.
For patient 8 we had perfect prediction for the interictal period, where we could
decrease the threshold, and again obtained 100% accuracy.
As we can see from table 1.4, our proposed method provides better accuracy than
all other related experiments. The reason for that we can find in the patient oriented
way of doing our experiment, where for each patient we defined a different model
based on their own training data.
In this section, we discuss the results and findings of our analysis and classification
of the selected studies. We discuss an overview of the selected studies and some
interactions among the taxonomy categories, presented further ahead.
The important thing related to our experiment is the big difference in time
execution of signal processing and testing when we used parallel processing. For
most cases, speed of execution for code with the parallel way is two times faster than
when we execute our code using the ordinary way.
Table 1.4 represents a comparison of average execution time for each patient. In
the results part we, also, present tables with a separate time for signal processing and
testing phase using a different way of execution. In table 1.4 we found the average
value for each patient for every type of execution and compared that time. What is
interesting to mention is that normal execution (serial processing) and MATLAB
parallel execution give almost the same execution time. On the other hand, we have
the code parallel as a better method which doubly reduces processing times.
Comparing our work with other experiments, we could not find anyone who
combines real-time and the parallel or distributed way of processing; because of that
we need to check our results with real-time prediction and distributed systems
separately. In the study of Liang et al (2010) they got accuracy of 92%, seizure
1-24
detection latency was 0.6 s, in comparison with our experiment where we got better
accuracy for the patient where we found our proposed method to be suitable and our
accuracy is 100% for prediction. Ali Shahidi et al (2010) in their study using
detection of epilepsy got accuracy around 90% latency of 7 s compared with our
experiment where we are making seizure prediction with 100% accuracy. Beside
these works, there are more experiments which are mentioned in literature review,
and all of them are focusing on real-time detection and none of those papers are
predicting epilepsy. Another group of papers are related to distributed systems, and
in the research of Teixeiraa et al (2014) they have 50% of predictions with a false
alarm of 1 in 6 h, compared with our proposed system where we have 100% accuracy
and no false alarm at all. In a paper of Dutta et al (2011) they reduced time
consumption and showed an increased speed from half an hour processing to 12 min
compared with our experiment and parallelization where we increase speed three
times. In the paper of Ježdík et al (2010) they decreased time consumption for signal
processing from 20 h using one unit to less than one hour when they used 18 units,
where we also show that speed with more cores could be faster by three or more
times.
As was mentioned before, in order to build the system, first we took ten-by-ten
random chunks of training data in order to generate the training dataset. Interictal
and preictal/ictal data was separately pre-processed. The pre-processing part
included loading data, WPD and MSCPA—de-noising and feature extraction and
in the end storing it into final datasets.
In the results for patient chb05, it could be seen that it took around 1500 s to pre-
process interictal training data and around 490 for preictal. There are about 17 h of
recorded interictal training data so it is no wonder that it took around 25 min. For
the preictal and ictal dataset there was about 6 h of recordings processed for around
8 min. For patient chb18 that amount of time is a bit less: 920 s for interictal and
450 s for ictal, so consecutively around 15 and 7 min. There were 15 files taken for
interictal training and four for ictal.
Data was pre-processed very well with around 70% of MSCPA average perform-
ance success and with 95 features extracted by WPD for both selected patients.
Those features were later used for training and testing data and they represent the
signal’s main characteristics. Mean, average power, standard deviation, skewness,
kurtosis and ratio of absolute mean values were calculated for each of the WPD
components we obtained. The only problem is that this process could have been
finished faster if we chose some other tools or techniques for processing. Using
Hadoop or Apache Spark, or perhaps introducing parallel processing would
probably decrease the number of seconds.
The next part was actual training and classifying combined ictal and interictal
prepared data. Matlab GUI was used for this purpose. It is a software based on
WEKA, and we would have used WEKA if it did not have the disadvantage of
inability to store the classifier. There are 10 possibilities for classifier algorithms
provided by Matlab GUI. Seven of them were chosen for the test: Naive Bayes,
SVM, kNN, random forest, random tree, ANN and rotation forest. As can be seen
in the results section, random forest showed the best performance which went even
1-25
up to 99% for patient chb05. If we do not count on Naive Bayes with a lot worse
results (still good though), and support the vector machine with the average result,
classifiers showed really great accuracy, especially for chb05 where they are above
90% accurate. However, some of them performed extremely slowly on given data,
where the training lasted even up to a couple of hours. That happened with SVM,
ANN and rotation forest. Even the leftover three algorithms were tried (out of those
10 mentioned to be provided by Matlab GUI), but they were so slow we could not
even get to the results part. Accuracy of classifiers was calculated with cross-
validation and confusion matrix (theory of sensitivity and specificity), both provided
by Matlab GUI.
In results, statistical classifying data is shown for the winner of the classifiers:
random forest. It was the most accurate algorithm for both patients. For chb05 it
could be seen in detail that only 83 out of 51 060 instances were classified wrongly,
which is a great result. For patient chb18, accuracy for random forest was about
94% with 3621 out of 69 000 falsely predicted results. And it is also important to
point out that the classification for random forest lasted less than 10 min. Those
were all the reasons why this classifier is recommended to be used for this kind of
data, and it was further used for this research for both patients.
The next part was loading the testing data and testing the seizure prediction
algorithm on it. Data was divided into 2048 samples (that is sample window size
really often used for FFT of signals because of its optimal performance), which
means data chunks were 8 s long. All 23 channels of those 8 s segments were then
inverted and loaded into the final matrix. That matrix was then divided into 8 min
long signals, for which we again performed de-noising and feature extraction. 8 min
long data chunks were chosen because the preictal part which had been chosen
before that was 48 min, and in this way it could be divided into six parts and
predictions for each of them could be calculated. Data instances that were obtained
for each of those data chunks, were then classified by random forest classifier
whether to represent ictal or interictal class—positive or negative prediction. If the
number of positive predictions per segment were higher than threshold value (a
constant we choose, by default it is half of the maximum number of instances), and if
there were at least three out of those five chunks with those results, we raised the
alarm.
In the case of patient chb05, it can be seen in the results for both tested seizures
that by using the threshold of 400, alarms were sent off earlier than should be right.
3.5 h for an alarm was over the one-hour-before the seizure limit, even though the
prediction was made. This could be happening for two reasons. The first is threshold.
Modifying the threshold can very much affect the result, for example, decreasing the
number of false alarms. But, in this case, even if the threshold was taken as 450 or
500, that would not change the overall result much, there would be perhaps fewer
alarms and that is it. The second reason why this was happening is probably because
interictal signals taken for the testing part represent the preictal part for actual
seizure that the patient had in the real-life situation. After some extra training, the
interictal data had been added but the system did not recognize it as a part of
the continuation of the preictal part. The solution is as follows: either taking all the
1-26
patient data (this would work well because the model is a bit overfitting), or
decreasing the amount of interictal testing data.
For the second seizure, good results were achieved where seizure was predicted
around 30 min before it started, even though there were some premature predictions.
In any case, results are better than those for the previous model. However, there is
still the necessity of solving the overtraining problem with perhaps better division of
training and testing data.
In the case of patient chb18 there were also two seizures. However, the overall
process had to be repeated twice. The reason for this is modification of the threshold.
In the case with the threshold of 400, there was one false alarm in the beginning.
That would ruin the system right away but was easily fixed by changing the
threshold to 450. After that there was no false alarm in the beginning. However, this
seizure prediction can hardly be counted as successful, considering that it is done too
early (five hours) and that in the preictal period there are no alarms.
For the second seizure the threshold was first 400, and then turned to 450 in
order to lose the false alarm at the beginning. In this case those prediction alarms
were sent off perhaps a bit too early. There was also seizure detection, where
alarms were raised during but not before the seizure. Unfortunately, seizures could
not be predicted properly for this patient’s data. Except for the reasons already
mentioned while discussing patient chb05, we can also add that there were more
consecutive seizures in real-life patient data where each of them have specific
preseizure data. In order to fix this there should be some changes made regarding
which interictal and ictal data to use for training or test. Except for these two,
during the experimentation period we also tried three other patients, with much
more successful results, but because of their specificity this paper covers these two
as the special cases.
The discussion part can be concluded with the execution time of the prediction
system. It can be seen that the testing part lasted about 4–5 h for both seizures of
patient chb05 and 3–4 h for patient chb18. This is understandable considering that
there was 18 h and 16 h of testing data for patients chb05 and chb18 sequentially.
This amount of testing time could be decreased by using other tools and techniques
specified for big data like Hadoop, Apache Spark, or if Matlab is preferred then to
use it with parallel processing.
Overall results are good, but as could be seen there is a lot of space for improvements.
In comparison to other researches based on this topic, combination of algorithms and
data were different in this research. There was no study that used random forest as the
classifier for a system like this at all—other classifiers like rotation forest, ANN or SVM
were preferred. DWT was also preferred instead of WPD usually, which is why WPD is
tried out in this research and it performed really well, for feature extraction as well as de-
noising as included in MSPCA process. All of those previous studies coped with the
same issues that occurred in this one as well, even though some of them used much more
powerful frameworks like Hadoop. They did however have better performance when it
comes to execution time, which is why in further research of the topic those systems like
Hadoop, specified for big data would probably be preferred.
1-27
1.7 Conclusion
In this informational era dominated by technology we are ‘floating’ in the huge
amounts of data and it is imperative to learn how to ‘deal’ with it in order to extract
as much useful information as possible. Healthcare data is perhaps the best example
because it represents the main and most important segment of our lives, which could
be improved a lot if we could make some use of the kept data from health records. In
this research we have been dealing with big data from Children’s hospital in Boston,
EEG signals recorded from 23 children who have epileptic seizures. The goal was to
build a system which can effectively predict new seizures. Many previous works
investigated the topic, only by using different techniques and algorithms. This work
represents a combination of wavelet packet decomposition for feature extraction,
multiscale principal component analysis for de-noising the data—for data prepara-
tion, then random forest classifier (which was around 96% accurate) for training the
data by using Matlab GUI software and in the end testing the written algorithm for
seizure predictions. This algorithm represent independent processing of subsets of
test data whose results in the end represent either positive or negative predictions.
This research was done for patients 5 and 18. In the results it could be seen that
predictions were made. However, predictions were made a lot earlier than they were
supposed to. This is bad, especially for patient 18, where there were no other
predictions in the preictal period. This could be fixed in future by making some
change in the division of training/test data and making sure it is not over-trained.
For patient chb05 there were some predictions around half an hour before the
seizures, which means that those earlier predictions could be considered as false
alarms. False alarms could cause a patient to lose faith in the whole system so this
problem should be overcome in the future. It should also be pointed out that this
system performed generally better on some other patients, and patients 5 and 18
were taken as special cases so we could in future avoid eventual possible problems.
For future work justifications in training/test data could be made, many more
patients could be involved in detailed research and data could be used from other
hospitals. Finally, rather than Matlab, frameworks for big data like Hadoop or
Apache Spark could be used for faster performance. However, even though there is a
lot of room for improvement, this model showed to be very promising and if justified
better in the future can be applied in healthcare.
References
Aarabi A, Fazel-Rezai R and Aghakhani Y 2009 A fuzzy rule-based system for epileptic seizure
detection in intracranial EEG Clin. Neurophysiol. 120 1648–57
Akben S B, Subasi A and Tuncel D 2011 Analysis of EEG signals under flash stimulation for
migraine and epileptic patients J. Med. Syst. 35 437–43
Ali Shahidi Z, Manouchehr J, Guy A D and Reza T 2010 Automated real-time epileptic seizure
detection in scalp EEG recordings using an algorithm based on wavelet packet transform
IEEE Trans. Biomed. Eng. 57 1639–51
1-28
Alickovic E, Kevric J and Subasi A 2017 Performance evaluation of empirical mode decom-
position, discrete wavelet transform, and wavelet packed decomposition for automated
epileptic seizure detection and prediction Biomed. Signal Process. Control 39 94–102
Bakshi B R 1998 Multiscale PCA with application to multivariate statistical process monitoring
AlChE 44 1596–610
Begg R, Lai D T and Palaniswami M 2008 Computational Intelligence in Biomedical Engineering
(Boca Raton, FL: CRC Press)
Berry M W and Browne M (ed) 2006 Lecture notes in Data Mining (Singapore: World Scientific)
Breiman L 2001 Random forests Mach. Learn. 45 5–32
Chisci L, Mavino A, Perferi G, Sciandrone M, Anile C, Colicchio G and Fuggetta F 2010 Real-
time epileptic seizure prediction using AR models and support vector machines IEEE Trans.
Biomed. Eng. 57 1124–32
Chua K C, Chandran V, Acharya U R and Lim C M 2011 Application of higher order spectra to
identify epileptic EEG J. Med. Syst. 35 1563–71
Costa R P, Oliveira P, Rodrigues G, Leitao B and Dourado A 2008 Epileptic Seizure
Classification Using Neural Networks with 14 Features (Berlin: Springer)
D’Alessandro M, Esteller R, Vachtsevanos G, Hinson A, Echauz J and Litt B 2003 Epileptic
seizure prediction using hybrid feature selection over multiple intracranial EEG electrode
contacts: a report of four patients IEEE Trans. Biomed. Eng. 50 603–15
Dutta H, Fiorletta H, Pooleery M, Diab H, German S and Waltz D 2011 A case-study on
learning from large-scale intracranial EEG data using multi-core machines and clusters Proc.
Third Workshop on Large Scale Data Mining: Theory and Applications
Fergus P, Hignett D, Hussain A, Al-Jumeily D and Abdel-Aziz K 2015 Automatic epileptic
seizure detection using scalp EEG and advanced artificial intelligence techniques BioMed
Res. Int. 2015 986736
Ghorbanian P, Devilbiss D M, Simon A J, Bernstein A, Hess T and Ashrafiuon H 2012 Wavelet
transform EEG features of Alzheimer’s disease in activated states 34th Annual Int. Conf. of
the IEEE EMBS
Ghosh-Dastidar S, Adeli H and Dadmehr N 2008 Principal component analysis-enhanced cosine
radial basis function neural network for robust epilepsy and seizure detection IEEE Trans.
Biomed. Eng. 55 512–8
Goldberger A L, Amaral L A, Glass L, Hausdorff J M, Ivanov P C, Mark R G, Mietus J E,
Moody G B, Peng C K and Stanley H E 2000 PhysioBank, PhysioToolkit, and PhysioNet:
components of a new research resource for complex physiologic signals Circulation 101
E215–20
Greco A, Mammone N, Morabito F C and Versaci M 2008 Kurtosis, Renyi’s entropy and
independent component scalp maps for the automatic artifact rejection from EEG data
World Acad. Sci. Eng. Technol. Int. J. Med. Health Biomed. Bioeng. Pharm. Eng. 2 344–8
Harikumar H and Narayanan B S 2003 Fuzzy techniques for classification of epilepsy risk level
from EEG signals Conf. on Convergent Technologies for Asia-Pacific Region TENCON
Haut S R, Swick C, Freeman K and Spencer S 2002 Seizure clustering during epilepsy monitoring
Epilepsia 43 711–5
Ježdík P, Čmejla R, Kršek P and Jahodová A 2010 Distributed computing system for EEG signal
processing using MATLAB as activex object in DCOM model Trans. Res. Biomed. Eng.
Jolliffe I T 2002 Principal Component Analysis (New York, USA: Springer)
1-29
Jukić S and Subasi A 2017 A MapReduce-based rotation forest classifier for epileptic seizure
prediction arXiv: 1712.06071
Jukić S, Кеčo D and Kevrić J 2018 Majority vote of ensemble machine learning methods for real-
time epilepsy prediction applied on EEG pediatric data TEM J. 7 313–18
Kevrić J and Subasi A 2015 Implementation of Novel Signal Decomposition and Data Mining
Techniques in EEG Signal Analysis
Khan Y, Farooq O, Sharma P and Rafiuddin N 2012 Latency study of seizure detection Adv.
Comput. Sci., Eng. Appl. 166 129–36
Liang S-F, Shaw F-Z, Young C-P, Chang D-W and Liao Y-C 2010 A Closed-loop Brain
Computer Interface for Real-time Seizure Detection and Control 32nd Int. Conf. IEEE
EMBS (Buenos Aires, Argentina)
Nasehi S and Pourghassem H 2013 Patient-Specific epileptic seizure onset detection algorithm
based on spectral features and IPSONN classifier 2013 Int. Conf. Communication Systems
and Network Technologies
Qidwai U, Malik A and Shakir M 2014 Embedded fuzzy classifier for detection and classification
of preseizure state using real EEG data 15th Int. Conf. Biomedical Engineering
Rodríguez-Bermúdez G and García-Laencina P J 2012 Automatic and adaptive classification of
electroencephalographic signals for brain computer interfaces J. Med. Syst. 36 S51–63
Schelter B, Winterhalder M, Maiwald T, Brandt A, Schad A, Timmer J and Schulze-Bonhage A
2006 Do false predictions of seizures depend on the state of vigilance? A report from two
seizure-prediction methods and proposed remedies Epilepsia 47 2058–70
Shoeb A H 2009 Application of machine learning to epileptic seizure onset detection and
treatment Ph.D. Thesis Massachusetts Institute of Technology
Stergiou C and Siganos D 1996 Introduction to neural networks (Accessed: 31 March 2004)
Teixeira A et al 2014 Epileptic seizure predictors based oncomputational intelligence techniques:
A comparative study with 278 patients Comp Methods Programs Biomed 114 324–36
Vapnik V 1995 The Nature of Statistical Learning Theory (New York: Springer)
Wang N and Lyu M R 2014 Extracting and selecting distinctive EEG features for efficient
epileptic seizure prediction IEEE J. Biomed. Health Inform. 19 1648–59
Winterhalder M, Maiwald T, Voss H U and Aschenbrenner-Scheibe R 2003 The seizure
prediction characteristic: a general framework to assess and compare seizure prediction
methods Epilepsy Behav. 4 318–25
1-30
IOP Publishing

Chapter 2
Delineation of epileptogenic zone
João Paulo Sant Ana Santos de Souza, Brunno Machado Campos,
Felipe Franco da Graça, Ricardo Brioschi, Larissa Núbia Nunes Vilany,
Wu Shin-Ting and Enrico Ghizoni
Epilepsy is a neurological condition with a considerable rate of refractoriness. For

drug-resistant cases, surgery is considered a valid treatment option. Delineation of
epileptogenic zones is a crucial step for planning surgical resection, however, it
remains an important challenge in medical practice. Basic principles, contributions
and limitations of semiology, scalp EEG, PET, MEG, SPECT, functional MRI,
subdural grids and stereoelectroencephalography (SEEG) are discussed here. In this
sense, the present chapter focuses on the state-of-art of non-invasive and invasive
modalities for the study of epileptic networks and localization of epileptogenic
zones.
2.1 Introduction
In 1993, the epileptogenic zone (EZ) concept was introduced as ‘the area of the
cortex that is necessary and sufficient for initiating seizures and whose removal (or
disconnection) is necessary for complete abolition of seizures’ [1]. This definition
evidenced the efforts to seek a cure in epilepsy and it sounded plausible that there
was a specific region in the brain that could be removed to overcome the disease.
Nonetheless, resection of the supposed EZ is not enough to achieve extinction of
seizures in all cases. Even after the most appropriate surgical treatment, patients
who suffer from epilepsy are not expected to present a risk of seizure recurrence
comparable to the risk of unprovoked seizures in normal individuals [2]. The term
‘cure’ must be used with caution. In substitution, the phrase ‘resolved’ has been
adopted to describe the condition in which a person no longer has epilepsy, although
it does not guarantee that it will not return [3].
Due to the failure in achieving universal seizure-freedom based on such limited
conception, new hypotheses have been raised [4]. The first of interest is based on the
idea that the ictal onset zone (IOZ) is connected with other region(s) of the brain,

yielding an epileptic network in which cortical regions are directly or indirectly

connected and influenced [5]. The second and complementary concept denies the
static nature of the EZ in favor of a dynamic EZ perception. In other words, a
cortical region that currently does not show participation in the primary organ-
ization of the epileptic discharge may become part of a future EZ. This cortical
region is represented by a potential EZ or potential IOZ [6]. Therefore, a complete
remission of seizures would encompass correct identification and resection of the
current EZ as well the potential EZ, rather than the disconnection of the current
epileptogenic zone solely.
The aim of this chapter is to briefly depict the contributions of invasive and non-
invasive techniques to understand the epileptic networks in the brain and to
formulate an anatomo-electro-clinical (AEC) hypothesis for surgical planning.
2.2 Seizure semiology

Localization always played a major role in neurosciences. Many different brain
maps were proposed over the years, with different areas representing a vast myriad
of functions. In the 1930s Penfield first described a detailed map of the motor area
after consistent in vivo experimentation. The same author had an important
contribution to seizure semiology with one of his classic works entitled ‘Epilepsy
and the Functional Anatomy of the Human Brain’, a milestone when other
complementary investigations were scarce [7].
Currently the advances in novel technologies implied in the surgical planning in
epilepsy are undeniable. However, as with any neurological disease, the detailed
clinical history (CH) is cardinal not only in the adequate diagnosis of epilepsy and
ruling out other causes of refractory seizures, but also in determining the EZ
localization [8, 9]. This is true considering its cost effectiveness once it does not
require any advanced method and might help the clinician in directing subsequent
investigations [10].
The International League Against Epilepsy (ILAE) recently revised the opera-
tional classification of seizure types giving special attention to the detailed
description of the episodes. This new publication reinforces the importance of the
first symptoms in classifying the seizure as generalized and focal [11].
For the focal seizures, the main concern of the present chapter, determining the
earlier symptoms is capital once they are the clinical translation of the cerebral
activity in an eloquent area that may be correspondent, or at least close, to the EZ.
A summary of the most common manifestations according to the origin of the
seizure is presented below:
• Temporal lobe seizures (TLS) are the most prevalent focal seizures in adults,
and the main cause of refractory epilepsy and need of surgical treatment. In
mesial temporal lobe epilepsy (MTLE), although some patients are not able
to remember the first symptoms before they lose perception, it is important to
access the information with family members or even in videos recorded by
them once some manifestations are brief. The MTLE classically presents with
2-2
a combination of psychic symptoms such as déjà vu and jamais vu, fear,

hyperventilation, ascending visceral and epigastric sensations, oral and
manual automatisms followed by dystonic posturing with or without bilateral
tonic clonic involvement. On the other hand, in the neocortical temporal lobe
epilepsy (NTLE), due to its different connectivity, auditory hallucinations
and vertigo are far more common and the seizure is usually shorter in
duration [9, 12, 13].
• Frontal lobe seizures (FLS) may have several different presentations, usually
with a shorter duration and with brief or even without post-ictal confusion. It
may present with prominent motor features (such as cycling movements and
walking) and without loss of awareness, and are more common during sleep
or close to sleep hours. Sometimes they present with complex behaviors that
might be confused with non-epileptic seizures. Other features can help the
clinician in a more precise localization in FLS. Speech arrest or other speech
alterations may point to the dominant opercular region while clonic activity
that spreads in a pattern that resembles the somatotopic distribution in the
motor cortex may point to the contralateral Rolandic area (the so-called
Jacksonian march). In a patient without loss of awareness, a seizure
beginning with tonic eye and head deviation points to the contralateral
premotor area [14, 15]. Other frontal areas such as the orbito-frontal region
may present with characteristics that mimic TLE, reinforcing the importance
of brain connections in seizure semiology.
• Occipital lobe seizures (OLS) are not as common as the FLS and TLS. They
typically present with visual symptoms such as elementary hallucinations
(positive), or amaurosis/visual field deficits (negative) and eye movement
sensations. It is important though to distinguish the positive visual symptoms
from a far more common condition, migraine with visual aura. In migraine
the visual symptoms usually last longer than 5 min and the images are black
and white and stripe-like while in OLS they last less and are generally colorful
and round [16].
• Parietal lobe epilepsies are even rarer than the OLS. Symptoms in these
patients may include sensory perturbations such as tingling, burning or
numbness. More complex features, such as sensation of spatial displacement
of a limb were also described [17].
• Insular lobe has a widespread connection with other brain areas, which is why
seizures originating in this region may have an extensive number of different
presentations including autonomic symptoms but also olfactory, gustatory
and somatosensory manifestations [18].
Besides the anatomical localization in lobes it is possible to have clues concerning

the lateralization of the crises with CH. Several publications have accessed possible
lateralizing signs, specially in temporal lobe epilepsy. Table 2.1 briefly presents some
of these findings.
2-3
Table 2.1. Some of the seizure characteristics that have been described in literature as having value in EZ
lateralization/localization [10, 11, 19, 20]. TL = temporal lobe, EZ = epileptogenic zone.
Clinical finding Most likely EZ lateralization/localization
Ictal asymmetric tonic limb posturing TL ipsilateral to the flexed limb

(figure of 4 sign) (patient ‘points’ to the EZ)
Ictal distonic/tonic limb posturing Contralateral to the limb
Peri-ictal water drinking Nondominant hemisphere
Unilateral eye blink Ipsilateral to the blinking eye
Ictal urinary urge Right temporal lobe
Postictal nose wiping TL ipsilateral to the limb
Gelastic seizure (Ictal laugh) Hypothalamus
2.3 Scalp electroencephalography (EEG) and long-term video-EEG

(VEEG) monitoring
2.3.1 Introduction
Electroencephalography (EEG) is a technique able to record the electrical activity of
the human brain and, therefore, it is an important instrument for the evaluation of
cerebral functioning. A standard EEG graph shows voltages on the vertical line and
time on the horizontal line, providing a real time exhibition of cerebral activity.
EEG is extensively used in evaluating consciousness impairment, seizures and
epilepsy, but its value goes beyond these classical applications. For example, it
can be used to monitor the depth of anesthesia in surgical procedures [21]. In this
section, we aimed to discuss the clinical importance of non-invasive EEG and long-
term video-EEG (VEEG) monitoring in the delineation of the ictal onset zone, and
its relation with the epileptogenic zone.
2.3.2 Terminology
The irritative zone (IZ) consists in the area of cortex capable of generating spikes or
sharp waves in the EEG, both characterized by a pointed peak and usually of
negative polarity. These electrophysiological alterations are waveforms distinct from
brain background activity, and are called interictal epileptiform discharges (IED)
(figures 2.1 and 2.2). These electrophysiological signs express an increased risk of
epilepsy [22]. While spikes have less than 80 ms, sharp waves have duration of 80–
200 ms [23]. IZ is not the same as EZ, but commonly overlaps with it, and it is a
good indicative of the region of seizure origin at the cortex [22]. In addition, high-
frequency oscillations (e.g. ripples, fast ripples) observed during interictal periods
were also described as potential biomarkers of the epileptogenic tissue [24]. Besides
that, there is an area of cortex that generates clinical seizures, so called ictal-onset
zone (IOZ) [25].
Symptoms occur when the electrical discharge affects an area of cortex function-
ally active. Therefore, it is plausible that the IOZ would present clinical features only
2-4
Figure 2.1. Interictal EEG of a 35 year-old patient with left temporal epilepsy. Scalp electrodes are attached
according to the International 10–20 system. Here we have indicated the IED, represented by sharp waves.
Images courtesy of Dr Fernando Cendes (University of Campinas). EEG = electroencephalography,
IED = interictal electrical discharges.
Figure 2.2. Interictal EEG of the 35 year-old patient from figure 2.1. Scalp electrodes are attached according to the
International 10–20 system. Here we have indicated the IED represented by a spike in the region of left temporal
lobe. Images courtesy of Dr Fernando Cendes (University of Campinas). EEG = electroencephalography,
IED = interictal electrical discharges.
if the ictal discharge spread to a symptomatogenic zone (SZ), defined as the ‘area of
cortex which, when activated, produces the initial ictal symptoms or signs’ [26]. This
region of the cortex can be determined by analyzing the initial seizure symptoma-
tology and the VEEG has a great value in this context.
Finally, the functional deficit zone is the ‘area of cortex that is not functioning
normally in the interictal period’ [26]. Neurological, neuropsychological and func-
tional imaging tests can be used to define this region.
2-5
2.3.3 Applications to epilepsy

Cortical pyramidal neurons, oriented perpendicularly to the brain’s surface, produce
the EEG. The summary of the excitatory and inhibitory postsynaptic potentials of
several groups of neurons shooting in a synchronous way is detectable by the EEG.
Multiple factors can interfere in the expression of epileptiform activity on non-invasive
EEG. Cerebral activity is attenuated by numerous biological filters, like meninges,
cerebrospinal fluid, skull, and scalp. The EEG is spread out more than its original
source. Other biologically generated electrical activities create voltage potentials that
distort the original cerebral activity. The location of the generator, recording
technique, state of the patient and seizure type are other biases. Thus, scalp EEG
can fail to record some IED, mainly when arising from deep or midline areas.
In spite of the caveats revised here, non-invasive EEG recordings still provide a
moderate assessment of the IOZ. It is considered an imperative step of the pre-surgical
evaluation [26]. Because seizure has generally spread outside the IOZ, it is assumed
that the interictal EEG furnishes more reliable localizing information than ictal EEG.
Evolution from the interictal to the ictal state is variable, so it may be hard to identify
this transition, chiefly when ictal activity is brief [27]. When the IED evolves into focal
seizure, we usually see sinusoidal activity in the beta, alpha, or theta range or repetitive
epileptiform discharges. The latter may advance in frequency, field, or amplitude
[28–32]. Isomorphic patterns, like repetitive IED, are not typical in focal epilepsy, but
can be present in some idiopathic generalized epilepsies [27].
The value of the EEG is greater in patients with temporal lobe seizures compared
to seizures of extratemporal origin. Walczak et al used rhythmic theta and alpha
activity, postictal slowing and activity at seizure onset to lateralize the IOZ;
47%–65% of extratemporal seizures were correctly lateralized while temporal lobe
seizures had an accuracy of 76%–83% [33]. Foldvary et al studied about 500 focal
seizures from 72 patients, and concluded that non-invasive EEG has correctly
localized 72% of cases, being more accurate for temporal than for extratemporal
lobe epilepsy [34].
2.3.3.1 Temporal lobe epilepsy (TLE)

In temporal lobe epilepsy, IED produces a pattern with the highest amplitudes at the
anterior temporal electrodes [35, 36]. Surface electrodes are not able to detect a great
number of mesial and neocortical temporal IEDs, although it was well-established
that IED has a great value in localizing temporal epileptogenesis. Furthermore, IED
confined to the anterior temporal region was correlated with a good postoperative
outcome in TLE surgery [37, 38].
In 80%–90% of patients with MTLE or unilateral hippocampal sclerosis (HS),
rhythmic theta or alpha activity (lateralized or more regularly localized to the
temporal region) occurred within the first seconds of electrographic or clinical onset
[31, 34, 39, 40]. Seizures initiating with rhythmic temporal theta activity is not
characteristic of none of the seizures arising from extratemporal zones [41].
Rhythmic theta at ictal onset is more common in patients with moderate or severe
hippocampal atrophy [42].
2-6
Today, a decision for surgical treatment is indicated for patients with MTLE
based on structural MRI (presence of unilateral hippocampal atrophy), routine
EEG, and the clinical history. In patients with focal epilepsy and a negative MRI,
other tests should be done [43].
2.3.3.2 Frontal lobe epilepsy (FLE)

More than a third of patients with frontal lobe epilepsy may not present any IED,
making clear the limited value of IED in extratemporal epilepsy [44, 45]. When
identified, IED is less well localized, multifocal, or even generalized [44, 46].
However, Ramantani et al designed a study comparing simultaneous subdural
and scalp EEG, and demonstrated that not only dorsolateral frontal, but also
orbitofrontal and mesial-frontal sources can be detectable in FLE with scalp EEG
[47]. A poor postoperative outcome was correlated with the pattern of generalized
IED, while a good prognostic was associated with unifocal IED [48, 49].
Ictal EEG in frontal lobe complex partial seizures can show a brief duration,
starting and finishing abruptly. Corresponding seizure is characterized by excessive
movement or tonic posturing, frequently resulting in obscuration of the EEG
[45, 50]. A third of cases do not show alteration in the EEG, and another third
present non-lateralized slowing, rhythmic activity, or repetitive spiking [50, 51].
Lateralized or localized patterns are observed in 33%–50% of the cases [45, 52].
Other patterns that have been described consisted of rhythmic alpha, beta, or theta
at or adjacent to the midline in the supplementary motor epilepsy, and generalized
suppression or paroxysmal fast activity as the pattern of ictal onset [34, 53].
2.3.3.3 Parietal lobe epilepsy (PLE)

Parietal lobe epilepsy is less prevalent. IEDs in these cases are typically widespread
and multifocal, or even bilateral [54, 55].
The epileptiform discharge can propagate to the temporal or frontal lobe. Then,
the ictal EEGs vary according to pathway discharge and may suggest a wrong
localization. Some alterations found in PLE are diffuse suppression, non-lateralized
rhythmic activity or lateralized changes [27, 55].
2.3.3.4 Occipital lobe epilepsy (OLE)

A great number of patients with occipital lobe epilepsy have electrophysiological
alterations, but epileptiform activity arising from the inferior and mesial occipital
surfaces is difficult to detect. Decreased alpha activity or focal slowing indicate
occipital dysfunction, and may help in OLE. Spikes and sharp waves in temporal or
temporo-occipital regions are the most frequent IEDs in this type of epilepsy
[56, 57]. Epileptiform activity restricted to the occipital lobe is not common,
different from the widespread and bilateral IED [57–59].
The electrophysiological and clinical expressions of OLE depend on the pathway
of discharge, as in the parietal lobe epilepsy. Ictal EEG commonly shows diffuse
suppression or rhythmic activity. The latter is usually generalized but may be
lateralized or maximal over the temporo-occipital region [58, 59].
2-7
2.3.4 Long-term video-EEG monitoring

EEG monitoring associated with a simultaneous video recording is usually
employed in epilepsy centers to elucidate the case. Prolonged samples of interictal
EEG data can be produced and analyzed, helping the diagnostic labor. VEEG is
often done after hospital admission, in a safe environment, under drug withdrawal
or sleep deprivation (pro seizure factors) [21].
This technique is more indicated for patients in need of clinical characterization
or seizure localization. In certain groups of patients, the diagnosis of epilepsy may be
challenging, due to atypical seizure semiology and/or infrequent clinical episodes.
Common non-epileptic mimickers of epilepsy, such as psychogenic non-epileptic
spells and syncope, can hinder the diagnosis. Some studies demonstrated that VEEG
leads to altered diagnosis in 45%–58% of patients and modifications in management
in approximately 75% of monitored patients [60]. In conclusion, VEEG remains a
useful study for seizure classification and localization of the EZ [21].
We defined previously the SZ as the area of cortex that when activated, produces
the symptomatology of epileptic seizures. The initial seizure symptoms are able to
provide fundamental information regarding SOZ localization. Auras, for example,
reflect the activation of the SZ, which is presumably close to the SOZ [61]. However,
if the seizure was born in a silent cortical region, and did not suffer propagation, the
epileptic activation would remain unnoticed.
Concomitant video analysis of epileptic seizures to localize and lateralize the SZ
was the object of numerous studies. It was observed that dystonic hand posturing
and forced head version could indicate a seizure onset in the contralateral hemi-
sphere [62, 63]. Unilateral manual automatisms were related with ipsilateral seizure
onset. Postictal aphasia was suggestive of a seizure onset in the speech-dominant
hemisphere [64], while preserved responsiveness during ictal automatisms was
detected in patients with temporal epilepsy of the non-speech dominant hemisphere.
Noachtar and Luders concluded that the positive predictive value of these signs of
lateralizing is greater than 80% [65].
However, a recent meta-analysis evaluated VEEG utility in identifying the EZ in
candidates for surgery [66]. Sensitivity and specificity were 70% and 40%, respec-
tively. In the subgroup of lesional TLE, sensitivity was higher (85%) compared to the
sensitivity (47%) observed in the lesional extra-temporal lobe epilepsy (ETLE)
group. In lesional TLE patients, the rate of seizure freedom was 74% when VEEG
was concordant with the site of surgical resection. In summary, the meta-analysis
showed that scalp EEG recordings are not enough in many occasions, such as in
lesional ETLE, requiring further invasive investigation.
2.4 PET-CT
2.4.1 Introduction
Positron emission tomography (PET) is a clinical nuclear medicine imaging modality
extremely sensitive to the positron-emitting radionuclide. The positrons interact with
surrounding electrons and release two photons detectable by the PET detector.
2-8
Injecting into a body small amounts of biologically active molecule (radioactive

tracer), it is possible to image the tissue concentration of the molecules of interest, and
hence to get a better understanding of its physiological function. If the radiolabeled
tracer is 18-fluorodeoxyglucose (FDG), a modified form of glucose, the imaged
concentration of the tracer may reflect the tissue metabolic activity [67].
Indeed, glucose is the main energy source of the brain tissue: 95% of the ATP
required for brain function is derived from this molecule. Once the tracer is injected
into the peripheric venous system, it reaches neuronal cells through an uptake
mediated by glucose transporters (GLUT), mainly type 1 (GLUT1) [68]. Within the
intracellular environment, the tracer undergoes a phosphorylation by the hexokinase
enzyme, resulting in the formation of FDG-6-phosphate. Not further processed in
significant magnitude, this metabolite remains trapped by the cell and leads to
imaging contrast.
Correlating the regional activities to the organ anatomy is, however, highly
desirable for the anatomical location of abnormal metabolic activities. Mounting a
CT scanner in the gaps of the rotating banks of bismuth germinate (BGO) block
detectors offered a hardware alternative to scan anatomical and functional data
simultaneously [69]. Today, almost all PET scans are performed together with CT
scans for more accurate diagnoses in clinical practice [70, 71]. In terms of
acquisition, the temporal resolution of PET-CT is approximately between seconds
to minutes [70].
Protocol to 18-FDG PET/CT encompasses the following steps:
• Between 4 to 6 h of fasting before the exam acquisition.
• Avoiding drugs that could interfere with the metabolism of the tracer, such as
insulin, corticoids, active psychotropic substances, caffeine, among others.
• Hydration and normal blood glucose levels (<150–200 mg dL−1) prior to
acquisition.
• Continuous EEG recording for patients, ideally 2 h before and until 20 min
after tracer injection. Time after a seizure interferes importantly in the glucose
uptake of the EZ.
• Resting for 30 min with no stimulations for the brain, in a quiet and dimly-lit
room. Patient must be oriented not to speak, read or be otherwise active.
• Radiolabeled tracer injection.
• Resting for at least 30 min after the latter step [72].
During a seizure occurrence, the EZ plays an electrical discharge that increases

the glucose uptake of this region. As a long time passes, the EZ returns to its
predominant metabolic state, such that the glucose uptake remains decreased.
Depending on the type and location of the epileptic lesion, it may take as much
as 48 h after a seizure for the brain to reach its baseline metabolic state [73].
The metabolic state of the region of interest (ROI) may be evaluated by different
means, including qualitative and quantitative assessment. In the first method, a
colored scale may define the intensity of the metabolic activity, in which hot colors
represent high metabolic regions and the cold colors represent the hypometabolic
ones. Alternatively, the metabolic active may be calculated by the standard uptake
2-9
value (SUV), which is defined by the ratio of average radioactivity concentration

obtained for each region to the injected dose per subject weight [74]. In other words,
SUV may be quantitatively given by SUV = ar′ , where r is the radioactivity
(w)
concentration within the ROI, a′ is the decay-corrected amount of injected tracer,
and w is the weight of the patient [75]. Metabolic changes may also be evaluated by
comparing SUV of analogous ROIs from different sides of the brain (SUV ratio).
In this instance, an area may be considered hyper- or hypometabolic compared to
the contralateral one as the SUV ratio between them increases or decreases,
respectively [76]. Semi-quantitative and quantitative analysis are important not
only to conclude for the seizure lateralization, but also to reduce intra- and
interobserver variability [77, 78].

In general, PET-CT is a desired tool when scalp-EEG and MRI fail to detect the
epileptogenic focus. The individual contribution of PET-CT for the diagnosis of
epilepsy vary according to the etiology. In a series of 97 patients with mesial
temporal lobe epilepsy (MTLE) due to hippocampal sclerosis who underwent
anteromesial temporal lobe resection, 85% achieved Engel I outcome, (mean
follow-up of >6 years). The presence of focal anteromesial temporal hypometabo-
lism was associated with Engel IA, whereas extratemporal metabolic changes were
associated with non-IA outcomes. In the right side, suboptimal outcomes were
associated with the involvement of ipsilateral mesial frontal and perisylvian
hypometabolism. Contralateral frontoinsular hypometabolism and posterior white
matter hypermetabolism showed association with poorer seizure control in cases of
left HS [79]. Another series of 64 patients with HS, with no other lesion shown by
MRI, and who were operated on with anterior temporal lobectomy, demonstrated
that extemporal glucose hypometabolism was a common finding, occurring most
frequently in the ipsilateral insula and frontal lobe, and that the extent of remote
hypometabolism was an independent predictor of surgical seizure control [80]. Irem
and colleagues studied a series of 141 patients who underwent anterior temporal
lobectomy (ATL), including 24 patients with negative MRI and FDG-PET positive
for unilateral temporal hypometabolism, as well as 117 patients with MRI positive
for mesial temporal sclerosis (MTS) and FDG-PET positive for unilateral temporal
hypometabolism. Comparisons between both profiles of patients showed no differ-
ences in Engel I surgical outcomes, suggesting PET as an important complement in
cases of MRI negative cases [81]. Concerning cases of TLE secondary to low-grade
tumor, concordance between PET and MRI was inferior to that showed between
256 channel dense-array electroencephalography source imaging (39.4% versus
72.7%) [82].
In a series of 29 patients suffering from FLE, with good seizure control (Engel I or
II) after surgery, FDG-PET/CT showed a sensitivity of 36% in patients without
structural lesion on MRI, while the sensitivity was 73% among those with structural
lesion as evidenced by MRI. According to pathological findings, etiology included
2-10
22 cases of cortical dysplasia, four cases of tumors, one case of cortical scars, and
two cases with old infarct [78].
Another interesting study evaluated the efficacy of applying a new patient
selection strategy for MRI-negative neocortical epilepsy surgery. This cohort
encompassed 153 patients who underwent ictal scalp EEG, interictal scalp EEG,
FDG-PET, and ictal-interictal SPECT. Authors compared two strategies for
surgical selection: at least one localizing result of presurgical studies, and concordant
results of two or more presurgical studies. The study demonstrated that the latter
strategy significantly improved the rate of seizure-free (75.5% versus 47.2% of
Engel I) and that concordance between two or more presurgical studies and PET
was related to seizure-free outcome.
To investigate the overall contribution of FDG PET in helping decision making
for epilepsy surgery, Rathore and colleagues evaluated 194 consecutive patients in
which MRI was normal or discordant with clinical and EEG data. Final localization
encompassed TLE (n = 64), FLE (n = 66), temporal plus epilepsy (n = 26) and other
ETE (n = 38). Abnormal PET was more frequent in TLE (63%), followed by ETE
(61%), and FLE (52%). PET data were useful in 103 (53%) patients, once it led
directly to surgery in 12 (6%) patients, helped planning intracranial EEG in 67 (35%)
patients and excluded 24 (12%) patients from further evaluation. In the same study,
focal hypometabolism on FDG PET increased the odds of being selected for surgery
or intracranial EEG five fold.
In a series of 26 subjects with parietal lobe epilepsy, Kim et al evaluated the
diagnostic sensitive of distinct modalities of non-invasive presurgical evaluation and
the surgical outcomes. The majority of cases (17 out of 26) included focal cortical
dysplasia (FCD) as revealed by pathological analysis, 14 patients achieved seizure-
freedom outcome, and 22 patients showed favorable seizure control. In the seizure-
free group, localization sensitivity was 64.3% by MRI, 50% by PET, 45.5% by ictal
SPECT [83].
Yang et al reported their experience in 35 consecutive cases of OLE secondary to
multiple causes and referred to surgical resection. In their sample, 30 patients
underwent invasive electroencephalography, 28 achieved Engel I or II, and 31
patients performed interictal FDG-PET preoperatively. From those who performed
PET, 27 (87.1%) patients showed lesional or perilesional hypometabolism, whereas
only four individuals showed normal or nonspecific findings [84]. A smaller study
conducted with OLE patients who underwent invasive monitoring and resective
surgery demonstrated that interictal FDG-PET correctly localized the epileptic
lesion in 8 of 16 seizure-free patients, and in three of nine non-seizure-free patients
[85].
A novel technique has tried to use an overlap of PET and MRI scans. However,
some results do not show any superiority either in terms of image quality or in
accuracy for localization of seizure focus compared to PET-CT [86]. A study
evaluated 50 patients with diagnosis of focal cortical dysplasia (FCD), in which the
hybrid PET-MRI showed metabolic changes, mainly hypometabolism, superim-
posed with MRI abnormal cortex in most of them [87]. It seems to be a promising
2-11
technique, but current literature still lacks information regarding PET-MRI

parameters for predicting surgical outcomes.
2.4.3 Conclusion
In conclusion, PET/CT has been considered to be a useful tool for the delimitation of
the EZ in different locations. The hypometabolism showed by the method in the
interictal phase is the most important finding for surgical planning and predictive
considerations. Nonetheless interictal hypometabolic regions may exceed the EZ or
the EZ may not be completely inside the hypometabolic region, motif by which PET
scan should be taken in conjunction with other non-invasive and invasive studies for
surgical planning [88, 89].
2.5 Magneto-electroencephalography (MEG)

2.5.1 Introduction
Magnetoencephalography is a modality of functional imaging. It is based on the
measuring of magnetic fields generated by massive synchronized intraneuronal
electric currents, being part of the preoperative assessment as well [90, 91]. Electrical
charges in movement produce a magnetic field oriented at a right angle to the flow of
current, the direction of which can be determined by the right-hand rule [92]. During
a MEG study, the patient is placed in a system equipped with 150–250 sensors
(or more) that cover the entire head and register the magnetic flux every 1–4 ms. An
isofield map in the sensor space is constructed and ultimately coregistered on the
patient’s high-resolution MRI (usually T1-weighted sequences) [93]. By doing so,
MEG is capable of informing areas that present interictal magnetic spikes and, then,
are suspected of involvement with an epileptogenic region. Tovar-Spinoza analyzed
the distribution of MEG spike sources (MEGSS) according to its number and
density. They defined as a ‘cluster’ the region comprising ⩾20 MEGSS within an
area 1 cm in diameter. In their experience, a single cluster frequently correlates with
the seizure onset zone, part of the symptomatogenic zone, and the active irritative
zone on intracranial recording [94, 95]. MEG can be performed also in the ictal
phase, although some challenges may be overcome for obtaining reliable results,
such as: seizure must occur during MEG recordings, the head movement must be
compensated, and movement-related magnetic artifacts must be suppressed.
Nonetheless, ictal MEG estimates IOZ with higher sensitivity and specificity than
interictal dipole clusters in the lobe resolution [96].
2.5.2 Applications and limitations

Despite many limitations, surgical outcomes from patients evaluated with MEG
show a valuable contribution for surgical planning. In a study conducted with
41 patients (29 cases of TLE and 12 of ETLE), interictal MEG was compared with
invasive monitoring in terms of accuracy to localize the EZ. Authors considered that
each method localized correctly the EZ when the hypothesized EZ overlapped
perfectly with the resected area and the patient was seizure-free postoperatively.
2-12
Interestingly, there was no statistically significant difference in the correct local-

ization of EZ between both modalities. Invasive monitoring and MEG localization
judgments were correct in 54% and 56% of the cases, respectively. However, results
suggested that MEG can be less beneficial relative to invasive monitoring in ETLE
than in TLE. One limitation that must be pointed out in the study is the absence of
comparisons between MEG and different submodalities of invasive monitoring [97].
The ability of MEG to localize EZ, however, seems to depend on the location of
epilepsy [98–100]. A study conducted with 38 patients found a concordance rate
between both interictal MEG spikes and interictal ECoG spikes ⩾90% in the
interhemispheric and fronto-orbital region, approximately 75% in the superior
frontal, central and lateral temporal regions, but only approximately 25% in the
mesial temporal regions [98]. Santiuste et al also verified an important difference
concerning the ability of MEG to detect spikes in mesial temporal lobe structures
and the non-mesial ones. Using invasive depth electrodes as reference, MEG was
able to detect and localize 95% of the neocortical spikes, but only 25%–60% of the
mesial spikes. Specifically, MEG detected 50% and 63% of the interictal spikes from
amygdala and hippocampus, respectively [101]. Less enthusiastic groups, in
contrast, report that MEG failed to detect paroxysmal activities localized in the
mesial temporal structures [100, 102].
There are many reasons for the inability of MEG in detecting interictal spikes.
Deeper structures need to generate many-fold stronger electrical energy on the
superficial cortex to present an equivalent probability of detection by this imaging
method [103]. MEG’s sensitivity also depends on the orientation of electrical
current, not identifying electrical sources purely orthogonal to the skull surface.
Because of such phenomenon, electrical discharges in the sulcal depths and gyral
crest, which tend to produce more radial currents, are less likely to be identified. On
the contrary, the more tangential currents from sulcal walls are, the more easily they
are detected [92, 103, 104]. Lastly, interactions between simultaneous electrical
sources in the brain may lead to a mutual cancelation of the magnetic fields. This is
on the basis of understanding the low sensitivity of MEG for MTLE: the spiral
anatomy of the hippocampus would produce magnetic fields that cancel each other
out [100].
Summarizing, MEG has been described as an important non-invasive study for
surgical planning. Nonetheless, its usefulness varies according to the localization of
seizure onset. The accuracy of MEG is higher for superficial and neocortical
structures, than for deep and mesial temporal lobe regions. Therefore, MEG is a
helpful complement but not a substitute for invasive monitoring.
2.6 Single photon emission computed tomography (SPECT)

2.6.1 Introduction
SPECT is a nuclear medicine imaging that enhances the tissue perfusion.
Analogously to PET, it detects the energy emitted by the radionuclide. Unlikely
PET, it measures directly the decay of the gamma rays (photons) emitted by a
gamma-emitted radionuclide. It has less contrast and spatial resolution, and hence
2-13
lower sensitivity. But, its radiotracers have longer half-lives and it is cheaper and
smaller [105]. So, SPECT is much more available and widely used in clinical
practice. It is well-established for the presurgical evaluation of patients with
refractory epilepsy [106]. In terms of acquisition, the temporal resolution of
SPECT is in the order of minutes, much longer than a PET scan.
There are two tracers commonly used in clinical practice: 99mtechnetium-
hexamethylene propylene amine oxime (99mTc-HMPAO) and 99mtechnetium-ethyl
cystainate dimer (99mTc-ECD) [107]. Both are lipophilic in nature and cross the
blood brain barrier (BBB) to get into the intracellular environment. Within the
neuron, the former is converted into a non-diffusible hydrophilic compound and
the latter undergoes an enzymatic conversion to ionized acid metabolites, leading to
tracer retention [108]. Differences between tracer metabolization and clearance
explain correspondent differences in localizing epileptogenic zones. In a small series
of 54 patients, hyperperfusion observed from 99mTc-HMPAO was significantly
higher than that from 99mTc-ECD in cases of TLE, while higher hyperperfusion and
sensitivity were observed with the first tracer in cases of neocortical epilepsies [109].
To understand the dynamic of brain blood perfusion and SPECT results
interpretation it is valid to recall the principles presented in the PET-CT topic.
During the epileptic electrical discharge, seizure onset zone shows an important
increase of neuronal activity, leading to an increase in glucose uptake, but also in the
regional cerebral blood flow (CBF) [110]. Afterwards, CBF returns to baseline
values or, in other words, it reduces importantly.
To improve the sensitivity of SPECT in locating the SOZ, Subtraction Ictal Single
photon emission computed tomography CO-registered to magnetic resonance
imaging (MRI) (SISCOM) was developed at the Mayo Clinic [111]. It consists in
subtracting two normalized SPECT studies, one ictal and the other interictal,
followed by co-registering the result with a high resolution anatomical MRI. The
interictal one may be acquired at patient admission, while the acquisition of the ictal
one is much more challenging. Ideally, the injection of the tracer must occur as fast
as possible after the beginning of a seizure episode. It is valid to monitor the patient
in order to act fast for tracer injection, which must be done in bolus. A recent study
revealed that tracer injection performed ⩽25 s after seizure onset optimized SPECT
results in pediatric patients with epilepsy [112]. It is recognized that the tracer for
injection must be at the bedside and, if possible, with trained nursing staff [113, 114].
This is particularly important considering that the tracer takes about 30 s to reach
the brain, besides the fact that post ictal switch (i.e. alteration from ictal hyper-
perfusion to postical hypoperfusion) takes approximately 1–2 min in temporal lobe
seizure, being even shorter in extratemporal seizures [108]. After radiotracer
injection, the patient may or may not be sedated to perform SPECT acquisition.
Finally, ictal–interictal subtraction of SPECT scans and co-registration with MRI
are required to localize areas of ictal blood flow change.
STATistical Ictal SPECT COregistered to MRI (STATISCOM), also developed
at the Mayo Clinic, is a newer technique that determines, on the basis of the
statistical parametric mapping [117], whether the ictal–interictal subtraction
2-14
difference is statistically different from the expected random variation between two
SPECT studies.

The usefulness of SPECT in conjunction with other non-invasive studies for surgical
planning has been recognized, but its individual power to help in decision-making
varies according to the protocol used in the study and the location of the supposed
EZ, besides it presents many limitations [115]. In the case of MTLE, ictal SPECT
correctly lateralized 80.6% of the cases with unilateral interictal spikes on EEG,
whereas this rate was only 55.0% in cases of bilateral interictal spikes [116]. In
another study, a group developed a quantitative manner to analyze whether the
ictal–interictal subtraction difference was statistically different from the expected
random variation between two SPECT studies. Then, the STATISCOM technique
was compared to the SISCOM method in a series of 87 consecutive patients with
TLE who underwent temporal lobectomy. As results, STATISCOM correctly
localized the TLE subtype (mesial versus lateral neocortical) in 68% of the patients,
which was significantly higher than the 24% using SISCOM. The probability of
seizure-free outcome was higher when STATISCOM correctly determined the TLE
subtype compared with cases in which subtype was indeterminate [117]. In a meta-
analysis which included seven studies regarding the participation of SPECT onto
patients with TLE undergoing surgical treatment, none has shown the importance of
this non-invasive technique in the postoperative prognosis, and in the case of MTS it
provided redundant information and/or did not assisted in surgical planning.
Considering patients with unremarkable MRI findings, the ictal EEG was associated
with better surgical outcome than SPECT [118]. In a setting of different types of
TLE, ictal SPECT showed distinct patterns of cerebral perfusion: hyperperfusion in
ipsilateral mesial and lateral regions in cases of lesional MTLE, bilateral temporal
lobe hyperperfusion in cases of foreign-tissue lesion in the lateral temporal lobe
epilepsy, and hyperperfusion restricted to ipsilateral anteromesial temporal region in
cases of non-lesional TLE [119]. In this study, ictal SPECT showed itself to be
helpful for subclassification of TLE, providing an insight into preferential pathways
of seizure propagation [119].
Concerning FLE, studies with SPECT showed its potential to localize seizures in
pediatric patients. Among 22 children with electroclinical features of FLE, ictal
SPECT showed unilateral frontal hyperperfusion in 20 (91%) of them, while
interictal SPECT showed unilateral frontal hypoperfusion in only two patients.
Besides, ictal hyperperfusion in frontocentral, medial frontal, and dorsolateral
frontal regions as shown by SPECT was significantly associated with asymmetric
tonic posturing, contralateral head/eye deviation, and unilateral clonic jerking.
However, ictal SPECT localization to the frontopolar and orbitofrontal was not
statistically associated with signs and symptoms [120]. A case report of a patient
with right frontal astrocytoma showed the results of two ictal SPECTs of this
patient. After ictal–interictal subtraction, lateral frontal hyperperfusion was seen
during a left face clonic seizure, while mesial frontopolar hyperperfusion was
2-15
observed during a subclinical seizure. This report suggested the importance of

simultaneous EEG for the correct interpretation of SPECT in epilepsy [121].
Regarding non-MTLE, a retrospective study analyzed 81 consecutive patients
with neocortical epilepsy who achieved favorable seizure-control after surgery,
including 36 patients with normal MRI. The authors then compared the perform-
ance of SPECT using ictal-interictal side-by-side visual analysis and subtraction
analysis. Capability of correctly localizing the epileptogenic lobe was higher with
subtraction than with side-by-side visual analysis (63.0% versus 58.9%), while the
diagnostic sensitivity using both methods in conjunction was 79.0%. Ictal SPECT
correctly localized the epileptogenic lobe more frequently than the subtraction
method when scalp EEG at the time of tracer injection revealed a localizing pattern.
An injection delay <20 s after seizure onset correlated with correct localization when
the subtraction image method was used. In cases of non-localizing/non-lateralizing
EEG at seizure onset, the subtraction method was superior to the visual analysis for
localizing FLE and PLE [122].
For parietal epilepsies, SPECT has also shown its importance. Ho and colleagues
studied 14 patients with PLE, nine of whom had structural lesions. Ictal SPECT
showed focal areas of parietal hyperperfusion in all 14 cases, corresponding with the
sites of structural lesions. Comparing ictal with interictal scans, perfusion increased
a mean of 25.5% in the parietal affected regions. Ictal SPECT localization also
correlated with two seizure patterns: an anterior syndrome characterized by
sensorimotor manifestations and a posterior syndrome characterized by complex
partial seizures (CPS) of the psychoparetic type [123].
Limitations of SPECT to localize EZ are also relevant in cases of OLE. Within a
subseries of 10 patients with drug resistant OLE, ictal SPECT failed to localize the
supposed EZ in seven of them [124]. A study conducted by another group analyzed a
series of OLEs who underwent ictal SPECT and intracranial monitoring, obtaining
similar results. Ictal SPECT correctly localized the lesion in only three out of
12 patients who achieved seizure-freedom outcome after surgery, and in four of
seven non-seizure-free patients [85].
For comparative purposes, ictal SPECT shows performance inferior to interictal
FDG-PET in cases of OLE. Kim and colleagues compared both methods concern-
ing the ability for lateralization and localization of the EZ. Epileptogenic hemi-
spheres were correctly lateralized in 13 of 17 patients with SPECT and in 14 of 15
with PET. However, SPECT correctly identified the EZ in only two (29%) of 17
patients against nine (60%) of 15 patients by interictal FDG-PET [125].
Interestingly, inferiority of SPECT in localizing the EZ in comparison with PET
goes beyond cases of OLE. An explanation is that the interictal phase is accom-
panied by more reduction in glucose metabolism than in reduction of blood flow
[114]. One study included 118 consecutive patients with multiple cases of
intractable epilepsy, secondary to multiple etiologies, in multiple locations
(MTLE and neocortical epilepsies). Using the pathological diagnosis as the standard
of reference, the rate of correct lateralization of ictal SPECT was 73%, against 85%
shown by PET and 72% shown by MRI. Considering only patients with good
outcomes after surgery, MRI, PET and ictal SPECT were correct in 77%, 86% and
2-16
78%, respectively. These results showed that SPECT and MRI had similar perform-
ances [126].
In conclusion, SPECT may be used as a complementary tool to evaluate the
lateralization of the EZ, although its individual contribution for delineation of the
epileptogenic regions is inaccurate to guide surgical resection or to avoid surgery.
2.7 Functional MRI (fMRI)

2.7.1 Introduction
Functional MRI is a modality sensible to changes on the oxygenation level in the
brain. The blood-oxygen-level dependent signal (BOLD signal) allows indirect
inferences regarding task-related brain activations and even activations and network
patterns during the rest condition. The brain does not have internal reserves of ATP
and oxygen [127, 128]. When a group of neurons is activated, an associated
hemodynamic response is triggered to supply its demands of energy and oxygen,
decreasing significantly the local concentration of deoxyhemoglobins (figure 2.3).
Due to the distinct magnetic ordering of the oxyhemoglobin (diamagnetic) and
deoxyhemoglobin (paramagnetic) the decreasing relative concentration of the
second increases the local MR signal, resulting in an indirect metabolic contrast
[127, 128].
In terms of imaging, the fMRI is a series of MRI images that may include the
whole brain and is acquired repeatedly and sequentially. The temporal resolution
(time to acquire one image of the whole brain) is around two seconds and the
number of sequential acquisitions may vary according to the final objective, from
one minute to one hour or more.
FMRI can provide information regarding interactional pattern between brain
regions, the functional connectivity methodology [129]. Based on the temporal
synchronism of the BOLD fluctuation among brain areas, the functional connec-
tivity can provide important information about the brain network’s organization
and its alterations on epilepsy subtypes, for example [129]. Recent studies demon-
strated throughout functional connectivity that MTLE patients with HS in different
sides present a complete distinct pattern of functional alterations, reinforcing the
Figure 2.3. Graphical representation of the haemodynamic response function of a stimulus. The MRI signs
fluctuates are correlated to the local relative concentration of oxy- and deoxyhemoglobins. The HRF cycle
after the trigger event is around 30 s.
2-17
method as an effective way to define and characterize the epileptic syndromes even
with subtle differences [130, 131].
Another technique widely applied over the past 20 years is the EEG acquired
concomitantly to the fMRI (EEG–fMRI technique) (figure 2.4) [132]. The EEG–
fMRI acquisition, with scalp-EEG is a noninvasive procedure, and uses no ionizing
radiation or intravenous contrast. The procedure requires MR compatible EEG
systems and specialized EEG and fMRI data processing steps. The method is a
powerful tool that combines the high temporal resolution of the EEG (in the scale of
milliseconds) with the high spatial resolution of the fMRI (in the scale of mm3) [132,
133]. The modality enables one to identify ‘WHEN’ the ictal or interictal discharges
occurred using the EEG to sequentially, search for temporally synchronized BOLD
alterations on the images (WHERE), being useful for SOZ localization and
syndrome characterization [132–134].
2.7.2 Application to epilepsy

Coan et al found that patients who underwent epilepsy surgery and had IED source
localization concordant to the surgical resection showed a good surgical outcome
(ILAE) in 81% of the cases, while those with discordant or absent EEG–fMRI IED
related focus presented a good outcome in only 21% [135]. Pittau et al, in a recent
study with epilepsy associated to FCD, showed that 69% of the studied group
Figure 2.4. Interictal epileptiform discharge (IED) source definition using EEG–fMRI. In this case, the result
from a left-temporal lobe epilepsy patient with MRI signs of left hippocampal sclerosis. The fMRI analysis
was based on EEG-informed left temporal sharp-waves events.
2-18
presented IED BOLD response localized on the FCD area [136]. Besides the SOZ
localization, EEG–fMRI is able to define adjacent irritative areas or a whole
network of interictal epileptiform discharges (IEDs) associated regions.
The method is still undergoing improvements and investigation but its results
have been used in several centers as clinical complementary information or a
confirmatory resource on epilepsy investigation and pre-surgical planning [135].
2.8 Subdural grids (SBG)

2.8.1 Introduction
Despite previously discussed non-invasive techniques to delineate epileptogenic
zones, some patients cannot yet benefit. It is estimated that approximately 25%–
50% of refractory epilepsy patients will have to undergo invasive EEG monitoring.
Objectives for invasive EEG monitoring with intracranial electrodes are lateral-
ization, localization of epileptogenic zones and mapping of eloquent areas [137]. It is
helpful in cases of ambiguous or discordant previous results, non-lesional or
extratemporal lobe epilepsy and functional mapping of relations to eloquent areas
(e.g. patients that cannot tolerate awake surgery in tumors) [138]. Advantages
include higher accuracy and fewer artifacts [137].
Subdural grids were popularized by W Penfield and H Jasper and materials
evolved over recent decades into flexible grids that permit long term (days, weeks),
extra-operative records.
The main indications for subdural exploration are as follows:
• documented drug-resistant epilepsy;
• significant impact of the seizures on patient’s quality life;
• expectation of seizure reduction after surgery;
• willingness of the patient to undergo multiple intracranial surgeries;
• superficial location of supposed epileptogenic lesion;
• eloquent areas in close relation with the hypothetical EZ [139, 140].
Grid sizes and shapes are chosen based on a patient’s previous investigation
(non-invasive), pointing towards suspected/hypothesized epileptic region. Electrodes
implantation is usually via general anesthesia craniotomy, with the use of intra-
operative neuronavigation and frameless stereotaxy for deep cortical sample targets.
Attention is recommended for head positioning, hair shaving, prophylactic anti-
biotics and use of steroids in post-operative. Measurements are performed, a
C-shaped dural incision is made and electrodes are positioned with attention to
superficial veins (due to risk of injury, compression and congestion). Regions of
adherence may relate to underlying bridging veins. It is recommended to implant
deep cortical samples with frameless stereotaxic and pial incision, before placement
of cortical targets. Grids or strips are measured and placed under irrigation to
prevent cortical laceration. If the clinician finds it difficult to localize a previously
epileptogenic lobe, one can have strip electrodes placed via burr holes on suspected
lobes and on a second operation perform a more guided craniotomy study, with the
recommendation of six weeks between procedures for decrease in infection risk.
2-19
Timing and monitoring of clinic semiology during records are also important
[137, 138].
Wires are then sutured to dural margins in the effort to avoid electrodes
displacement and digital photographs are taken to confront with posterior acquired
data. The neurosurgeon proceeds to water-tight dural suture, passage of wires
through burr holes, closing with a free bone margin, proper identification of colors,
numbers and codings coming out through tunneling within the scalp flap, sutured
with purse-string stitches, subgaleal drain and careful dressings and head wrap [137].
Some services replace bone flap on the second surgery with the removal of electrodes
only. Subgaleal tunnel and purse-string stitches aim to prevent CSF leaks. Secure
head wraps may have a submandibular sling to minimize chance of displacement
with seizures [138].
Post-operative care includes intensive care unit (ICU) stay, radiographic imaging
of the grid placement (x-rays, CT and MRI with three-dimensional reconstruction).
Wires are then attached to EEG equipment and patients may stay in epilepsy
monitoring units. Anti-epileptic drugs are adjusted according to the epilepsy clinical
team. Steroids, antibiotics and sterile head dressing changes are performed over the
following days. After proper data acquisition, which may last up to a few weeks,
surgical grid removal is performed. EZ resection may occur on the same occasion
depending on epilepsy conference decision. Grids are delicately removed under
irrigation, wires are disconnected and can be pulled out of the field. Closing then
proceeds in standard fashion, with attention to closing all previous exit sites in order
to prevent possible CSF leaks and infection.
Specific considerations for the pediatrics population include trained staff, addi-
tional time to explain to family members about each step of the process, sometimes
the need for general anesthesia to obtain 3T MRI, age appropriate surgical
equipment and testings, besides additional care for the risk of children removing
wires and electrodes [137].
Results of surgeries planned from subdural grids explorations are the reward, but
the cost of such procedure may vary between $77 000 to $130 000 [140]. This may
also partially justify the limited sample sizes in the majority of studies.

A study with 16 patients with TLE, including 11 subjects with normal MRI (group I)
and five patients with bilateral MTS (group II), evaluated the outcomes after
subdural grids evaluation. All of them exhibited bilateral independent temporal lobe
spiking, were implanted bilaterally with 32 contacts subdural grids and underwent
resection in the side indicated by chronic electrocorticography (ECoG). Eight
patients from group I became seizure-free while three patients achieved Engel II
outcome. In group II, four patients (80%) were seizure-free and only one showed
Engel class II after surgery [141]. Using ECoG recordings, Abel and colleagues
assessed ten consecutive patients with TLE who underwent cortico amygdalohip-
pocampectomy. In seven patients (70%) the ictal onset included the temporal pole
(TP). From those seven patients, three did not reach Engel I after selective surgery
2-20
and only two showed evident abnormalities of temporal polar cortex in MRI. In two
patients with evidence of MTS and no involvement of TP in ictal onset as
determined by ECoG, authors obtained Engel I after surgery. This data demon-
strated the usefulness of SBG and ECoG to evaluate the participation of TP in TLE
and showed that SOZ in TP does not necessarily correlate with preoperative
neuroimaging abnormalities [142].
The feasibility, safety and utility of subdural invasive monitoring was also
demonstrated for interhemispheric EZ. Bekellis et al applied interhemispheric grid
electrodes monitoring in 50 patients, who previously underwent non-invasive studies
(MRI, scalp-EEG, ictal SPECT and interictal PET). Only one patient showed
transient leg weakness, with no other major complications. A total of 21 (55.3%)
patients was submitted to medial frontal resective surgery, nine of whom (43%)
showed normal MRI results previous to surgery. In all cases, localization of
supposed EZ was possible only because of data from the interhemispheric grids,
and the extent of resection was tailored based on such data as well. Interhemispheric
grids were helpful to exclude medial frontal epilepsies in 17 patients (who underwent
other types of resection) and precluded resection in two patients due to motor cortex
involvement with EZ. Twelve patients did not undergo resection because of
nonlocalizable or multifocal disease. After 2 years of follow-up, 27 patients (71%)
who underwent surgical resection achieved Engel class I [143].
In a recent review paper, Bingamam et al discussed the value of subdural
electrodes to monitor the EZ in pediatric patients. The ability to localize the EZ
may approach high rates but seizure-free depends on surgical experience and the
extension of resection [144]. Vale et al also observed that patients with positive
lesional preoperative MRI presented improved outcomes regarding seizure control
compared to patients in whom MRI was normal before resective surgery [145].
2.8.3 Complications
Overall risk of complications is 9%–22%, with 5%–6% for depth cortical electrodes
[146, 147]. Possible complications are CSF leaks, infection, hemorrhage, among
others.
CSF leakage incidence varies from 0.5%–2% to 30% according to studies. It may
be prevented by attention to electrodes exit sites [146, 147]. Treatment consists in
reinforcement of suture, local wrapping with betadine-soaked gauze, and in some
cases, temporary lumbar drain [137]. Another associated complication is infection,
in 4%–5% of the cases [147]. Infection may consist in superficial wound infection,
meningitis, epi- or subdural abscess, osteomyelitis and intraparenchymal brain
abscess. Risk factors include longer length of implantation and increased number
of electrodes [148]. Treatment varies from isolated antibiotics to reoperation with
removal of grids, culture sampling, copious irrigation, debridation and IV anti-
biotics. In chronic infection, removal of bone flap may be indicated.
Clinically relevant cerebral edema is reported at 2%–3% risk [147]. Patients might
develop headaches, neurologic deficits, loss of consciousness and fatal herniation.
Local irritation evolves to subsequent inflammatory edema. Pediatric population is
2-21
in greater risk for this complication. Measures to minimize such a risk include wide
opening of the dura, loose bone closure, steroids administration perioperative.
Treatment will consist in removal of electrodes and close patient monitoring [137].
Radiographic relevant hemorrhage rate is present in 16% of the cases, and
clinically relevant hemorrhage in 7%–8% [147]. Risk factors are intraoperative
cortical injury and venous occlusion or disruption [149]. It may be associated with
brain edema and subdural hematoma, and cause prejudice to recordings and clinical
decline.
Other relevant complications might be thromboembolic events, pneumonia and
other medical conditions from hospitalization and immobilization. Displacement,
removal and fracture of implanted electrodes are of greater risk in patients with a
motor semiology of seizures and post-ictal confusion. Activity beyond the edge of
the grid and necessity of multiple procedures are also associated with a greater risk
of complications [137, 147].
2.8.4 Conclusions
Subdural grids are a form of invasive EEG monitoring and may be useful in
delineating the epileptogenic zone, mainly in cases of superficially located SOZ in
close anatomical relation to eloquent cortex. It should be performed by a trained
team, and may also be useful in the pediatrics population.
2.9 Stereoelectroencephalography (SEEG)

2.9.1 Introduction
After using the aforementioned non-invasive modalities in an attempt to delineate
the suspected EZ, a multidisciplinary team must discuss the results in conjunction.
A group composed of neurologist, neurosurgeon, neuroradiologist and neuro-
psychologist interprets the results from seizure semiology, EEG, PET-CT,
SPECT, and MEG in order to formulate an anatomo-clinical-electro (AEC)
hypothesis for the epileptic network. According to the content already discussed
in this chapter, a precise localization of where the epileptogenic zone occurs is quite
difficult. The objective is to integrate knowledge from different experts and, thus, to
formulate the widest and most accurate hypothesis to identify both the current EZ
and the areas that could eventually assume the role of EZ [139]. By electing the areas
supposed to be involved in the EZ, the multidisciplinary team chooses the areas that
are going to be monitored by electrodes implantation. The strategy of implantation
is individually addressed and they try to understand which regions of the brain are
involved in the primary organization of seizure to guide a tailored resective surgery.
General indications for invasive monitoring include:
• MRI negative cases: absence of a structural lesion hampers confirmation or
even causes doubt about the electrical hypothesis generated by scalp-EEG
and VEEG [150].
• Electroclinical and MRI data discordance: in some cases, results shown by
MRI do not agree with those shown by EEG, PET and SPECT. This may
occur in cases of deeply seated lesions as periventricular nodular heterotopia
2-22
and deep sulcal lesions [151, 152]. Additionally, patients with FCD show
interictal spikes that range in their distribution from lobar to lateralized, and
are difficult to localize to diffuse in the majority of cases [153–155]. Moreover,
interictal spikes may occupy a spatial distribution greater than that shown by
MRI preoperative assessment or intraoperative visual inspection [156].
Another possibility is the presence of two anatomical lesions such that the
location of one or both is discordant to the location suggested by the
electroclinical hypothesis, or both lesions are located in the same functional
network with no clear evidence whether one or more is epileptogenic.
• Overlap with eloquent cortex: the AEC hypothesis elaborated by the multi-
disciplinary team may include at least part of the eloquent cortex or be in close
relation to it. Defining the precise boundaries of the EZ in this situation is of
paramount importance to preclude or refer to resective surgery. A common
condition that may overlap with the eloquent cortex is FCD [156–163].
2.9.2 SEEG versus SBG

Both methods of invasive monitoring have shown advantages compared to solely
non-invasive management. The decision between these two modalities of invasive
evaluation prior to resection may vary according to the experience of the surgical
team. However, there is a recent tendency in favor of SEEG rather subdural grids
and strips. Some advantages shown by SEEG include fewer postoperative compli-
cations, besides shorter recovery time based on length of stay [164].
It is of note that the advantages of SEEG in relation to subdural grids and depth
electrodes (SGDE) go beyond the better profile of surgical complications. In fact,
SEEG is more suitable for remote and multilobular areas, without the need for
craniotomies and immediate resective surgery. This approach leads to more time for
the multidisciplinary team to discuss resective strategy and to the patient’s reflection
about the surgical treatment, allowing a more informed consent.
Maybe the more pronounced advantage of SGDE explorations is the ability of
this method to cover and sample hypothetical EZ and contiguous functional cortex
simultaneously. With such capacity, SGDE is suitable for preservation of eloquent
cortex and helps to preclude resective surgeries that may cause permanent deficits
[165, 166]. However, this advantage of SGDE is relative. First, precision of subdural
functional mapping is far from being validated. Second, SEEG functional mapping
cannot be dissociated from electroclinical localization, besides the fact that SEEG
may be complemented by DTI technique and/or awake craniotomies, diminishing
the disadvantages pointed out by pro-subdural groups [158].
SBDE implantations are an open procedure and, within a surgical view, permit a
better management of eventual hemorrhage. On the other hand, bleeding is a much less
incidental occurrence in SEEG procedures compared with SBDE implantations [167].
The main disadvantage of subdural monitoring is the difficulty of sampling deep
structures such as insula, orbitofrontal cortex, cingulate gyrus and depth of sulci, for
example. Added to the intrinsic inability to record subcortical and white matter
regions, this limitation exposes also the inability of subdural electrodes to define the
2-23
spatiotemporal dynamic of the epileptic network. On contrary, SEEG allows

coverage of extensive and deep seated region with minimal associated morbidity
[158, 168–173]. By recording cortical, subcortical and deep structures, SEEG has the
potential of defining how the epileptic electric signals travel over time, with a higher
spatiotemporal definition of the network involved in the seizure primary
organization.
Overall, SEEG seems to be a better choice for invasive monitoring and the
following reasons corroborate this assertion:
1. SEEG permits to record deep-seated or difficult to cover areas such as mesial
temporal lobe structures, perisylvian areas, cingulate gyrus, mesial interhe-
mispheric locations, ventromedial prefrontal regions, insula and the depth of
sulci.
2. Resective surgery guided by SEEG monitoring for patients in whom subdural
monitoring failed showed 60% of seizure-free outcome after 11 months
follow-up [169].
3. The need for bihemispheric (normally extensive) explorations. This may be
the case of focal epilepsies arising from interhemispheric or deep insular
regions as well as the temporo-parieto-occipital junction [174].
4. The AEC hypothesis suggests the need for extensive explorations (tempor-
oparietal or frontoparietal ones), mainly associated with normal MRI.
As one may conclude, indications for invasive explorations may vary according to
the AEC hypothesis generated by non-invasive explorations of the hypothesized EZ,
and thus represent a decision made on each particular situation, as synthesized in the
decision-making flowchart in figure 2.5.
2.9.3 Patterns of explorations

Once SEEG has been decided as a more suitable invasive exploration modality,
there are some patterns of explorations that may be considered in order to avoid
cover potential EZ not completely evidenced by non-invasive studies [139, 175]:
• Limbic networks explorations: in the majority of cases, use of invasive
explorations is not necessary when semiological and electricalphysiological
hypothesis suggest typical non-dominant MTLE and imaging studies are
concordant to show a clear lesion (e.g. unilateral hippocampal sclerosis). On
the other hand, temporal plus epilepsies show an increased risk of surgical
failure (5.06 times greater compared to unilateral TLE) [176]. This explains
why SEEG may be required for patients in whom the supposed EZs (probably
involving the temporal lobes) are suspected of involving extratemporal
areas as well. In these cases, preferential directions of discharge involve
temporo-insular-anterior perisylvian areas, the temporo-insular-orbitofrontal
areas, or the posterior temporal, posterior insula, temporo-basal, and parietal
and posterior cingulate areas. Sampling of extratemporal limbic areas must be
wide enough to identify an eventual extratemporal origin of seizures that was
not clearly suspected by non-invasive studies solely.
2-24
Figure 2.5. Algorithm for delineation of epileptogenic zone (EZ). EEG = electroencephalography,
MRI = magnetic resonance imaging, PET = positron emission tomography, SPECT = single photon emission
computed tomography, MEG = magnetoelectroencephalography, fMRI = functional MRI, SBG = subdural
grids, SEEG = stereoelectroencephalography.
2-25
• Frontal-parietal networks explorations: both frontal and parietal lobes are

relatively large regions, possibly demanding a high number of electrode
contacts. Nonetheless, specific cases may allow a more restricted coverage.
Cases of orbitofrontal epilepsies often require the investigation of the gyrus
rectus, the frontal polar areas, the anterior cingulate gyrus, and the temporal
pole. Seizures suspected of arising from mesial premotor cortex are evaluated
by sampling at least the rostral and caudal parts of the supplementary motor
area (SMA), the pre-SMA area, different portions of the cingulate gyrus and
sulcus, the primary motor cortex, and the mesial and dorsolateral parietal
cortex. By doing so, epileptogenic regions in the frontal and parietal lobes
may be sampled, even when they are relatively small. Cases in which fronto-
parietal networks are suspected to be involved in the EZ and non-invasive
studies failed in lateralizing the EZ may require bilateral and symmetrical
explorations. Explorations of rolandic and neighboring regions are often
required when there is the need to define the posterior resection margin in
frontal networks explorations, the anterior resection margin in parieto-
occipital explorations, and when the EZ may be within or very close to
rolandic regions. By intracerebral electrical stimulation, this pattern of
explorations (sampling the central sulcus as well as ascending and descending
fibers related to this region) permits one to assess the participation of rolandic
regions in the ictal discharge and to obtain a functional mapping.
• Posterior quadrant network explorations: in cases of posterior quadrant
epilepsies, simultaneous participation of occipital, parietal and posterior
temporal structures is a common event, as well as multidirectional spread
to both supra and infra sylvian areas or even to the contralateral hemisphere
during ictal episodes. In this instance, posterior quadrant network explora-
tions may sample mesial and dorsolateral surfaces of occipital lobe, infra and
supra calcarine regions, posterior temporal, posterior peri-sylvian, basal
temporo-occipital, and posterior parietal regions, including the posteroinfe-
rior parietal lobule and the posterior precuneus.
2.9.4 Technique description

Before electrode implantation, a stereo-contrasted volumetric T1-weighted MRI is
obtained and fused with stereotactic DynaCT (Siemens AG) scans and angiographic
images [172]. The resultant image is transferred to a robot’s native planning
software. This is necessary to evaluate the vascular disposition of the patient and
to program an implantation strategy that both respects the idiosyncratic vascular
variations and allows one to test the AEC hypothesis generated by non-invasive
studies with the maximal possible sampling rate. Electrodes are usually inserted
orthogonally to facilitate the correlation between anatomo-electrophysiologic data
and to avoid deviations on electrode trajectory that could be seen with high
angulated implantations. If the orthogonal implantation is at risk of causing a
vascular injury, the direction of electrode implantation is redefined. It is mandatory
to amplify coronal, axial and sagittal views in order to perform a profound anatomic
2-26
study and to avoid intersections between vessels and electrodes trajectory. Vessels
allocated in the brain surface are under higher risk of damage because they are more
difficult to identify and less mobile compared to the cisternal ones. It is also
recommended that implantation trajectories avoid passing through bone cells to
prevent liquor fistulae. Another recommendation concerns the number of electrode
implantations: a single non-orthogonal implantation for multiple target points is
preferable to multiple orthogonal implantation for covering the same target points.
This aims to minimize the risk of complications verified by multiple implantations.
Finally, implantation strategy is re-checked using a 3D cranial reconstruction
capability [139].
The implantation starts by positioning the patient under general anesthesia with
the cranium 70 cm distant from the base of the robotic arm, ideally. The patient’s
table and the robot arm are locked, while the head holder device is secured to the
robot. Then, robot laser is calibrated and the preoperative volumetric MRI is
coregistered with the patient using a semi-automatic laser based facial recognition
tool, in which preset facial landmarks are manually selected with the laser until there
is a satisfactory match between the manually entered landmarks and the automatic
registered ones [139].
Once the patient is already prepped and draped in a sterile standard fashion way,
the robotic arm locks in an initial setwork position (150-mm distant from the target
point), and the drilling platform (i.e. a 2.5 mm diameter cannula previously
attached) initiates a movement towards the cranium and the target point. The
drilling platform reaches a previously calculated position and, then, the neuro-
surgeon introduces a 2 mm handheld drilling (Stryker, MFI Medical Equipment,
San Diego, CA) through the platform to create a pinhole. After skin perforation, a
monopolar device is introduced through the same platform to coagulate the vessels.
The next step is to reintroduce the drilling to perforate the skull bone. The length of
the drilling necessary to perforate the inner limit of the bone (when there is an abrupt
fall in the resistant against drilling dislocation) is measured. At this point, the
neurosurgeon introduces an insulated dural perforator using a monopolar cautery at
low settings. The introduced length of the dural perforator must exceed in one finger
(5–8 mm, approximately) the previously measured length of the drill. A guiding bolt
(Ad-Tech, Racine, WI) is firmly screwed into the pinhole at the end of the process
[172].
After the distance from the drilling platform to the retained bolt is automatically
measured, this value is subtracted from the standardized 150 mm platform to the
target distance. The resulting difference is recorded and corresponds to the final
length of the electrode that must be implanted. This process is repeated for each
trajectory.
Once all bolts are screwed and the electrode length of each insertion is calculated,
the electrode insertion itself takes place. The neurosurgeon changes the gloves and
then inserts a small stylet into the parenchyma (2 mm of diameter). The stylet is
inserted until a depth corresponding to the electrode length of insertion, in a
rotational and gentle manner guided by the implanted bolt. The stylet is also gently
pulled back, and a premeasured electrode is inserted through the implantation bolt.
2-27
The guiding line attached to the electrode is finally pulled back. In the final phase of
surgery, a fluoroscopic study is performed intraoperatively to evaluate the success of
electrode implantations. Curves or tortuosity in the electrode trajectory denote it has
not reached the target point and a reinsertion attempt should be tried. The last step is
to test whether all electrodes are recording the electrical activity of the brain
properly.
2.9.5 Surgical outcomes

Seizure control rates after invasive monitoring with SEEG technique have shown
rewarding results among different centers worldwide. Moreover, the complications
observed in the postoperative period translate the safety of this technique [177].
González-Martínez et al reported the experience of 200 patients who underwent
SEEG investigations using either the conventional Leksell frame-based stereotaxy
(n = 122) or frameless stereotaxy under robotic guidance (n = 78). Collectively, they
performed a total of 2663 tailored electrode implantations after non-invasive studies
with video-EEG, MRI, PET, SPECT, and neuropsychological assessment. The
epileptogenic zone was confirmed by SEEG in 154 patients (77%), of which 134
(87%) underwent resection of the supposed EZ. From those who underwent
resection, a considerable number of 61 patients (67.8%) was seizure-free after an
average follow-up of 2.4 years. The most prevalent etiology was FCD type I
(n = 55). Complication rates per electrode were minimal: 0.08% of wound infection,
0.08% of hemorrhage, and 0.04% transient neurological deficit [178].
Goldstein et al evaluated the contribution of SEEG for pediatric patients. They
studied 25 patients who underwent SEEG evaluation (average of 8 days of
monitoring), of whom 10 patients had previous resection or ablations. SEEG
localized the EZ in 23 (92%) out of 25 pediatric patients. A total of 15 patients
showed a minimum 6 months follow-up, of whom 8 patients (53%) were seizure-free.
There were no major complications and minor complications were found in only
three patients [179]. Analogously, Taussig et al evaluated the results of SEEG in
65 pediatric patients. In this study, MRI was not contributory in 20% of the cases,
and 78% of patients with a low probability of having surgery before SEEG received
surgery. Engel class I outcome was achieved in 67% of patients who underwent
surgical resection after SEEG investigation. No complications were reported in the
series [180].
When applied to patients with polymicrogyria (PMG), SEEG also provides an
important contribution. In 49 patients with PMG who underwent SEEG evaluation,
the invasive monitoring showed that in only 8 patients (16%) the PMG lesion was
fully concordant with the supposed EZ, whereas in 29 patients (60%) the supposed
EZ involved only part of the PMG, in seven patients (14%) it involved the entire
PMG and remote cortical areas, and in five patients (10%) it involved only remote
cortical areas. In those patients who experienced surgical resection after SEEG
evaluation, the author obtained Engel I outcome in 85% (n = 10) of the cases in
which the supposed EZ was fully concordant with the PMG and in 100% (n = 3) of
the cases in which the EZ involved only remote cortical areas [181].
2-28
When bilateral exploration is required, SEEG may be an alternative to invasive

exploration. Steariade et al analyzed 184 patients who underwent bilateral SEEG
implantations. A number of 106 of 184 patients were referred to resective surgery. A
total of 81 patients underwent resection and had more than 1 year follow-up. From
these, 26 (32%) patients remained seizure-free and 44 (54%) patients experienced
>50% of seizure reduction. Predictors of seizure-freedom were: single seizure type,
short epilepsy duration, use of ⩽2 antiepileptic drugs at the time of surgery, AEC
hypothesis involving the frontal lobe, sentinel electrode placement only, and lack of
overlap between IOZ and the eloquent cortex [182].
Concerning surgical complications, a meta-analysis elucidated that SEEG has a
safe profile of complications. In a total of 2624 SEEG procedures (22 085 implanted
electrodes), there were 121 cases of complications, encompassing 40 patients with
hemorrhages, 11 patients with permanent neurologic deficits, 28 patients with
infections, 11 cases of implant malfunction, and five deaths. This study supports
that SEEG presents a lower rate of surgical complications in comparison with other
invasive monitoring modalities [183].
2.9.6 Conclusions
In cases of invasive monitoring, SEEG proved to be a safe and effective procedure
simultaneously. Its advantages encompass the possibility of recording both hemi-
spheres, extensive networks, depth of sulci, white matter pathways and deep
structures. With such aspects, SEEG permits the observation of the epileptic
network dynamics with an optimal spatiotemporal correlation. Finally, this invasive
exploration method is useful in guiding tailored surgical resections with a rewarding
seizure control.
2.10 Illustrative case

Herein, we will discuss a case in which the algorithm here proposed was applied and
helped successfully in resection surgical planning. An 18-year-old right-handed
woman presented with seizures characterized by déjà vu and burning smell
sensations followed by confusion and semi-purposeful movements. Seizure semi-
ology sometimes included eye blinking initially then subtle oral automatisms
followed by confusion. She recalls speech disturbances postictally as well. Scalp-
EEG showed sharp waves arising from left mesial temporal regions (maximum on
the left sphenoidal electrode) and intermittent slow waves in the left fronto-temporal
electrodes. Structural FLAIR-weighted images showed subtle increased signal in
mesial temporal structures (more evident on the left side), and vertical hippocampi.
Interictal PET showed mild to moderate hypometabolism in the anterior and mesial
temporal regions bilaterally, more prominent on the left side and minimally reduced
FDG uptake in the left orbitofrontal regions. Ictal SPECT revealed hyperperfusion
in the left temporal region, left temporoparietal region, and right dorsolateral frontal
central region (figure 2.6).
First, AEC hypothesis suggested lateralization of the supposed EZ in the left side,
however, non-invasive studies were not strongly concordant regarding EZ
2-29
Figure 2.6. Imaging findings of the illustrative case. (a) Coronal FLAIR-weighted image with increased signal
in mesial temporal regions and vertical hippocampi. (b) Interictal PET (coronal cut) demonstrating
hypometabolism in the left temporal pole. (c) Ictal SPECT study (axial section) demonstrating hyperperfusion
in the left temporal lobe. (d) Blue dots demonstrate the entry point of each SEEG electrode in a schematic 3D
reconstruction of the brain. Reproduced from [184]. With permission of Springer.
localization. Second, non-invasive evaluation suggested involvement of deep struc-

tures and more than one lobe. For these motives, invasive monitoring with SEEG
seemed plausible and indicated.
Based on the non-invasive studies evaluation, electrodes implantation strategy
aimed to assess the following questions: (1) the involvement of mesial temporal
structures and its association (or not) with temporal pole, neocortical areas,
entorhinal cortex, superior temporal gyrus and dorsolateral temporal neocortex,
(2) the involvement of the orbitofrontal cortex and its involvement with mesial and
neocortical structures, (3) participation of the anterior insula in the epileptic
network, and (4) if dorsolateral temporal neocortex is involved, there is a need to
define the caudolateral extension of the EZ. A frontotemporal implantation strategy
was planned and detailed in the original publication of de Souza et al [184].
The SEEG evaluation demonstrated frequent sharp waves in the left temporal
pole, amygdala and hippocampus. One habitual seizure was recorded. SEEG onset
consisted in rhythmical 4–5 Hz activity in the temporal pole followed by early spread
to amygdala and subsequently to lateral temporal neocortex (in 20 s). Initially,
mesial structures were not involved in the discharge. Such findings guided a surgical
resection towards temporal pole and amygdala, sparing the hippocampus.
Intraoperatively, multiple small temporal pole encephaloceles were found. En bloc
resection of temporal pole and amygdala was performed, and the patient remained
seizure-free with no neuropsychological deficits (12 months follow-up).
2-30
This case exemplifies how to integrate non-invasive studies to formulate an AEC

hypothesis, how to choose an invasive monitoring modality, and how to plan a
surgical resection strategy based on the supposed EZ informed by extraoperative
exploration.
2.11 Conclusion
Delineation of the epileptogenic zone remains a challenge for proper surgical
treatment. Semiology and non-invasive studies contribute decisively to the formu-
lation of an AEC hypothesis. This will guide not only the modality of invasive
exploration, but also the strategy of electrode implantations. Based on the invasive
exploration, a patient tends to benefit from a tailored surgical resection strategy,
optimizing the chances of a seizure-freedom outcome.
References
[1] Walczak T 1994 Surgical Treatment of the Epilepsies 2nd edn ed J Engel Jr (New York:
Raven Press) 1993 Ann. Neurol. 35 252
[2] Berg A T, Testa F M and Levy S R 2011 Complete remission in nonsyndromic childhood-
onset epilepsy Ann. Neurol. 70 566–73
[3] Fisher R S et al 2014 A practical clinical definition of epilepsy Epilepsia 55 475–82
[4] Jehi L 2018 The epileptogenic zone: concept and definition Epilepsy Curr. 18 12–6
[5] Holmes M D and Tucker D M 2013 Identifying the epileptic network Front. Neurol. 4 84
[6] Lüders H O, Najm I, Nair D, Widdess-Walsh P and Bingman W 2006 Definition and
localization of the epileptogenic zone The epileptogenic zone: general principles Epileptic
Disord. Suppl. 2 S1–9
[7] Magiorkinis E, Diamantis A, Sidiropoulou K and Panteliadis C 2014 Highights in the
History of Epilepsy: The Last 200 Years Epilepsy Res. Treat. 2014 1–13
[8] Tufenkjian K and Lüders H O 2012 Seizure semiology: its value and limitations in
localizing the epileptogenic zone J. Clin. Neurol. 8 243–50
[9] Coan A C and Cendes F 2013 Understanding the spectrum of temporal lobe epilepsy:
contributions for the development of individualized therapies Expert Rev. Neurother. 13
1383–94
[10] Jan M M S and Girvin J P 2008 Seizure semiology: value in identifying seizure origin Can.
J. Neurol. Sci. 35 22–30
[11] Fisher R S et al 2017 Operational classification of seizure types by the International League
Against Epilepsy: Position paper of the ILAE Commission for Classification and
Terminology Epilepsia 58 522–30
[12] Kennedy J D and Schuele S U 2012 Neocortical Temporal Lobe Epilepsy J. Clin.
Neurophysiol. 29 366–70
[13] Alvim M K M, Morita M E, Yasuda C L, Damasceno B P, Lopes T M, Coan A C, Ghizoni
E, Tedeschi H and Cendes F 2018 Is inpatient ictal video-electroencephalographic
monitoring mandatory in mesial temporal lobe epilepsy with unilateral hippocampal
sclerosis? A prospective study Epilepsia 59 410–19
[14] Tinuper P and Bisulli F 2017 From nocturnal frontal lobe epilepsy to Sleep-Related
Hypermotor Epilepsy: A 35-year diagnostic challenge Seizure 44 87–92
2-31
[15] Marashly A, Ewida A, Agarwal R, Younes K and Lüders H O 2016 Ictal motor sequences:
Lateralization and localization values Epilepsia 57 369–75
[16] Williamson P D, Thadani V M, Darcey T M, Spencer D D, Spencer S S and Mattson R H
1992 Occipital lobe epilepsy: Clinical characteristics, seizure spread patterns, and results of
surgery Ann. Neurol. 31 3–13
[17] Salanova V 2012 Parietal Lobe Epilepsy J. Clin. Neurophysiol. 29 392–6
[18] Obaid S, Zerouali Y and Nguyen D K 2017 Insular Epilepsy J. Clin. Neurophysiol. 34
315–23
[19] Leutmezer F, Serles W, Lehrner J, Pataraia E, Zeiler K and Baumgartner C 1998 Postictal
nose wiping: a lateralizing sign in temporal lobe complex partial seizures Neurology 51
1175–7
[20] Trinka E, Walser G, Unterberger I, Luef G, Benke T, Bartha L, Ortler M and Bauer G
2003 Peri-ictal water drinking lateralizes seizure onset to the nondominant temporal lobe
Neurology 60 873–6
[21] St Louis E K, Frey L C, Britton J W, Frey L C, Hopp J L, Korb P, Koubeissi M Z, Lievens
W E, Pestana-Knight E M and St Louis E K 2016 Electroencephalography (EEG): An
Introductory Text and Atlas of Normal and Abnormal Findings in Adults, Children, and
Infants (Chicago, IL: American Epilepsy Society)
[22] Binnie C D and Stefan H 1999 Modern electroencephalography: its role in epilepsy
management Clin. Neurophysiol. 110 1671–97
[23] Celesia G G and Chen R C 1976 Parameters of spikes in human epilepsy Dis. Nerv. Syst. 37
277–81
[24] Gotman J 2010 High frequency oscillations: the new EEG frontier? Epilepsia 51 63–5
[25] Ebersole J S and Pedley T A 2003 Current practice of clinical electroencephalography 3rd
edn Eur. J. Neurol. 10 604–5
[26] Lüders H O, Najm I, Nair D, Widdess-Walsh P and Bingman W 2006 The epileptogenic
zone: general principles Epileptic Disord. 8 S1–9
[27] Cascino G D, Hulihan J F, Sharbrough F W and Kelly P J Parietal lobe lesional epilepsy:
electroclinical correlation and operative outcome Epilepsia 34 522–7
[28] Blume W T, Young G B and Lemieux J F 1984 EEG morphology of partial epileptic
seizures Electroencephalogr. Clin. Neurophysiol. 57 295–302
[29] Gibbs F A, Gibbs E L and Lennox W G 2002 Epilepsy: a paroxysmal cerebral dysrhythmia
Epilepsy Behav. 3 395–401
[30] Geiger L R and Harner R N 1978 EEG patterns at the time of focal seizure onset Arch.
Neurol. 35 276–86
[31] Risinger M W, Engel J, Van Ness P C, Henry T R and Crandall P H 1989 Ictal localization
of temporal lobe seizures with scalp/sphenoidal recordings Neurology 39 1288–93
[32] Sharbrough F W 1993 Scalp-recorded ictal patterns in focal epilepsy J. Clin. Neurophysiol.
10 262–7
[33] Walczak T S, Radtke R A and Lewis D V 1992 Accuracy and interobserver reliability of
scalp ictal EEG Neurology 42 2279–85
[34] Foldvary N, Klem G, Hammel J, Bingaman W, Najm I and Lüders H 2001 The localizing
value of ictal EEG in focal epilepsy Neurology 57 2022–8
[35] Alarcon G, Guy C N, Binnie C D, Walker S R, Elwes R D and Polkey C E 1994
Intracerebral propagation of interictal activity in partial epilepsy: implications for source
localisation J. Neurol. Neurosurg. Psychiatry 57 435–49
2-32
[36] Cooper R, Winter A L, Crow H J and Walter W G 1965 Comparison of subcortical, cortical
and scalp activity using chronically indwelling electrodes in man Electroencephalogr. Clin.
[37] Cascino G D, Trenerry M R, So E L, Sharbrough F W, Shin C, Lagerlund T D, Zupanc
M L and Jack C R 1996 Routine EEG and temporal lobe epilepsy: relation to long-term
EEG monitoring, quantitative MRI, and operative outcome Epilepsia 37 651–6
[38] Chung M Y, Walczak T S, Lewis D V, Dawson D V and Radtke R Temporal lobectomy
and independent bitemporal interictal activity: what degree of lateralization is sufficient?
[39] Ebner A and Hoppe M 1995 Noninvasive electroencephalography and mesial temporal
sclerosis J. Clin. Neurophysiol. 12 23–31
[40] Williamson P D, French J A, Thadani V M, Kim J H, Novelly R A, Spencer S S, Spencer
D D and Mattson R H 1993 Characteristics of medial temporal lobe epilepsy: II. Interictal
and ictal scalp electroencephalography, neuropsychological testing, neuroimaging, surgical
results, and pathology Ann. Neurol. 34 781–7
[41] Rosenow F and Lüders H 2004 Presurgical assessment of the epilepsies with clinical
neurophysiology and functional imaging (Amsterdam: Elsevier)
[42] Vossler D G, Kraemer D L A, Knowlton R C, Kjos B O, Rostad S W, Wyler A R, Haltiner
A M, Hasegawa H and Wilkus R J 1998 Temporal ictal electroencephalographic frequency
correlates with hippocampal atrophy and sclerosis Ann. Neurol. 43 756–62
[43] Iwasaki M, Jin K, Nakasato N and Tominaga T 2016 Non-invasive Evaluation for
Epilepsy Surgery Neurol. Med. Chir. (Tokyo) 56 632–40
[44] So E L 2000 Integration of EEG, MRI, and SPECT in localizing the seizure focus for
epilepsy surgery Epilepsia 41 S48–54
[45] Provini F, Plazzi G, Tinuper P, Vandi S, Lugaresi E and Montagna P 1999 Nocturnal frontal
lobe epilepsy. A clinical and polygraphic overview of 100 consecutive cases Brain 122 1017–31
[46] Cukiert A, Buratini J A, Machado E, Sousa A, Vieira J O, Argentoni M, Forster C and
Baldauf C 2001 Results of surgery in patients with refractory extratemporal epilepsy with
normal or nonlocalizing magnetic resonance findings investigated with subdural grids
[47] Ramantani G, Dümpelmann M, Koessler L, Brandt A, Cosandier-Rimélé D, Zentner J,
Schulze-Bonhage A and Maillard L G 2014 Simultaneous subdural and scalp EEG
correlates of frontal lobe epileptic sources Epilepsia 55 278–88
[48] Holmes M D, Kutsy R L, Ojemann G A, Wilensky A J and Ojemann L M 2000 Interictal,
unifocal spikes in refractory extratemporal epilepsy predict ictal origin and postsurgical
outcome Clin. Neurophysiol. 111 1802–8
[49] Janszky J, Jokeit H, Schulz R, Hoppe M and Ebner A 2000 EEG predicts surgical outcome
in lesional frontal lobe epilepsy Neurology 54 1470–6
[50] Laskowitz D T, Sperling M R, French J A and O’Connor M J 1995 The syndrome of
frontal lobe epilepsy: characteristics and surgical management Neurology 45 780–7
[51] Quesney L F 1991 Preoperative electroencephalographic investigation in frontal lobe
epilepsy: electroencephalographic and electrocorticographic recordings Can. J. Neurol.
Sci. 18 559–63
[52] Zaatreh M M, Spencer D D, Thompson J L, Blumenfeld H, Novotny E J, Mattson R H
and Spencer S S 2002 Frontal lobe tumoral epilepsy: clinical, neurophysiologic features and
predictors of surgical outcome Epilepsia 43 727–33
2-33
[53] Morris H H, Dinner D S, Lüders H, Wyllie E and Kramer R 1988 Supplementary motor
seizures: clinical and electroencephalographic findings Neurology 38 1075–82
[54] Boesebeck F, Schulz R, May T and Ebner A 2002 Lateralizing semiology predicts the
seizure outcome after epilepsy surgery in the posterior cortex Brain 125 2320–31
[55] Salanova V, Andermann F, Rasmussen T, Olivier A and Quesney L F 1995 Parietal lobe
epilepsy. Clinical manifestations and outcome in 82 patients treated surgically between 1929
and 1988 Brain 118 607–27
[56] Sveinbjornsdottir S and Duncan J S Parietal and occipital lobe epilepsy: a review Epilepsia
34 493–521
[57] Aykut-Bingol C, Bronen R A, Kim J H, Spencer D D and Spencer S S 1998 Surgical
outcome in occipital lobe epilepsy: Implications for pathophysiology Ann. Neurol. 44 60–9
[58] Williamson P D, Thadani V M, Darcey T M, Spencer D D, Spencer S S and Mattson R H
1992 Occipital lobe epilepsy: clinical characteristics, seizure spread patterns, and results of
surgery Ann. Neurol. 31 3–13
[59] Salanova V, Andermann F, Olivier A, Rasmussen T and Quesney L F 1992 Occipital lobe
epilepsy: electroclinical manifestations, electrocorticography, cortical stimulation and out-
come in 42 patients treated between 1930 and 1991 Surgery of occipital lobe epilepsy Brain
115 1655–80
[60] Ghougassian D F, d’Souza W, Cook M J and O’Brien T J 2004 Evaluating the utility of
inpatient video-EEG monitoring Epilepsia 45 928–32
[61] Palmini A and Gloor P 1992 The localizing value of auras in partial seizures: a prospective
and retrospective study Neurology 42 801–8
[62] Kotagal P, Lüders H, Morris H H, Dinner D S, Wyllie E, Godoy J and Rothner A D 1989
Dystonic posturing in complex partial seizures of temporal lobe onset: a new lateralizing
sign Neurology 39 196–201
[63] Wyllie E, Lüders H, Morris H H, Lesser R P and Dinner D S 1986 The lateralizing
significance of versive head and eye movements during epileptic seizures Neurology 36
606–11
[64] Gabr M, Lüders H, Dinner D, Morris H and Wyllie E 1989 Speech manifestations in
lateralization of temporal lobe seizures Ann. Neurol. 25 82–7
[65] Ebner A, Dinner D S, Noachtar S and Lüders H 1995 Automatisms with preserved
responsiveness: a lateralizing sign in psychomotor seizures Neurology 45 61–4
[66] Kobulashvili T et al 2018 Diagnostic and prognostic value of noninvasive long-term video-
electroencephalographic monitoring in epilepsy surgery: A systematic review and meta-
analysis from the E-PILEPSY consortium Epilepsia 59 2272–83
[67] Ter-Pogossian M M, Phelps M E, Hoffman E J and Mullani N A 1975 A Positron-
Emission Transaxial Tomograph for Nuclear Imaging (PETT) Radiology 114 89–98
[68] Sarikaya I 2015 PET studies in epilepsy Am. J. Nucl. Med. Mol. Imaging 5 416–30
[69] Townsend D W, Wensveen M, Byars L G, Geissbuhler A, Tochon-Danguy H J, Christin A,
Defrise M, Bailey D L, Grootoonk S and Donath A 1993 A rotating PET scanner using
BGO block detectors: design, performance and applications J. Nucl. Med. 34 1367–76
[70] Townsend D W 2008 Combined positron emission tomography-computed tomography: the
historical perspective Semin. Ultrasound. CT. MR 29 232–5
[71] Lynch T B and Kim E E 2008 PET/CT in Clinical Practice J. Nucl. Med. 49 685
[72] Varrone A et al 2009 EANM procedure guidelines for PET brain imaging using [18F]FDG,
version 2 Eur. J. Nucl. Med. Mol. Imaging 36 2103–10
2-34
[73] Leiderman D B, Albert P, Balish M, Bromfield E and Theodore W H 1994 The dynamics of
metabolic change following seizures as measured by positron emission tomography with
fludeoxyglucose F 18 Arch. Neurol. 51 932–6
[74] Ding Y-S, Chen B-B, Glielmi C, Friedman K and Devinsky O 2014 A pilot study in
epilepsy patients using simultaneous PET/MR Am. J. Nucl. Med. Mol. Imaging 4 459–70
[75] Kinahan P E and Fletcher J W 2010 Positron emission tomography-computed tomography
standardized uptake values in clinical practice and assessing response to therapy Semin.
Ultrasound. CT. MR 31 496–505
[76] Alkonyi B, Chugani H T and Juhász C 2011 Transient focal cortical increase of interictal
glucose metabolism in Sturge-Weber syndrome: implications for epileptogenesis Epilepsia
52 1265–72
[77] Peter J, Houshmand S, Werner T J, Rubello D and Alavi A 2015 Applications of global
quantitative 18F-FDG-PET analysis in temporal lobe epilepsy Nucl. Med. Commun. 37 1
[78] Kim Y K, Lee D S, Lee S K, Chung C K, Chung J-K and Lee M C 2002 (18)F-FDG PET
in localization of frontal lobe epilepsy: comparison of visual and SPM analysis J. Nucl.
Med. 43 1167–74
[79] Chassoux F, Artiges E, Semah F, Laurent A, Landré E, Turak B, Gervais P, Helal B-O and
Devaux B 2017 18F-FDG-PET patterns of surgical success and failure in mesial temporal
lobe epilepsy Neurology 88 1045–53
[80] Wong C H, Bleasel A, Wen L, Eberl S, Byth K, Fulham M, Somerville E and Mohamed A
2010 The topography and significance of extratemporal hypometabolism in refractory
mesial temporal lobe epilepsy examined by FDG-PET Epilepsia 51 1365–73
[81] Yıldırım Capraz I, Kurt G, Akdemir Ö, Hirfanoglu T, Oner Y, Sengezer T, Atay Kapucu
L O, Serdaroglu A and Bilir E 2015 Surgical outcome in patients with MRI-negative,
PET-positive temporal lobe epilepsy Seizure 29 63–8
[82] Feng R, Hu J, Wu J, Lang L, Ma C, Jiang S, Sun B, Gu X and Pan L 2017 Comprehensive
preoperative work-up and surgical treatment of low grade tumor/benign lesion related
temporal lobe epilepsy J. Clin. Neurosci. 39 203–8
[83] Kim D W, Lee S K, Yun C-H, Kim K-K, Lee D S, Chung C-K and Chang K-H 2004
Parietal lobe epilepsy: the semiology, yield of diagnostic workup, and surgical outcome
[84] Yang P-F, Jia Y-Z, Lin Q, Mei Z, Chen Z-Q, Zheng Z-Y, Zhang H-J, Pei J-S, Tian J and
Zhong Z-H 2015 Intractable occipital lobe epilepsy: clinical characteristics, surgical
treatment, and a systematic review of the literature Acta Neurochir. (Wien) 157 63–75
[85] Kun Lee S, Young Lee S, Kim D-W, Soo Lee D and Chung C-K 2005 Occipital Lobe
Epilepsy: Clinical Characteristics, Surgical Outcome, and Role of Diagnostic Modalities
[86] Paldino M J, Yang E, Jones J Y, Mahmood N, Sher A, Zhang W, Hayatghaibi S,
Krishnamurthy R and Seghers V 2017 Comparison of the diagnostic accuracy of PET/MRI
to PET/CT-acquired FDG brain exams for seizure focus detection: a prospective study
Pediatr. Radiol. 47 1500–7
[87] Maldonado A, Alvarez J, Perez M and Vilarego F 2012 PET-MRI in presurgical evaluation
of refractory epilepsy associated with focal cortical dysplasia Soc. Nucl. Med. 53 40
[88] Pianou N and Chatziioannou S 2019 Imaging with PET/CT in Patients with Epilepsy
Epilepsy Surgery and Intrinsic Brain Tumor Surgery (Cham: Springer International
Publishing), pp 45–50
2-35
[89] Alkonyi B, Juhász C, Muzik O, Asano E, Saporta A, Shah A and Chugani H T 2009
Quantitative brain surface mapping of an electrophysiologic/metabolic mismatch in human
neocortical epilepsy Epilepsy Res. 87 77–87
[90] Idris Z, Kandasamy R, Reza F and Abdullah J M 2014 Neural oscillation, network,
eloquent cortex and epileptogenic zone revealed by magnetoencephalography and awake
craniotomy Asian J. Neurosurg. 9 144–52
[91] Baumgartner C, Pataraia E, Lindinger G and Deecke L 2000 Magnetoencephalography in
focal epilepsy Epilepsia 41 S39–47
[92] Magnetoencephalography in the presurgical evaluation of epilepsy. - PubMed - NCBI.
[Online]. Available: http://ncbi.nlm.nih.gov/pubmed/25555504.
[93] Simos P G, Rezaie R and Papanicolaou A C 2019 Applications of magnetoencephalog-
raphy in epilepsy and tumor surgery Epilepsy Surgery and Intrinsic Brain Tumor Surgery
(Cham: Springer International Publishing), pp 51–65
[94] Tovar-Spinoza Z S, Ochi A, Rutka J T, Go C and Otsubo H 2008 The role of magneto-
encephalography in epilepsy surgery Neurosurg. Focus 25 E16
[95] Oishi M, Kameyama S, Masuda H, Tohyama J, Kanazawa O, Sasagawa M and Otsubo H
2006 Single and multiple clusters of magnetoencephalographic dipoles in neocortical
epilepsy: significance in characterizing the epileptogenic zone Epilepsia 47 355–64
[96] Medvedovsky M et al 2012 Sensitivity and specificity of seizure-onset zone estimation by
ictal magnetoencephalography Epilepsia 53 1649–57
[97] Papanicolaou A C, Pataraia E, Billingsley-Marshall R, Castillo E M, Wheless J W, Swank
P, Breier J I, Sarkari S and Simos P G 2005 Toward the substitution of invasive
electroencephalography in epilepsy surgery J. Clin. Neurophysiol. 22 231–7
[98] Agirre-Arrizubieta Z, Huiskamp G J M, Ferrier C H, van Huffelen A C and Leijten F S S
2009 Interictal magnetoencephalography and the irritative zone in the electrocorticogram
Brain 132 3060–71
[99] Oishi M, Kameyama S, Masuda H, Tohyama J, Kanazawa O, Sasagawa M and Otsubo H
2006 Single and multiple clusters of magnetoencephalographic dipoles in neocortical
epilepsy: significance in characterizing the epileptogenic zone Epilepsia 47 355–64
[100] Shigeto H, Morioka T, Hisada K, Nishio S, Ishibashi H, Kira D, Tobimatsu S and Kato M
2002 Feasibility and limitations of magnetoencephalographic detection of epileptic dis-
charges: simultaneous recording of magnetic fields and electrocorticography Neurol. Res. 24
531–6
[101] Santiuste M, Nowak R, Russi A, Tarancon T, Oliver B, Ayats E, Scheler G and Graetz G
2008 Simultaneous magnetoencephalography and intracranial EEG registration: technical
and clinical aspects J. Clin. Neurophysiol. 25 331–9
[102] EEG and MEG in mesial temporal lobe epilepsy: where do the spikes really come from? -
PubMed - NCBI. [Online]. Available: http://ncbi.nlm.nih.gov/pubmed/?term=EEG+and+
MEG+in+mesial+temporal+lobe+epilepsy%3A+where+do+the+spikes+really+come+
from%3F
[103] A quantitative assessment of the sensitivity of whole-head MEG to activity in the adult
human cortex. - PubMed - NCBI. [Online]. Available: http://ncbi.nlm.nih.gov/pubmed/?
term=A+quantitative+assessment+of+the+sensitivity+of+whole-head+MEG+to+activity+
in+the+adult+human+cortex
[104] A study of dipole localization accuracy for MEG and EEG using a human skull phantom. -
PubMed - NCBI. [Online]. Available: http://ncbi.nlm.nih.gov/pubmed/9751287
2-36
[105] Hutton B F 2014 The origins of SPECT and SPECT/CT Eur. J. Nucl. Med. Mol. Imaging
41 3–16
[106] Jayalakshmi S, Sudhakar P and Panigrahi M 2011 Role of single photon emission
computed tomography in epilepsy Int. J. Mol. Imaging 2011 803920
[107] Koyama M, Kawashima R, Ito H, Ono S, Sato K, Goto R, Kinomura S, Yoshioka S, Sato
T and Fukuda H 1997 SPECT imaging of normal subjects with technetium-99m-HMPAO
and technetium-99m-ECD J. Nucl. Med. 38 587–92
[108] Paesschen W V 2004 Ictal SPECT Epilepsia 45 35–40
[109] Lee D S, Lee S K, Kim Y K, Lee J S, Cheon G J, Kang K W, Kim E S, Chung J-K and Lee
M C 2002 Superiority of HMPAO ictal SPECT to ECD ictal SPECT in localizing the
epileptogenic zone Epilepsia 43 263–9
[110] Knowlton R C, Lawn N D, Mountz J M and Kuzniecky R I 2004 Ictal SPECT analysis in
epilepsy: subtraction and statistical parametric mapping techniques Neurology 63 10–5
[111] Brinkmann B H, O’Brien T J, Mullan B P, O’Connor M K, Robb R A and So E L 2000
Subtraction ictal SPECT coregistered to MRI for seizure focus localization in partial
epilepsy Mayo Clin. Proc. 75 615–24
[112] Stamoulis C, Verma N, Kaulas H, Halford J J, Duffy F H, Pearl P L and Treves S T 2017
The promise of subtraction ictal SPECT co-registered to MRI for improved seizure
localization in pediatric epilepsies: affecting factors and relationship to the surgical outcome
Epilepsy Res. 129 59–66
[113] Burneo J G, Vezina W, Romsa J, Smith B J and McLachlan R S 2007 Evaluating the
development of a SPECT protocol in a Canadian epilepsy unit Can. J. Neurol. Sci. 34 225–9
[114] Kim S and Mountz J M 2011 SPECT imaging of epilepsy: an overview and comparison
with F-18 FDG PET Int. J. Mol. Imaging 2011 813028
[115] Beleza P and Pinho J 2011 Frontal lobe epilepsy J. Clin. Neurosci. 18 593–600
[116] Velasco T R et al 2002 Accuracy of ictal SPECT in mesial temporal lobe epilepsy with
bilateral interictal spikes Neurology 59 266–71
[117] Kazemi N J, Worrell G A, Stead S M, Brinkmann B H, Mullan B P, O’Brien T J and So
E L 2010 Ictal SPECT statistical parametric mapping in temporal lobe epilepsy surgery
Neurology 74 70–6
[118] Jones A L and Cascino G D 2016 Evidence on use of neuroimaging for surgical treatment
of temporal lobe epilepsy JAMA Neurol. 73 464
[119] Ho S S, Berkovic S F, McKay W J, Kalnins R M and Bladin P F 1996 Temporal lobe
epilepsy subtypes: differential patterns of cerebral perfusion on ictal SPECT Epilepsia 37
788–95
[120] Harvey A S, Hopkins I J, Bowe J M, Cook D J, Shield L K and Berkovic S F 1993 Frontal
lobe epilepsy: clinical seizure characteristics and localization with ictal 99mTc-HMPAO
SPECT Neurology 43 1966–80
[121] Correia P, Vollmar C, Rémi J, la Fougere C and Noachtar S 2018 Ictal SPECT reveals
different epileptogenic zones in frontal lobe epilepsy Epileptic Disord. 20 447–50
[122] Lee S K, Lee S-Y, Yun C-H, Lee H-Y, Lee J-S and Lee D-S 2006 Ictal SPECT in
neocortical epilepsies: clinical usefulness and factors affecting the pattern of hyperperfusion
Neuroradiology 48 678–84
[123] Ho S S, Berkovic S F, Newton M R, Austin M C, McKay W J and Bladin P F 1994
Parietal lobe epilepsy: clinical features and seizure localization by ictal SPECT Neurology
44 2277–84
2-37
[124] Jobst B C, Williamson P D, Thadani V M, Gilbert K L, Holmes G L, Morse R P, Darcey

T M, Duhaime A-C, Bujarski K A and Roberts D W 2010 Intractable occipital lobe
epilepsy: Clinical characteristics and surgical treatment Epilepsia 51 2334–7
[125] Kim S K, Lee D S, Lee S K, Kim Y K, Kang K W, Chung C K, Chung J K and Lee M C
2001 Diagnostic performance of [18F]FDG-PET and ictal [99mTc]-HMPAO SPECT in
occipital lobe epilepsy Epilepsia 42 1531–40
[126] Won H J, Chang K H, Cheon J E, Kim H D, Lee D S, Han M H, Kim I O, Lee S K and
Chung C K 1999 Comparison of MR imaging with PET and ictal SPECT in 118 patients
with intractable epilepsy Am. J. Neuroradiol. 20 593–9
[127] Friston K J, Holmes A P, Poline J B, Grasby P J, Williams S C, Frackowiak R S and
Turner R 1995 Analysis of fMRI time-series revisited Neuroimage 2 45–53
[128] Ogawa S, Lee T M, Kay A R and Tank D W 1990 Brain magnetic resonance imaging with
contrast dependent on blood oxygenation Proc. Natl. Acad. Sci. U.S.A. 87 9868–72
[129] Biswal B, Yetkin F Z, Haughton V M and Hyde J S 1995 Functional connectivity in the
motor cortex of resting human brain using echo-planar MRI Magn. Reson. Med. 34 537–41
[130] Waites A B, Briellmann R S, Saling M M, Abbott D F and Jackson G D 2006 Functional
connectivity networks are disrupted in left temporal lobe epilepsy Ann. Neurol. 59 335–43
[131] de Campos B M, Coan A C, Lin Yasuda C, Casseb R F and Cendes F 2016 Large-scale
brain networks are distinctly affected in right and left mesial temporal lobe epilepsy Hum.
Brain Mapp. 37 3137–52
[132] Lemieux L, Salek-Haddadi A, Josephs O, Allen P, Toms N, Scott C, Krakow K, Turner R
and Fish D R 2001 Event-related fMRI with simultaneous and continuous EEG:
description of the method and initial case report Neuroimage 14 780–7
[133] Coan A C, Campos B M, Beltramini G C, Yasuda C L, Covolan R J M and Cendes F 2014
Distinct functional and structural MRI abnormalities in mesial temporal lobe epilepsy with
and without hippocampal sclerosis Epilepsia 55 1187–96
[134] Mulert C and Lemieux L 2010 EEG-fMRI: physiological basis, technique, and applications
(Berlin: Springer)
[135] Coan A C et al 2016 EEG-fMRI in the presurgical evaluation of temporal lobe epilepsy
J. Neurol. Neurosurg. Psychiatry 87 642–9
[136] Pittau F, Ferri L, Fahoum F, Dubeau F and Gotman J 2017 Contributions of EEG-fMRI
to assessing the epileptogenicity of focal cortical dysplasia Front. Comput. Neurosci. 11 8
[137] Serletis D 2019 Invasive electroencephalography in epilepsy Epilepsy surgery and intrinsic
brain tumor surgery (Cham: Springer), pp 107–14
[138] Voorhies J M and Cohen-Gadol A 2013 Techniques for placement of grid and strip
electrodes for intracranial epilepsy surgery monitoring: Pearls and pitfalls Surg. Neurol. Int.
4 98
[139] de Souza J P S S and Gonzalez-Martinez J 2019 Stereo-electroencephalography (SEEG) in
the diagnosis and evaluation of medically intractable epilepsy Epilepsy Surgery and Intrinsic
Brain Tumor Surgery (Cham: Springer), pp 115–25
[140] Skoch J, Adelson P D, Bhatia S, Greiner H M, Rydenhag B, Scavarda D and Mangano F T
2017 Subdural grid and depth electrode monitoring in pediatric patients Epilepsia 58 56–65
[141] Cukiert A, Sousa A, Machado E, Buratini J A, Forster C and Argentoni M 2000 Results of
surgery in patients with bilateral independent temporal lobe spiking (BITLS) with normal
MRI or bilateral mesial temporal sclerosis (MTS) investigated with bilateral subdural grids
Arq. Neuropsiquiatr. 58 1009–13
2-38
[142] Abel T J, Woodroffe R W, Nourski K V, Moritani T, Capizzano A A, Kirby P, Kawasaki

H, Howard M and Werz M A 2018 Role of the temporal pole in temporal lobe epilepsy
seizure networks: an intracranial electrode investigation J. Neurosurg. 129 165–73
[143] Bekelis K, Radwan T A, Desai A, Moses Z B, Thadani V M, Jobst B C, Bujarski K A, Darcey
T M and Roberts D W 2012 Subdural interhemispheric grid electrodes for intracranial
epilepsy monitoring: feasibility, safety, and utility: clinical article J. Neurosurg. 117 1182–8
[144] Bingaman W E and Bulacio J 2014 Placement of subdural grids in pediatric patients:
technique and results Child’s Nerv. Syst. 30 1897–904
[145] Vale F L, Pollock G, Dionisio J, Benbadis S R and Tatum W O 2013 Outcome and
complications of chronically implanted subdural electrodes for the treatment of medically
resistant epilepsy Clin. Neurol. Neurosurg. 115 985–90
[146] Swartz B E, Rich J R, Dwan P S, DeSalles A, Kaufman M H, Walsh G O and Delgado-
Escueta A V 1996 The safety and efficacy of chronically implanted subdural electrodes: a
prospective study Surg. Neurol. 46 87–93
[147] Schmidt R F, Wu C, Lang M J, Soni P, Williams K A, Boorman D W, Evans J J, Sperling
M R and Sharan A D 2016 Complications of subdural and depth electrodes in 269 patients
undergoing 317 procedures for invasive monitoring in epilepsy Epilepsia 57 1697–708
[148] Wiggins G C, Elisevich K and Smith B J 1999 Morbidity and infection in combined
subdural grid and strip electrode investigation for intractable epilepsy Epilepsy Res. 37
73–80
[149] Lee W S, Lee J K, Lee S A, Kang J K and Ko T S 2000 Complications and results of
subdural grid electrode implantation in epilepsy surgery Surg. Neurol. 54 346–51
[150] So E L and Lee R W 2014 Epilepsy surgery in MRI-negative epilepsies Curr. Opin. Neurol.
27 206–12
[151] Battaglia G, Chiapparini L, Franceschetti S, Freri E, Tassi L, Bassanini S, Villani F,
Spreafico R, D’Incerti L and Granata T 2006 Periventricular nodular heterotopia:
classification, epileptic history, and genesis of epileptic discharges Epilepsia 47 86–97
[152] Tassi L et al 2004 Electroclinical, MRI and neuropathological study of 10 patients with
nodular heterotopia, with surgical outcomes Brain 128 321–37
[153] Devaux B et al 2008 Epilepsy surgery in France: evaluation Neurochirurgie 54 453–65
[154] Guenot M and Isnard J 2008 La thermocoagulation multiple de foyers épileptogènes guidée
par la SEEG (thermo-SEEG) Neurochirurgie 54 441–7
[155] Marnet D, Devaux B, Chassoux F, Landré E, Mann M, Turak B, Rodrigo S, Varlet P and
Daumas-Duport C 2008 Surgical resection of focal cortical dysplasias in the central region
Neurochirurgie 54 399–408
[156] Kellinghaus C, Möddel G, Shigeto H, Ying Z, Jacobsson B, Gonzalez-Martinez J, Burrier
C, Janigro D and Najm I M 2007 Dissociation between in vitro and in vivo epileptogenicity
in a rat model of cortical dysplasia Epileptic Disord. 9 11–9
[157] Adelson P D, O’Rourke D K and Albright A L 1995 Chronic invasive monitoring for
identifying seizure foci in children Neurosurg. Clin. N. Am. 6 491–504
[158] Gonzalez-Martinez J, Bulacio J, Alexopoulos A, Jehi L, Bingaman W and Najm I 2013
Stereoelectroencephalography in the ‘difficult to localize’ refractory focal epilepsy: early
experience from a North American epilepsy center Epilepsia 54 323–30
[159] Lo Russo G, Tassi L, Cossu M, Cardinale F, Mai R, Castana L, Colombo N and Bramerio
M 2003 Focal cortical resection in malformations of cortical development Epileptic Disord.
5 S115–23
2-39
[160] González-Martínez J A, Srikijvilaikul T, Nair D and Bingaman W E 2007 Long-term

seizure outcome in reoperation after failure of epilepsy surgery Neurosurgery 60 873–80
discussion 873–80
[161] Tassi L et al 2002 Focal cortical dysplasia: neuropathological subtypes, EEG, neuro-
imaging and surgical outcome Brain 125 1719–32
[162] Srikijvilaikul T, Najm I M, Hovinga C A, Prayson R A, Gonzalez-Martinez J and
Bingaman W E 2003 Seizure outcome after temporal lobectomy in temporal lobe cortical
dysplasia Epilepsia 44 1420–4
[163] Francione S, Kahane P, Tassi L, Hoffmann D, Durisotti C, Pasquier B and Munari C
1994 Stereo-EEG of interictal and ictal electrical activity of a histologically proved
heterotopic gray matter associated with partial epilepsy Electroencephalogr. Clin.
[164] Abisogun A, Fallil Z, Mehta A and Harden C 2014 Stereo-electroencephalography versus
subdural grids and depth electrode placements: when to use which? (P6.335) Neurology 82
P6.335
[165] Najm I M, Bingaman W E and Lüders H O 2002 The use of subdural grids in the
management of focal malformations due to abnormal cortical development Neurosurg.
Clin. N. Am. 13 87–92 viii–ix
[166] Widdess-Walsh P, Jeha L, Nair D, Kotagal P, Bingaman W and Najm I 2007 Subdural
electrode analysis in focal cortical dysplasia: predictors of surgical outcome Neurology 69 660–7
[167] Burchiel K and Raslan A 2018 Functional neurosurgery and neuromodulation (Amsterdam:
Elsevier)
[168] Vadera S, Marathe A R, Gonzalez-Martinez J and Taylor D M 2013
Stereoelectroencephalography for continuous two-dimensional cursor control in a brain-
machine interface Neurosurg. Focus 34 E3
[169] Vadera S, Mullin J, Bulacio J, Najm I, Bingaman W and Gonzalez-Martinez J 2013
Stereoelectroencephalography following subdural grid placement for difficult to localize
epilepsy Neurosurgery 72 723–9 discussion 729
[170] González-Martínez J and Lachhwani D 2014 Stereoelectroencephalography in children
with cortical dysplasia: technique and results Childs Nerv. Syst. 30 1853–7
[171] González-Martínez J, Mullin J, Bulacio J, Gupta A, Enatsu R, Najm I, Bingaman W,
Wyllie E and Lachhwani D 2014 Stereoelectroencephalography in children and adolescents
with difficult-to-localize refractory focal epilepsy Neurosurgery 75 258–68 discussion 267–8
[172] González-Martínez J, Mullin J, Vadera S, Bulacio J, Hughes G, Jones S, Enatsu R and
Najm I 2014 Stereotactic placement of depth electrodes in medically intractable epilepsy
J. Neurosurg. 120 639–44
[173] Mullin J P, Shriver M, Alomar S, Najm I, Bulacio J, Chauvel P and Gonzalez-Martinez J
2016 Is SEEG safe? A systematic review and meta-analysis of stereo-electroencephalography-
related complications Epilepsia 57 386–401
[174] Ho A L, Feng A Y, Kim L H, Pendharkar A V, Sussman E S, Halpern C H and Grant G A
2018 Stereoelectroencephalography in children: a review Neurosurg. Focus 45 E7
[175] Chassoux F, Navarro V, Catenoix H, Valton L and Vignal J-P 2018 Planning and
management of SEEG Neurophysiol. Clin. 48 25–37
[176] Barba C, Rheims S, Minotti L, Guénot M, Hoffmann D, Chabardès S, Isnard J, Kahane P
and Ryvlin P 2015 Temporal plus epilepsy is a major determinant of temporal lobe surgery
failures Brain 139 444–51
2-40
[177] Iida K and Otsubo H 2017 Stereoelectroencephalography: indication and efficacy Neurol.
Med. Chir. 57 375–85
[178] Serletis D, Bulacio J, Bingaman W, Najm I and González-Martínez J 2014 The stereotactic
approach for mapping epileptic networks: a prospective study of 200 patients J. Neurosurg.
121 1239–46
[179] Goldstein H E, Youngerman B E, Shao B, Akman C I, Mandel A M, McBrian D K,
Riviello J J, Sheth S A, McKhann G M and Feldstein N A 2018 Safety and efficacy of
stereoelectroencephalography in pediatric focal epilepsy: a single-center experience
J. Neurosurg. Pediatr. 22 444–52
[180] Taussig D, Chipaux M, Lebas A, Fohlen M, Bulteau C, Ternier J, Ferrand-Sorbets S,
Delalande O and Dorfmüller G 2014 Stereo-electroencephalography (SEEG) in 65
children: an effective and safe diagnostic method for pre-surgical diagnosis, independent
of age Epileptic Disord. 16 280–95
[181] Maillard L G et al 2017 Stereoelectroencephalography and surgical outcome in poly-
microgyria-related epilepsy: A multicentric study Ann. Neurol. 82 781–94
[182] Steriade C, Martins W, Bulacio J, Morita-Sherman M E, Nair D, Gupta A, Bingaman W,
Gonzalez-Martinez J, Najm I and Jehi L 2019 Localization yield and seizure outcome in
patients undergoing bilateral SEEG exploration Epilepsia 60 107–20
[183] Mullin J P, Shriver M, Alomar S, Najm I, Bulacio J, Chauvel P and Gonzalez-Martinez J
2016 Is SEEG safe? A systematic review and meta-analysis of stereo-electroencephalography-
related complications Epilepsia 57 386–401
[184] de Souza J P S A S, Mullin J, Wathen C, Bulacio J, Chauvel P, Jehi L and Gonzalez-
Martinez J 2018 The usefulness of stereo-electroencephalography (SEEG) in the surgical
management of focal epilepsy associated with ‘hidden’ temporal pole encephalocele: a case
report and literature review Neurosurg. Rev. 41 347–54
2-41
IOP Publishing

Chapter 3
Dyslexia: behavioral and biological correlates
and treatment
Kathleen H Nielsen, Todd L Richards and Virginia W Berninger
In this chapter we draw on findings of programmatic research for over two decades
in a multidisciplinary research center on specific learning disabilities in which the
same inclusion criteria for dyslexia were used in all the studies. First, we explain how
correlations between behavioral measures (phenotypes) and results of different kinds
of genetics analyses identified associations between phenotypes and the genetics
bases of dyslexia. Second, we review correlations between brain imaging tasks and
structural and functional brain variables and how these changed from before to after
instructional interventions for students with dyslexia. Third, we propose future
research on the relationships between genetics variables and brain variables. We also
discuss applications of the various research findings for behavioral assessment of
dyslexia and instructional intervention for dyslexia.
3.1 Introduction
Dyslexia is defined by the International Dyslexia Association as a specific reading
disability that is neurobiological in origin and affects real word reading and
decoding accuracy and fluency, and spelling [1, 2]. It is defined by the World
Health Organization as a specific reading disorder that is brain based with
significant impairment of reading skills despite normal intelligence, educational
and social–cultural opportunities [3].
Different research groups have operationalized the definition of dyslexia some-
what differently for research purposes. Some use low achievement, some use
underachievement compared to intellectual ability, and some like the programmatic
research featured in this chapter use inclusion criteria that combine low achievement
and underachievement in defining dyslexia. In the interdisciplinary research pro-
gram featured in this chapter the same inclusion criteria were used for the purpose of
ascertaining samples in all studies; then two kinds of relationships were investigated

(a) between behavioral markers of dyslexia (phenotypes) and genetic indicators of

dyslexia; and (b) relationships between behavioral tasks in and out of an imaging
scanner and various structural and functional brain measures. The results of these
two kinds of studies of correlations between behavioral and biological variables have
application to assessment of dyslexia, as explained in the chapter. In addition,
effective treatment for dyslexia based on changes in behavioral achievement
measures and changes in brain measures before and after treatment for participants
with dyslexia compared to participants without dyslexia who meet criteria for
typical reading and spelling have application to instructional intervention for
dyslexia, as is also discussed in this chapter. In addition, future research is proposed
to investigate the relationships between the genetic and brain variables. Such
research would add to scientific knowledge of the correlations among biological
variables associated with dyslexia.
3.2 Definitional issues

Not all reading disabilities are dyslexia. Students with a variety of developmental
disabilities or neurological disorders may have learning problems that include but
are not restricted to the hallmark impairments in dyslexia: word level decoding of
unknown words and reading familiar words and spelling written words [4]. Also,
often there is a family history of reading issues reported by one or both parents for
their side(s) of the family, as is found in disorders with a genetic basis. Thus, in the
multi-generational family study of dyslexia [5] featured in this chapter parents were
interviewed regarding family history of dyslexia to confirm a multigenerational
history of dyslexia. Then developmental, medical, and educational histories were
obtained to rule out any conditions except a struggle to learn to read and spell in the
early primary grades. Children who qualified their families for participation
(probands in grades 1–6) had to be low achieving (below the population mean)
and also underachieving (at least one standard deviation) below their verbal
comprehension index score on a test of intellectual abilities, which had to be at
least within the average range (standard score of 90, 25th percentile or higher). Some
of the participants were twice exceptional (being intellectually gifted and having
dyslexia) [6, 7].
In more recent programmatic research, based on what was learned from the
family genetics study, only students in grades 4–9 (upper elementary, middle school,
and transition to high school) with persisting problems in reading and writing
despite early and ongoing interventions were studied to differentiate dyslexia and
other specific learning disabilities affecting writing and reading in otherwise typically
developing students (dysgraphia and oral and written language learning disability,
OWL LD) [8]. For example, students with OWL LD struggle with oral language
learning beginning in the preschool years and may struggle with written language
learning including reading during the school years and often have more difficulty
with reading comprehension than do students with dyslexia [9]. It is estimated that
15%–20% of the population have symptoms of reading disabilities but not all of
these may be dyslexia [10].
3-2
3.3 Phenotypes (behavioral markers) and genetic bases of dyslexia

3.3.1 Phenotypes
Behavioral measures of reading and writing were used to assess phenotypes, the
behavior expression of underlying genetic variables in dyslexia [11, 12]. Impaired
single real word reading, pseudoword (nonword) reading/decoding, and written
spelling are behavioral signs of dyslexia that have been linked to genetic markers
[13]. There is also evidence that components of verbal working memory that support
language learning play an important role in word reading, decoding, and spelling
and may be impaired in individuals with dyslexia: phonological coding (storing and
processing heard and spoken words), orthographic coding (storing and processing
viewed and written words), morphological coding (storing and processing base
words and affixes at the beginning and end of base words), phonological loop
(integrating orthographic codes with oral names), orthographic loop (integrating
orthographic codes with producing written letters), and supervisory attention [5].
Supervisory attention includes focused attention (inhibition of what is irrelevant)
and switching attention (changing the focus of attention) [5, 14].
3.3.2 Genetic findings

Genetic research used to investigate dyslexia has employed a variety of methods:
twin studies, aggregation studies, and segregation studies to show that dyslexia has a
genetic basis and chromosomal linkage studies and DNA variations/alleles to
identify gene candidates [13]. Real word reading accuracy has been associated
through joint segregation and linkage studies with chromosome 15q [15] and real
word reading rate has been associated with DYX1C1 [5]. A linkage study also
associated accuracy of real word reading with chromosome 13 [16]. In a gene locus
study pseudoword reading (decoding) accuracy was associated with the FOXP2
SNP gene locus [17]. A linkage study identified an association between rate of
decoding and chromosome 2q [18]. Joint segregation and linkage studies with
adjustment for verbal intelligence (VIQ) have associated spelling with 6q but
without adjustment for VIQ with 15q, 2q and 9q [19]. Based on quantitative
transmission disequilibrium and normality testing, phonological coding was asso-
ciated with CNTNAP2 for rs2710102 [17]. Orthographic coding, based on a spelling
word choice task, was associated with the DCDC2 allele [5]. Based on joint
segregation and linkage studies, the phonological loop assessed with rapid letter
naming was associated with DYX3 locus on chromosome 2p and 10q [20].
Although these studies are promising, results have been difficult to replicate
probably due to different criteria used in identifying individuals with dyslexia who
participate in these studies [13]. One reason for focusing in this chapter on a specific
programmatic research program over two decades is that the criteria for diagnosing
dyslexia were kept the same over the years in order to identify phenotypes and
genotypes when criteria were kept constant. The inability to replicate findings across
genetic studies may also reflect the great variability in genetic makeup of families
where dyslexia is present.
3-3
3.4 Assessment of dyslexia

Assessment of dyslexia should include measures of word level identification of real
words (accuracy and rate), word decoding of pseudowords (accuracy and rate), and
spelling as well as components of working memory (phonological, orthographic, and
morphological coding, phonological and orthographic loops, and supervisory
attention). The purpose of assessment is not only for diagnosis (identifying dyslexia)
but also for planning appropriate instruction and accommodations. It is important
to keep in mind that such instructional and accommodation needs will change at
different grade levels as curriculum changes across the grades [4].
3.4.1 Family history and neurodevelopmental history

Diagnosis of dyslexia should begin with parental interview regarding familial history
of learning issues, developmental history to determine if prenatal and birth histories
or postnatal developmental history may identify other causes of learning disorders
other than dyslexia [4] and environmental factors including educational opportu-
nities to learn. For example, other genetic disorders, birth trauma, head or other
neurological injury, or other medical problems may be identified that could be
responsible for the reading problems rather than dyslexia. Or, environmental issues
related to family or home stressors, lack of appropriate parenting, or lack of home
literacy experiences could be contributing to problems in learning to read.
3.4.2 Educational history

The educational background of the individual is an important factor in assessment
and may affect the phenotypic expression of dyslexia. Risk factors for dyslexia are
ideally identified early in formal schooling and appropriate intervention linked with
the results of early screening are implemented to prevent the hallmark reading and
spelling problems associated with dyslexia. Unfortunately, many schools do not
have early screening and intervention programs in place, and often dyslexia is not
identified or treated until later grades. Moreover, schools differ in the kinds of
services they offer to students with learning disabilities and teachers and other
educational professionals differ in whether they have received professional prepa-
ration in identifying dyslexia and teaching students with dyslexia. Thus, students
evaluated for dyslexia may have received intervention but not necessarily appro-
priate intervention.
3.4.3 Language
Language is not a unitary construct. Language by ear (listening), language by mouth
(oral expression), language by eye (reading), and language by hand (writing) may
each contribute to learning to read and spell. Also, language is multi-leveled with
subword, word, syntax, and text units. Students with dyslexia often have the most
difficulty with language by ear and mouth at the subword level (phonological) and
language by eye and hand at the subword level (orthographic) and word level
(reading and spelling) [4, 8, 9].
3-4
3.4.4 Motor
Many have argued that multisensory instruction is the appropriate way to teach
students with dyslexia. Yet, research shows that students with dyslexia have
difficulty with motor movement and motor coordination [21]. Research also shows
that motor systems are involved in the brain systems for language by ear, by mouth,
by eye, and by hand [22].
3.5 Phenotypic markers

3.5.1 Verbal comprehension
Research has shown that verbal reasoning/comprehension factors of tests of
intellectual abilities are the best predictor of academic achievement [4, 23]. Thus,
many researchers rely on the verbal comprehension index score instead of overall IQ
score to determine levels of expected achievement. On the one hand, a verbal
comprehension score or IQ score of 90 or above is often used in research studies
because often individuals who have scores below this level (lower limit of average
range) tend to have reading issues that have a different etiology than dyslexia [24].
On the other hand, high verbal reasoning has been known to mask the presence of
dyslexia in individuals who read within the average range but have intellectual
ability to read at a superior or very superior level [6, 7].
3.5.2 Assessing word level reading and spelling skills

Tests of oral reading of single real words on a list without context clues are used to
assess both accuracy and rate of single word identification. Tests of oral reading of
single pseudowords on a list are used to assess accuracy and rate of pseudoword
reading of pronounceable nonwords. Tests of spelling dictated single real words and
pseudowords are used to assess accuracy of written word spelling.
3.5.3 Assessing verbal working memory components most often impaired

in dyslexia [5]
Phonological coding for storing and processing heard and spoken words enables
development of phonological awareness, which is the skill that allows the sounding
out of phonemes or subword sound units in words used in mapping written spelling
onto spoken words or spoken words onto written spelling. Assessment of phono-
logical coding is often based on asking students to listen to and repeat heard
pseudowords. Assessment of phonemic awareness associated with word reading,
decoding, and spelling is often based on asking students to repeat a heard pseudo-
word without a designated phoneme.
Orthographic coding for storing and processing read and spelled words enables
development of orthographic awareness, which is the skill that allows identifying
component letters and their positions and sequences in written words and plays an
important role in learning to read and write. In the early grades this skill can be
assessed by asking students to briefly view a written real word and then decide, after
it disappears, if a displayed letter or letter group was in the previously displayed
3-5
word or all the displayed letters were in that prior word. In the later grades this skill
can be assessed with the same tasks but for an initially displayed visual pseudoword
or with an orthographic choice task (choose the correctly spelled word from choices
all of which sound like a real word when pronounced but only one of which is a
correctly spelled real word).
The phonological loop supports integration of a letter or letter group (ortho-
graphic code) with names (word-level phonological code) under timed conditions.
The phonological loop enables learning to read orally, rapidly and automatically
without slow strategic decoding. The orthographic loop supports rapid automatic
writing of alphabet letters under timed conditions. The orthographic loop enables
learning to spell and compose fluently.
Focused attention is assessed by asking students to name the color of ink in which
a color word is printed, for example, the word red printed in green ink or red printed
in red ink and comparing the amount of extra time it takes if the color of the ink and
the color of the word name are different rather than the same. Switching attention is
assessed by comparing the time costs in rapid naming of orthographic stimuli of the
same category (letters) versus the time costs of rapidly naming orthographic stimuli
of switching categories (alternating letters and numbers).
3.6 Brain imaging behavioral and brain correlates

3.6.1 Brain imaging methods
Multiple brain imaging methods and techniques have been used at the multi-
disciplinary research center featured in this chapter to study dyslexia. These include
MR spectroscopy, structural MRI (to measure volumes and shapes of brain
structures), functional MRI during an imaging task, resting-state functional con-
nectivity (with no task), functional connectivity during an imaging task, diffusion
tensor imaging, electroencepholography, and magnetoencepholography. These
techniques have been used to study the differences between typical readers/spellers
without dyslexia and participants with dyslexia in brain parameters as well as
correlate the brain imaging parameters with behavioral test scores. The imaging
findings have been consistent with the findings of the genetics studies that the
hallmark impairments of students with dyslexia are problems in real word reading,
decoding, and spelling [25–40].
Examples of imaging methods used in the learning disability research center
featured in this article have involved a two hour visit for training and scanning in
each of two different sessions on a Philips 3T Achieva scanner. (1) The reading
session (tasks organized by levels of language) included an MPRAGE T1-weighted
structural scan, a resting-state fMRI scan, and a reading-related task-based fMRI
connectivity scan. (2) The writing session (tasks organized by levels of language)
included a resting-state fMRI scan, DTI scan, and writing-related task-based fMRI
connectivity scan. In order to have the participants perform a writing task during the
fMRI scan, a new fMRI compatible writing pad was created [41]. As part of this
protocol the following MRI series were scanned: (1) 3-plane scout view with gradient
echo pulse sequence: TR/TE 9.8/4.6 ms; field of view 250 × 250 × 50 mm; acquisition
3-6
time 30.3 s; (2) reference scan (used in parallel imaging) with gradient echo pulse
sequence: TR/TE 4.0/0.75 ms; field of view 530 × 530 × 300 mm; acquisition time
44.4 s; (3) fMRI scan with echo-planar gradient echo pulse sequence (single shot):
TR/TE 2000/25 ms; field of view 240 × 240 × 99 mm; slice orientation transverse,
acquisition voxel size 3.0 × 3.08 × 3.0 mm; acquisition matrix 80 × 80 × 33; slice
thickness 3.0, SENSE factor in the AP direction 2.3; epi factor 37; bandwidth in the
EPI frequency direction 1933 Hz, SoftTone factor 3.5, sound pressure 6.1 dB, 5
dummy scans; fold over direction AP, dynamic scans 387, acquisition time 13:08
min s−1; and (4) diffusion tensor imaging (DTI) with echo-planar spin-echo diffusion
pulse sequence: TR/TE 8593/78 ms, slice orientation transverse, field of view 220 ×
220 × 128 mm, voxel size 2.2 × 2.2 × 2.0 mm, b values 0 and 1000, output images 1 b
value at 0 and 32 b values at 1000 with 32 different diffusion vector non-colinear
directions, SoftTone factor 4.0, sound pressure 3.1 dB, bandwidth in the EPI
frequency direction 1557.7 Hz, epi factor 57, acquisition time 9:35.7 min s−1.
Next, DTI scans were collected during which no task was performed. Five DTI
parameters of white matter integrity were measured—fractional anisotropy (FA);
relative anisotropy (RA); axial diffusivity (AD); radial diffusivity (RD); and mean
diffusivity (MD) in 10 brain regions—bilateral optical radiation, corticospinal tract,
inferior longitudinal fasciculus, superior longitudinal fasciculus, and cingulum.
3.6.2 Example brain imaging tasks linked to behavioral phenotypes outside scanner
Subword and word coding: Four tasks used in one study [28] to investigate fMRI
activation in specific brain locations and functional connectivity between locations
in students with and without dyslexia were: (1) phonological processing;
(2) orthographic processing; (3) morphological processing without phonological
shifts; and (4) morphological processing with phonological shifts. Results confirmed
triple word form theory [29] (uniqueness of phonological, orthographic, and
morphological coding), an important finding for learning to read and spell
English, a morphophonemic orthography, for both students with and without
dyslexia. See table 3 in [28] for description and examples of each set of on\off tasks
for isolating each of the unique processes.
Cascading levels of language: A levels-of-language study [37] drew on contribu-
tions from developmental psycholinguistics and neuroimaging to study the multiple
levels (units) of language in the reading brain. Developmental psycholinguistic
research has shown that there is a cascading progression from smaller to increasingly
larger units of language: sub-word sounds to spoken single words to two-word
combinations to multi-word syntax in clauses to multi-syntax/clausal constructions
in text. Each successive level draws on a higher level (unit) of language than the prior
one. Results showed unique connectivity between adjacent levels of language.
For subword grapheme–phoneme judgments, the participant was instructed to
think about the small sounds that could go with each pair of single letters, a single
letter and a letter group, or letter groups and then press yes if each letter and/or letter
group in a pair presented on the screen can stand for the same sound or no if both in
3-7
a pair cannot stand for the same sound. Example of a yes pair is ‘c’ and ‘s’. Example
of no pair is ‘d’ and ‘m’.
For lexical judgments about spelling, the participant was instructed to press yes if
a written word on the screen is a correctly spelled real word, but press no if the
written word on the screen is not a correctly spelled word, even though when
pronounced, it sounds like a real word. Example of a yes item is ‘bus.’ Example of a
no item is ‘eer.’
For one kind of syntactic judgment, the participant was instructed to press yes if
the sentence could be a real sentence that is meaningful because all the words are
spelled correctly and make sense in the sentence, but to press no if the sentence is not
meaningful because one of the words is not spelled correctly for how it is used in the
sentence. The ‘no’ items differed from the ‘yes’ items by only one word which was a
homonym foil. This is an example of a yes sentence: ‘The bee, which buzzes, can sting
you.’ This is an example of a no sentence: ‘The bee, witch buzzes, can sting you.’
For lexical judgments about whether words had true affixes or non-affix foils, the
participant was instructed to press yes if the word has a true affix, but to press no if
the word has the same spelling as an affix but is not an affix. This is an example of a
yes item: untie. This is an example of a no item: under.
For a second kind of syntax judgment, the participant was instructed to press yes
if the sentence is meaningful but press no if not because of an affix error on one
word. This is an example of a yes item: He was unfit physically. This is an example of
a no item: He was unfitted physically.
For multi-sentence text judgments, five written sentences were presented on the
monitor one at a time (each presented for constant time interval). The participant
was instructed to read each of the first four sentences that appeared on the monitor
one at a time and then press yes if the fifth sentence that appeared is true based on
the four prior sentences read or no if it is false. The last one was always a statement
about the accumulated text so far that could be answered true (yes) or false (no).
Example set for a true response follows:

Sentence 1: John handed Bill a note.
Sentence 2: It was from Sarah.
Sentence 3: Sarah had written that she wanted to talk to Bill.
Sentence 4: Bill frowned when he read the note.
Sentence 5: True or False? (press key to answer)
Bill was not pleased with what Sarah had written. (True)
Example set for a false response follows:

Sentence 1: Tomorrow is the day of the picnic.
Sentence 2: If it rains, the picnic will be cancelled.
Sentence 3: Amy listens to the weather report.
Sentence 4: She hopes it will rain.
Sentence 5: True or False? (press key to answer)
Amy wants to go to the picnic. (False)
3-8
3.6.3 Brain imaging before and after treatment

Many studies showed changes in brain imaging parameters from before to after
treatment of those with dyslexia; afterwards they did not differ from those without
dyslexia. Two representative findings include the following.
After instructional intervention for dyslexia that included both phonological loop
training (saying phoneme sounds that go with letters close in time) and orthographic
loop training (writing letter units that go with phoneme sounds close in time), fMRI
functional connectivity to middle frontal gyrus (a working memory region)
normalized on a phonological decoding task during imaging [31].
The cingulo-opercular network plays important roles in control of language
processing [42] and cognition [43]. In response to instruction, the cingulo-opercular
network changed and the changes were predicted by phenotypes for working
memory components supporting language learning [44]. An example of the changes
are shown in figure 3.1 for connectivity between the dorsal cingulate and the inferior
frontal gyrus, which increased after intervention during an fMRI task of word
specific spelling.
3.7 Future research directions

Future programmatic research should move beyond correlations between behavioral
and biological (genetic or brain) measures to correlations between biological
measures (genetics and brain). Table 3.1 proposes, based on findings reported in
this chapter from the programmatic line of research for which ascertainment
procedures for operationalizing the definition of dyslexia were kept constant, those
genetic-brain measure correlations that might be investigated.
3.8 Assessment–instruction links for treatment

3.8.1 Educational and clinical applications
Multiple disciplines have relevant expertise to contribute to evidence-based diag-
nosis of dyslexia and planning evidence-based instructional treatment. Thus, it is
important that interdisciplinary teams participate in school settings to identify
students with dyslexia, plan and implement interventions for dyslexia, evaluate
response to those interventions, and modify the intervention if students do not
respond [4]. Instructional interventions have been validated on the basis of changes
on behavioral measures and also on the basis of changes in brain imaging from
before to after intervention [4, 44–46]. For example, students benefit from instruc-
tion that teaches to all levels of language close in time [46] and from both explicit
skills instruction and self-regulation instruction [44, 45]. Also, students with
diagnosed dyslexia who do not respond to interventions may benefit from partic-
ipating in genetics and imaging studies in medical settings.
3.8.2 Biological and environmental risk factors

A word of caution is in order. Both biological and environmental variables may pose
risk factors for individuals with dyslexia. The definition used by the World Health
3-9
Figure 3.1. Effects of intervention on brain connectivity.
Organization (WHO) and the International Dyslexia Association (IDA) describes

dyslexia as a neurobiological disorder. As such, it is beneficial when assessing for
dyslexia to ask about family history of reading and spelling difficulties because of the
high risk for dyslexia of children who have a parent with the disorder [47]. However,
parents with dyslexia may be less able to provide home literacy opportunities for
their children, thus compounding their children’s risk for reading and spelling
problems [47]. Thus, children with dyslexia may be both at environmental risk and
genetic risk. It is important to obtain educational history to determine what has been
3-10
Table 3.1. Changes in Cingular-Opercular Network in Response to Spelling Instruction Correspondences (arranged in same rows) between behavioral phenotypes
(column 1 on left) used in genetics research (column 2) and fMRI tasks (column 3) used in fMRI brain imaging research (column 4) in programmatic research program
on dyslexia. See table notes for behavioral tests used to assess phenotypes in column 1 See reference list for numbered citations of research studies.
Phenotype Genetic method and findings Brain imaging tasks Brain imaging method and findings
Reading real words

(word identification)
Accuracy real Joint segregation and linkage: On and Off tasks for phoneme, Comparison of dyslexia and non-dyslexia
word readinga Chromosome 15q [15] morpheme (+ or − phonological groups first to identify unique activation for
shifts), and orthographic each of four word forms (see tasks) and
mapping [27] second to compare each word form to each
other word form: Identified cross-word form
mapping networks [27]
Rate real word Comparison of dyslexics with and Not available Not available
readingb without DYX1C1 Elk Allele: differed
in this location [5]
3-11
Reading pseudowords (decoding)
Accuracy Association of SNPs with phenotype: On task: View and orally decode Compare On and Off Tasks for dyslexia and
pseudo-word FOXP2 SNP (rs7782412) correlated written pseudo-words Off task: non-dyslexia groups: Dyslexia group under-
readingc with this phenotype [17] Say if 2 heard pseudowords same activated in BA19/V5 [30]
[30]
Rate pseudo- Linkage: Chromosome 2 signal for rate Not available Not available
word readingd of decoding [18]
Spelling
Spellinge Joint segregation and Linkage: with Write with special pen a letter in Functional connectivity analyses comparing
VIQ adjustment, chromosome 6q; blank in letter string to create dyslexia and control groups: Dyslexics had
without VIQ adjustment, correctly spelled word [36] stronger connectivity in left occipital gyrus
chromosomes, 15q, 2q, 9q [19] and cerebellum [36]
(Continued)
Table 3.1. (Continued )
Phenotype Genetic method and findings Brain imaging tasks Brain imaging method and findings
Working memory components
Phonological Quantitative transmission On Task: listen to and repeat Compare On and Off Tasks for dyslexia and
codingf disequilibrium and normality testing: aural pseudoword Off Task: Say control groups: dyslexics over-activated in
replicated association of nonword if two oral pseudowords are the bilateral frontal and temporal and left
repetition and CNTNAP2 for same [30] parietal [30]
rs2710102 [17]
Orthographic Comparison of dyslexics with DCDC2 On Task: Press yes if 2 words are Compare On and Off Tasks for dyslexia and
codingg allele and non-dyslexics: differed in this correctly spelled Off Task: Press control groups: Dyslexics activated in 11 of
location on choosing correct spelling yes if 2 letter strings match the 20 regions controls did but in 6 regions
among foils [5] (Richards et al [28]) controls did not [28]
3-12
Phonological Joint segregation and linkage: RAN Not available Not available
looph (rapid automatic naming of letters) has
DYX3 locus on chromosome
2p [20]
Orthographic Comparison of dyslexics with DCDC2 Writing with special pen letter Comparison of dyslexia and control groups:
loopi allele or DYX1C1 Elk Allele and that follows visually displayed dyslexics had stronger connectivity than
non-Dyslexics: differed in both letter [27] controls in cerebellum [27]
locations on automatic fast writing of

letters [5]
Switching Joint segregation and linkage: RAS Make judgments on tasks at Correlations between RAS and connectivity
attentionj (rapid switching between letters and multiple levels of language: unique compared to the next higher level:
numerals) was associated with locus Subword, Word, Syntax, and subword (grapheme-phoneme) from left
DYX8 on chromosome 1p; dyslexics Multi-Sentence [46] supramarginal in bilateral cerebellum; for
with and without DCDC2 allele and other levels [46]
non-dyslexics differed on RAS [20]
Note that each of the phenotypes assessed with behavioral and genetic and/or brain measures corresponds to a component of the working memory system supporting
language learning [5].
Behavioral measures used to assess phenotypes in Column 1.
a
Woodcock Reading Mastery or Woodcock Johnson 3 word identification [48, 49].
b
TOWRE sight word efficiency [50].
c
Woodcock Reading Mastery or Woodcock Johnson 3 word attack [48, 49].
d
TOWRE phonolocal decoding efficiency [50].
e
WRAT3 spelling WIAT 2 or WIAT 3 spelling [51, 52].
3-13
f
CTOPP nonword repetition [53].
g
Prepublication UW version of word choice.
h
Prepublication Wolf rapid automatic naming (RAN) or published Wolf and Denckla RAN [54].
i
Prepublication of UW writing alphabet from memory.
j
Prepublication Wolf rapid automatic switching letters and numerals (RAS) or published Wolf and Denckla RAS [54].
provided not only at school but also at home. Both parental resources and school
resources may determine the kinds of interventions possible for a student with a
genetic based learning disorder such as dyslexia.
References
[1] International Dyslexia Association 2002 Definition of Dyslexia Retrieved from dyslexiaida.
org
[2] Lyon G R, Shaywiz S and Shaywitz B 2003 A definition of dyslexia Ann. Dyslexia 53 1–14
[3] World Health Organization (WHO) ICIDH-2 2010 The international classification of
impairments Activities and Participation
[4] Berninger V W 2015 Interdisciplinary Frameworks for Schools: Best Professional Practices for
Serving the Needs of all Students (Washington, DC: American Psychological Association)
Companion Websites with Readings and Resources and Advisory Panel. All royalties go to
Division 16 to support these websites and develop future editions.
[5] Berninger V, Raskind W, Richards T, Abbott R and Stock P 2008 A multidisciplinary
approach to understanding developmental dyslexia within working-memory architecture:
Genotypes, phenotypes, brain, and instruction Dev. Neuropsychol. 33 707–44
[6] Lyman R, Sanders E, Abbott R and Berninger V 2017 Translating interdisciplinary research
on language learning into identifying specific learning disabilities in verbally gifted and
average children and youth J. Behav. Brain Res. 7 227–46
[7] Berninger V and Abbott R 2013 Differences between children with dyslexia who are and are
not gifted in verbal reasoning Gift. Child Q. 57 223–33
[8] Berninger V, Richards T and Abbott R 2015 Differential diagnosis of dysgraphia, dyslexia,
and OWL LD: Behavioral and neuroimaging evidence Read. Writ. 28 1119–53
[9] Silliman E and Berninger V 2011 Cross-disciplinary dialogue about the nature of oral and
written language problems in the context of developmental, academic, and phenotypic
profiles Top. Lang. Disord. 31 6–23
[10] Katusic S K, Colligan R C, Barbaresi W J, Schaid D J and Jacobsen S J 2001 Incidence of
reading disability in a population-based birth cohort’, 1976-1982, Rochester, Minnesota
Mayo Clin. Proc. 76 1081–92
[11] Nielsen K, Abbott R, Griffin W, Lott J, Raskind W and Berninger V 2016 Evidence-based
reading and writing assessment for dyslexia in adolescents and young adults Learn. Disabil.
21 38–56
[12] Peterson R L and Pennington B F 2012 Seminar: Developmental dyslexia Lancet 26 1997–
2007
[13] Raskind W, Peters B, Richards T, Eckert M and Berninger V 2012 The Genetics of reading
disabilities: From phenotype to candidate genes Front. Psychol. 3 601
[14] Altemeier L, Abbott R and Berninger V 2008 Executive functions for reading and writing in
typical literacy development and dyslexia J. Clin. Exp. Neuropsychol. 30 588–606
[15] Chapman N, Igo R, Thomson J, Matsushita M, Brkanac Z, Hotzman T, Berninger V,
Wijsman E and Raskind W 2004 Linkage analyses of four regions previously implicated in
dyslexia: confirmation of a locus on chromosome 15q Am. J. Med. Genet./Neuropsychiatr.
Genet. 131B 67–75
Chapman N, Igo R, Thomson J, Matsushita M, Brkanac Z, Hotzman T, Berninger V,
Wijsman E and Raskind W 2004 Am. J. Med. Genet. Suppl. 03174 9999:1
3-14
[16] Igo R, Chapman N, Berninger V, Matsushita M, Brkanac Z, Rothstein J, Holtzman T,

Nielsen K, Raskind W and Wijsman E 2006 Genomewide scan for real-word reading
subphenotypes of dyslexia: Novel chromosome 13 locus and genetic complexity Am. J. Med.
Genet./Neuropsychiatr. Genet. 141B 15–27
[17] Peters B, Raskind W, Matsushita M, Lisowski M, Vu T, Berninger V, Wijsman E and
Brkanac Z 2011 Replication of CNTNAP2 association with nonword repetition and support
for FOXP2 association with timed reading and motor activities in a dyslexia family sample
J. Neurodev. Disord. 3 39–49
[18] Raskind W, Igo R, Chapman N, Berninger V, Thomson J, Matsushita M, Brkanac Z,
Holzman T, Brown M and Wijsman E 2005 A genome scan in multigenerational families
with dyslexia: Identification of a novel locus on chromosome 2q that contributes to
phonological decoding efficiency Mol. Psychiatry 10 699–711
[19] Rubenstein K, Matsushita M, Berninger V, Raskind W and Wijsman E 2011 Genome scan
for spelling deficits: Effects of verbal IQ on models of transmission and trait gene localization
Behav. Genet. 41 31–42
[20] Rubenstein K, Raskind W, Berninger V, Matsushita M and Wijsman E 2014 Genome scan
for cognitive trait loci of dyslexia: rapid naming and rapid switching of letters, numbers, and
colors Am. J. Med. Genet. Part B: Neuropsychiatr. Genet. 165 345–56
[21] Nielsen K, Henderson S, Barnett A L, Abbott R D and Berninger V 2018 Movement issues
identified in movement ABC2 checklist parent ratings for students with persisting dysgra-
phia, dyslexia, and OWL LD and typical literacy learners Learn. Disabil. 23 10–23
[22] Berninger V, Richards T, Nielsen K, Dunn M, Raskind W and Abbott R 2018 Behavioral
and brain evidence for language by ear, by mouth, by eye, and by hand and motor skills in
literacy learning Int. J. School Educ. Psychol. 7 182–200
[23] Vellutino F R 1991 Introduction to three studies on reading acquisition: Convergent findings
on theoretical foundations of code-oriented versus whole language approaches to reading
instruction J. Educ. Psychol. 83 437–43
[24] Batshaw M L, Roizen N J and Lotrecchiano G R (ed) 2012 Children With Disabilities
(Baltimore, MD: Paul H. Brookes)
[25] Richards T L, Corina D and Serafini S 2000 The effects of a phonologically-driven
treatment for dyslexia on lactate levels as measured by Proton MRSI Am. J. Neuroradiol.
21 916–22
[26] Richards T L, Berninger V W, Aylward E H, Richard A, Thompson J, Nagy W, Carlilse J F,
Dager S R and Abbott R D 2002 Reproducibility of proton MR spectroscopic imaging
(PEPSI): comparison of dyslexic and normal-reading children and effects of treatment on
brain lactate levels during language tasks Am. J. Neuroradiol. 23 1678–85
[27] Richards T, Berninger V, Nagy W, Parsons A, Fields K and Richards A 2005 Brain
activation during language task contrasts in children with and without dyslexia: Inferring
mapping processes and assessing response to spelling instruction Educ. Child Psychol. 22
62–80
[28] Richards T, Aylward E, Berninger V, Field K, Parsons A, Richards A and Nagy W 2006
Individual fMRI activation in orthographic mapping and morpheme mapping after ortho-
graphic or morphological spelling treatment in child dyslexics J. Neurolinguist. 19 56–86
[29] Richards T, Aylward E, Raskind W, Abbott R, Field K, Parsons A, Richards A and Nagy W
2006 Converging evidence for triple word form theory in child dyslexia Dev. Neuropsychol. 30
547–89
3-15
[30] Richards T, Berninger V, Winn W, Stock P, Wagner R, Muse A and Maravilla K 2007
Functional MRI activation in children with and without dyslexia during pseudoword aural
repeat and visual decode: before and after treatment Neuropsychology 21 732–41
[31] Richards T L and Berninger V W 2008 Abnormal fMRI connectivity in children with
dyslexia during a phoneme task: Before but not after treatment J. Neurolinguist. 21 294–304
[32] Richards T, Stevenson J, Crouch J, Johnson L C, Maravilla K, Stock P, Abbott R and
Berninger V 2008 Tract-based spatial statistics of diffusion tensor imaging in adults with
dyslexia Am. J. Neuroradiol. 29 1134–9
[33] Richards T, Berninger V Winn W et al 2009 Differences in fMRI activation between children
with and without spelling disability on 2-back/0-back working memory contrast J. Writ. Res.
1 93–123
[34] Richards T L, Berninger V W, Stock P, Altemeier L, Trivedi P and Maravilla K 2009
Differences between good and poor child writers on fMRI contrasts for writing newly taught
and highly practiced letter forms Read. Writ. 24 493–516
[35] Richards T L, Berninger V W and Fayol M 2009 fMRI activation differences between 11-
year-old good and poor spellers’ access in working memory to temporary and long-term
orthographic representations J. Neurolinguist. 22 327–53
[36] Richards T L et al 2005 Contrasting brain patterns of writing-related DTI parameters, fMRI
connectivity, and DTI-fMRI connectivity correlations in children with and without
dysgraphia or dyslexia Neuroimage Clin. 8 408–21
[37] Richards T, Nagy W, Abbott R and Berninger V 2016 Brain connectivity associated with
cascading levels of language J. Syst. Integr. Neurosci. 2 https://oatext.com/Brain-
Connectivity-Associated-with-Cascading-Levels-of-Language.php#Article.
[38] Richards T, Pettet M, Askren M, Mestre Z, Grabowski T, Yagle K, Wallis P, Northey M,
Abbott R and Berninger V 2017 ERPs while judging meaningfulness of sentences with and
without homonym or morpheme spelling foils: Comparing 4th to 9th graders with and
without spelling disabilities Dev. Neuropsychol. 42 284–97
[39] Richards T L, Berninger V W, Yagle K J, Abbott R D and Peterson R D 2017 Changes in
DTI diffusivity and fMRI connectivity cluster coefficients for students with and without
specific learning disabilities in written language: Brain’s response to writing instruction
J. Nat. Sci. 3 e350
[40] Richards T L, Abbott R D, Yagle K, Peterson D, Raskind W and Berninger V W 2017 Self-
government of complex reading and writing brains informed by cingulo-opercular network
for adaptive control and working memory components for language learning J. Syst. Integr.
Neurosci. 3 1–12
[41] Reitz F, Richards T, Wu K, Boord P, Askren K, Lewis T and Berninger V 2013 A low-cost,
computer-interfaced drawing pad for fMRI studies of dysgraphia and dyslexia Sensors 13
5099–108
[42] Vaden K I Jr, Kuchinsky S E, Cute S L, Alhstrom J B, Dubno J R and Eckert J R 2013 The
cingulo-opercular network provides word-recognition benefit J. Neurosci. 33 18979–86
[43] Dosenbach N U et al 2007 Distinct brain networks for adaptive and stable task control in
humans Proc. Natl. Acad. Sci. U. S. A. 104 11073–8
[44] Richards T L, Abbott R D, Yagle K, Peterson D, Raskind W and Berninger V 2017 Self-
government of brain’s response to instruction informed by cingulo-opercular network for
adaptive control and working memory components for language learning J. Syst. Integr.
Neurosci. 3 1–12
3-16
[45] Berninger V, Richards T and Abbott R 2017 Brain and behavioral assessment of executive
functions for self-regulation of levels of language in reading brain J. Nat. Sci. 3 e464
[46] Berninger V and Wolf B 2016 Dyslexia, Dysgraphia, OWL LD, and Dyscalculia: Lessons
from Teaching and Science 2nd edn (Baltimore: Paul H. Brookes) Also available as e-book.
2nd printing 2019
[47] van Bergen E, de Jong P F, Maassen B and van der Leij A 2014 The effect of parents’ literacy
skills and children’s preliteracy skills on the risk of dyslexia J. Abnorm. Child Psychol. 42
1187–2000
[48] Woodcock R 1987 Woodcock Reading Mastery Test, Revised (WRMT-R) (Circle Pines,
MN: American Guidance Service)
[49] Woodcock R 1990 Woodcock-Johnson Psycho-Educational Battery Revised Tests of
Achievement (Chicago: Riverside)
[50] Torgeson J K, Wagner R K and Rashotte C A 1999 Test of Word Reading Efficiency
(TOWRE) (Austin, TX: Pro-Ed)
[51] Wilkinson G 1993 Wide Range Achievement Tests 3rd edn (Wilmington, DE: Wide Range)
[52] Wechsler D 2009 Wechsler Individual Achievement Test 3rd edn (San Antonio, TX: Pearson)
[53] Wagner R, Torgeson J and Rashotte C 1999 Comprehensive Test of Phonological Processing
(CTOPP) (Austin, TX: Pro-Ed)
[54] Wolf M and Denkla M 2005 RAN/RAS Rapid Automated Naming and Rapid Alternating
Stimulus Tests (Austin, TX: Pro-Ed)
3-17
IOP Publishing

Chapter 4
Cholesterol and oxidized cholesterol derivatives:
potential biomarkers of Parkinson’s disease?
Thomas Nury, Gerard Lizard and Anne Vejux
Neurodegenerative diseases represent an important social issue with an increasing

number of patients concerned. The most current neurodegenerative diseases are
Alzheimer’s disease, Parkinson’s disease and multiple sclerosis. Parkinson’s disease
is a gradual neurodegenerative disorder characterized by the loss of dopaminergic
neurons in the substantia nigra and the accumulation of α-synuclein (Lewy bodies).
In order to propose better management of these patients, it is necessary to better
understand the mechanisms involved in the development of these pathologies and in
particular to search for biomarkers allowing either the diagnosis or the follow-up of
treatments. One possibility envisaged is to study the involvement of lipids and more
particularly cholesterol and its oxidized derivatives oxysterols. Studies have shown
changes in the lipid profile, the treatment with drugs regulating cholesterol and the
development of Parkinson’s disease. Modifications in oxysterol levels in plasma or
cerebrospinal fluid in patients with Parkinson’s disease have also been identified.
Various experimental data benefiting from technical progresses realized to study the
oxisome (oxysterol profile) suggest that oxysterols may play a role in the develop-
ment of the disorder by interacting at the level of cell death, oxidative stress or
inflammation or on α-synuclein accumulation. Consequently, oxysterols could be
considered as new therapeutic targets and used as biomarkers.
4.1 Introduction
Neurodegenerative diseases are a major cause of dependency with a strong impact
on the life quality of people affected but also on their relatives, whether they are
carers or not. Unfortunately, the treatments available for these diseases are currently
only intended to reduce symptomatic disorders and are of varying effectiveness.
Neurodegenerative diseases include two sub-families: demyelinating neurodegener-
ative diseases (such as multiple sclerosis (MS), peroxisomal leukodystrophies

(X-ALD)) and non-demyelinating neurodegenerative diseases (including

Alzheimer’s disease (AD), Parkinson’s disease (PD), Huntington’s disease,
Niemann–Pick disease, amyotrophic lateral sclerosis). Afflictions of brain functions
or more generally of the nervous system characterize these pathologies. PD affects
approximately 6.3 million people worldwide (WHO) and a doubling of the number
of people with Parkinson’s disease is expected in many countries between 2005 and
2030, due to the increase in life expectancy. PD is, after stroke, the main cause of
motor disability in the elderly. Some nations, such as France, which sees the number
of patients with PD increase (167 000 affected by PD in 2015; https://dataviz.
santepubliquefrance.fr/parkinson/), have developed national plans on neurodege-
nerative diseases focused on access to care and support for patients and their carers,
as well as on the development and better coordination of research. As the world’s
population ages, it is becoming necessary and urgent to develop effective therapies
targeting the origins of the neurodegenerative diseases rather than the symptoms,
particularly PD. In order to identify new avenues, it is necessary to know the factors
that may be involved in the development of these pathologies. One of these factors
might be the lipids. Indeed, many biological functions involve lipids for their
regulation; however, they can become toxic at certain concentrations, a phenom-
enon described as lipotoxicity. To support the hypothesis of targeting lipids as a new
therapeutic strategy, we can look at other diseases such as AD or Niemann–Pick
disease where a link has been established between the presence of a modified/
distorted lipid profile and the disease [1–3]. To study the lipid profile, various
analyses are carried out: study of fatty acid metabolism (synthesis, β-oxidation,
elongation and/or desaturation), cholesterol metabolism (synthesis, transformation
into bile acids, esterification, oxidation) and analyses of phospholipids. Based on
these studies, it has been shown that in AD, there is an increase in stearoyl-CoA
desaturase activity (SCD, key enzyme in fatty acid metabolism), especially isomers
SCD1, SCD5a/SCD5b and an enhancement in ELOVL6 (elongation of very long
chain fatty acids protein 6, catalyze the first reaction of long-chain fatty acids
elongation cycle) activity associated with an increase in C18:1/C18:0 and C16:1/
C16:0 (desaturation index) ratios and an increase in C24:1 in the brain, respectively
[4]. Quantity of very long chain fatty acid: C22:0 (docosanoic acid); C24:0
(lignoceric acid); C26:0 (hexacosanoic acid) are enhanced in the brain and plasma
[5]. In PD, there are already some data that tend to highlight the involvement of
lipids in this pathology, including reduced levels of polyunsaturated fatty acids in
raft fractions, as were important decreases in the fatty acid 18:1 (n-9) associated with
an augmentation in stearic acid (18:0) [6, 7]. In PD patients and mild cognitive
impairment, higher phospholipid levels are observed in patients with significant
cognitive impairment than in control patients, due to modifications in the lipid cell
membrane composition leading to their damage and dysfunction [8]. A positive
relationship would be possible between phospholipid levels and cognitive impair-
ment. Other lipids can also be modified during the development of neurodegener-
ative diseases. Concerning the lipids present in the brain, cholesterol and its
oxidation products, oxysterols, can play a role in the evolution of neurodegenerative
diseases, especially in AD, MS and Niemann–Pick disease [9–12]. In PD, we can
4-2
hypothesize that cholesterol and oxysterols could be implicated in the pathophysi-

ology, and some of these molecules might eventually become potential biomarkers.
4.2 Parkinson’s disease

Parkinson’s disease (PD), described by James Parkinson in 1817, is a chronic,
sporadic progressive neurodegenerative disease. The main symptoms of the disease
are motor symptoms: rest tremor, bradykinesia (slow movement), hypertonia
extrapyramidal (stiffness) and postural instability. Non-motor symptoms are also
frequent but more variable depending on the patients and the duration of evolution:
neuropsychiatric symptoms (depression, cognitive disorders and dementia, halluci-
nations), disorders of the REM sleep, neurovegetative symptoms (sphincter disor-
ders, hypotension orthostatic), gastrointestinal symptoms (constipation,
hypersalivation), sensory symptoms (pain, anosmia) [13]. The term ‘Parkinson’s
disease’ refers to a single disease, but a clinical heterogeneity is often mentioned and
could be explained by different pathophysiological mechanisms. Early onset PD is
usually distinguished from later onset PD and forms with predominant tremor are
distinguished from forms with non-dominant tremor (also identified with postural
instability and walking disorders) [14, 15].
Neuropathological symptoms in PD include the loss of dopaminergic neurons in
the substantia nigra (SN) and intraneuronal α-synuclein inclusions termed Lewy
bodies [16, 17]. Following the autopsy, it was established that misdiagnosis is
common, particularly with other Parkinsonian syndromes [18]. Indeed, other
neurodegenerative diseases, such as multi-system atrophy or progressive supra-
nuclear paralysis, are also characterized by the presence of Parkinson’s syndrome;
they are distinguished from PD by the presence of associated symptoms and poor
response to treatment. These diseases are rare and their risk factors have rarely been
studied. The neuroleptic use is another common cause of Parkinson’s syndrome and
should still be sought. Moreover, motor symptoms generally arrive after severe
neurodegeneration and the clinical diagnosis arrives late once 50%–60% of dop-
aminergic neurons are lost in the SN, which decreases the effect of possible
neuroprotective drugs [19]. Non-dopaminergic lesions have also been observed.
Thus, a loss of noradrenergic, cholinergic, serotonergic neurons has been observed in
the brains of patients with PD [20]. These lesions are believed to be responsible for
non-motor symptoms: onset of dementia in late stages of the disease and loss of
norepinephrine-containing neurons in the coercive locus, depressive behaviors and
loss of serotonergic neurons in the Raphe nucleus and walking and balance disorders
associated with cholinergic degeneration of the pedonculopontine nucleus [21].
The biological mechanisms of PD are based on three major dysfunctions:
disturbance of mitochondrial function combined with oxidative stress and an
augmentation in misfolded proteins (defects of the ubiquitin proteasome system)
[22–25]. Inflammation can also contribute to the development of the pathology [26].
For mitochondria, a deficiency in complex I of the respiratory chain is present both
in neurons and glia of the substantia nigra, at the level of skeletal muscle and
platelets [27, 28]. The inhibition of complex I induce an increase of apoptotic
4-3
processes related to mitochondria, associated with a diminution in ATP production

and an increase of free radical generation, these different processes are responsible
for neuronal cell death [28, 29]. Mitochondrial dysfunction and dopamine metab-
olism induce the oxidative stress present at the level of the substantia nigra. In the
nigral neurons of PD patients, different oxidative damages are present: increased
iron, oxidation of proteins, lipid peroxidation (high levels of cholesterol hydro-
peroxide, malondialdehyde (MDA), hydroxynonenal (HNE)), and DNA alterations
(high levels of 8-hydroxydeoxyguanine (OH8dG)) [24, 28, 30]. During post-mortem
studies, decreased levels of GSH and augmented levels of GSSG were detected in
substantia nigra pars compacta. The use of anti-oxidants or molecules capable of
reducing oxidative stress have been evaluated during clinical investigations; some
molecules are not effective, such as melatonin, α-lipoic acid, or β-carotene, with no
effect on the disease [30]. With consumption of vitamin E, the onset of PD can be
delayed but not the symptoms [31, 32]. The use of coenzyme Q10 acts positively on
motor function. During a clinical phase III trial, the use of coenzyme Q10 combined
with vitamin E or creatine induces a significant reduction in the depletion of
dopamine and loss of tyrosine hydroxylase neurons, and diminue lipid peroxidation
and α-synuclein aggregation in the neurons of the substantia nigra [30, 33]. The third
dysfunction, alteration in ubiquitin-proteasome system activity can be considered
the origin of Lewy bodies or Lewy body-like inclusions [34, 35]. At the level of the
substantia nigra of patients, there were observed structural and functional defects at
the level of the 26S proteasome [36, 37]. In PD, inflammatory mediators such as
tumor necrosis factor-α (TNF-α), interleukin 1β, nitric oxide or ROS, secreted by
microglia, have been shown to modulate the progression of neuronal death [38]. All
inflammatory or toxic mediators secreted by microglial and astrocytic cells exert
toxic activity preferentially on dopamine neurons. This is associated with glutamate
release and increased excitotoxicity (damage and death of neurons after excessive
stimulation of glutamatergic receptors) [39]. The neuroinflammation and excitotox-
icity of glutamate create a vicious circle that aggravates nigro-striated degeneration
in PD [39, 40].
Motor impairment is still the primary clinical argument for the diagnosis of PD,
but non-motor symptoms (NMS) often arrive years before motor symptoms and
advance with the disease, worsening the life quality for PD patients [41]. Moreover,
neuropathological changes in the form of aggregated α-synuclein can also be
detected in peripheral tissues (including skin, the gastrointestinal tract and olfactory
bulb) before the onset of motor symptoms [42]. Therefore, NMS and peripheral α-
synuclein deposits could provide a way for an earlier diagnosis and more effective
neuroprotective interventions. Only the anatomo-pathological examination of the
brain can establish the diagnosis of PD with certainty. Currently, the diagnosis is
only clinical and based on the evaluation of the two types of symptoms: motor
symptoms, still called Parkinsonism, and non-motor symptoms. Additional tests,
such as brain magnetic resonance imaging or copper tests, can be used on a case-by-
case basis, most often to rule out other causes of Parkinsonian syndromes with focal
atrophies, while positron emission tomography (PET) and single-photon emission
computerized tomography (SPECT) with specific dopaminergic tracers can confirm
4-4
the nigral degeneration and striatal dopamine deficiency. Nevertheless, the disease is
definitively confirmed after autopsy and it can exclude differential diagnoses, such as
multiple system atrophy or progressive supranuclear palsy, whose symptoms are
similar to those of PD. Lack of diagnostic tools explains why in vivo biomarkers are
urgently needed to differentiate PD from other types of neurodegenerative diseases
and make the diagnosis more reliable, especially in the early stages.
4.3 Cholesterol and Parkinson’s disease

Cholesterol, a lipid of the sterol family, is found in animal origin products
(figure 4.1). The main food sources are cheeses, egg yolk, beef and pork.
Cholesterol synthesis takes place in the endoplasmic reticulum. From the reticulum,
cholesterol is distributed to the other membrane compartments. Synthesis begins
with the condensation of three acetate molecules into hydroxy-methyl-glutarate
(HMG). It is the HMG-CoA synthase (hydroxymethylglutaryl-coenzyme A syn-
thase) in the presence of coenzyme A, an enzyme cofactor, which carries out this
synthesis. Hydroxymethylglutarate is then reduced to mevalonate. Mevalonate is
then decarboxylated to 5-carbon isoprenoid (isopentenyl pyrophosphate and dime-
thylallyl pyrophosphate). The condensation of 6 isoprenoid molecules finally results
in squalene. Finally, the squalene is subjected to the action of the squalene cyclase
which creates cholesterol cycles from the unsaturations present in the squalene.
In the brain, cholesterol is produced locally, mainly by astrocytes. Indeed, the
brain is protected by a blood–brain barrier that prevents the entry of cholesterol-rich
lipoproteins. Cerebral cholesterol is part of the composition of myelin synthesized by
oligodendrocytes but is also involved in synaptic function. Since cholesterol cannot
come from an extracerebral source, changes in its concentration will reflect a
dysfunction in its metabolism. Because of its involvement in the brain functions
mentioned above, it is important to consider the effects of this lipid in neuro-
degenerative diseases and particularly in PD and to evaluate whether modulations of
its metabolism can be causes or consequences for PD [43]. In China, serum levels of
total cholesterol, LDL-C, HDL-C and triglycerides in PD patients are lower than in
controls [44]. The authors make the hypothesis that lipids may be markers of PD
severity [44]. In 2017, at the Pennsylvania State University College of Medicine, a
study, realized on plasma from an NINDS PD Biomarkers Program cohort, showed
that low levels of total cholesterol and LDL are combined with PD [45]. Another
prospective Singapore Chinese Health study assessed the association between
Figure 4.1. Structures of cholesterol and oxysterols mainly present in the brain.
4-5
dietary intake of cholesterol and major fatty acids and the risk of PD [46]. A reduced
risk of PD was combined with an augmented dietary intake of cholesterol [46]. In
some clinical work in the USA, The Netherlands and other countries, high rank of
total cholesterol was combined with a lower risk of PD, while in other works, the
opposite was demonstrated. Using PubMed database, Gudala et al in 2013,
evaluated the link between serum cholesterol and PD; after the analysis of eight
relevant studies, no correlation was demonstrated between serum cholesterol and
PD [47]. An augmentation in cholesterol in the visual cortex but not in the anterior
cingulate cortex has been highlighted, with the reduction of levels of two cholesterol
precursors, lathosterol and 7-dehydrocholesterol, compared to controls [48].
Changes in cholesterol metabolism could be a medical target by turning to the
treatment with drugs that modify cholesterol levels in the body. Satines (simvastatin,
lovastatin, pravastatin) can reduce α-synuclein aggregation, and this in a more-
or-less important way depending on the drug [49–51]. When using a drug for
extracting cholesterol (α-cyclodextrin), in vitro, a decrease in α-synuclein levels at
the membrane level was obtained, as well as its accumulation in the cell bodies
of neurons and synapses, thus reducing α-synuclein aggregation [52]. The 2-
hydroxypropyl-β-cyclodextrin (HPβCD) has been shown to indirectly stimulate
the autophagy-lysosomal pathway, that augmented, in a human neuroglioma cell
line, the autophagic clearance of α-synuclein aggregates [53, 54]. The association of
β-cyclodextrin with curcumin increases the capacity of the curcumin in inhibiting
α-synuclein aggregation and by making preformed aggregates disappear [54–56].
Alpha-synuclein, which contains a cholesterol binding domain, could be induced by
cholesterol in one cell-culture study [49, 57]. A molecule used as a drug to lower
cholesterol levels, statins (or HMG-CoA reductase inhibitors) inhibit the aggrega-
tion of α-synuclein in a neuronal culture and α-synuclein aggregation increases with
the addition of exogenous cholesterol, which suppresses the growth of neurons [49].
On the model of 3D5-cells, food deprivation causes the aggregation of α-synuclein
associated with apoptosis, which is linked with ER stress and SREBP1 activation,
followed by an increase in cholesterol synthesis [58]. Cholesterol, in PD, could act as
a protector against cell death induced by lysosomal membrane permeabilization and
as a stimulator of α-synuclein accumulation [59]. In this study, however, treatment
with lovastatin does not prevent the accumulation of α-synuclein although it
decreases 1-methyl-4-phenylpyridinium (MPP+)-induced cell death by inhibiting
ROS production [59, 60]. The role of statins is still subject to debate, some
presenting them as neuroprotective and others as increasing the risk of developing
PD. The reader can get an idea of the controversy by reading the recent studies
referenced below [61–69]. Paul et al also showed that hypercholesterolemia in an
MPTP model leads to the loss of dopaminergic neurons in the nervous system via a
modification of mitochondrial functions and antioxidant homeostasis [70]. Using the
C. elegans model, it has been shown that neutral cholesterol ester hydrolase 1
(NCEH-1) is able to modulate cholesterol metabolism and protect against toxic
accumulation of synuclein [71]. It is also proposed that some poorly folded proteins
such as GBA protein could impact the correct functioning of lysosomes via
cholesterol accumulation, thus preventing autophagy from functioning to prevent
4-6
synucleinopathies [72]. In an MPP+ model, hypercholesterolemia leads to a

reduction in the number of neurons through increased depolarization of mitochon-
drial membrane potential [73]. Paul et al in a study in hypercholesterolemic mice
show a dysfunction of the mitochondrial complex I associated with the production
of hydroxyl radicals in nigra [74].
These different facts carry the involvement of cholesterol in the pathophysiology
of PD. However, cholesterol can be oxidized and lead to the formation of oxidized
derivatives such as oxysterols.
4.4 Oxysterols and the brain

Oxysterols are produced by the oxidation of cholesterol and contain 27 carbon
atoms. Oxysterols are important signaling molecules for the regulation of the overall
metabolism of cholesterol, bile acids and steroid hormones. Oxysterols, present
in vivo, may have an endogenous origin (synthesis by autoxidation or enzymatically)
or be from food. Oxysterols are present in cholesterol-rich foods: eggs, dried egg
powder, present in many commercial products [75], milk powder, in the composition
of products for children [76], clarified butter (or ghee), cheeses, dairy products, red
meat (beef, pork, veal), ham, liver, kidneys, dried or tinned fish (cod, anchovies,
herring). All products containing cholesterol that are dehydrated, subject to
radiation or high temperatures and those that are cooked in the presence of oxygen
may also contain oxysterols [77, 78]. Cholesterol oxidation can occur on the sterol
nucleus and more particularly on positions 4, 5, 6 and 7 but also on the cholesterol
side chain at positions 22, 24, 25 and 27. Autoxidation of cholesterol occurs
preferentially on the sterol ring, forming 7-hydroxycholesterols (α and β),
7-ketocholesterol, epoxides 5.6 (5α, 6α or 5β, 6β) and cholestantriol [79, 80]. In
the review of Vejux et al, a simplified representation of the auto-oxidation of the
cholesterol is present [81]. Within the tissue, reactive oxygen species (ROS), reactive
nitrogen species (RNS), the Fenton reaction (Fe(II) + H2O2 → Fe(III) + HO− +
HO•), or enzymatic processes (cytochrome P450 (CYP) enzymes) can generate
oxysterols [82, 83]. Some oxysterols can be produced by both processes:
25-hydroxycholesterol and 7α-hydroxycholesterol [84]. Two distinct pathways are
implicated in cholesterol autoxidation: free radical pathways (hydrogen abstraction at
position C7 by HO• or ONOO−) and a non-free radical pathway (involving oxygen,
ozone and molecular chlorine) [85, 86]. Auto-oxidation of the cholesterol side-chain is a
marginal phenomenon. Indeed, side-chain is located in the hydrophobic region of
lipoproteins. It is thus protected from this process. The production of oxysterols such as
27-hydroxycholesterol, 25-hydroxycholesterol or 24(S)-hydroxycholesterol is therefore
mainly the result of enzymatic oxidation. The enzymes involved in the synthesis of
oxysterols belong to the cytochrome P450 (CYP) family at the exception of
25-hydroxylase. The enzyme CYP3A4, 27-hydroxylase (CYP27A1), 7α-hydroxylase
(CYP7A1), 25-hydroxylase, 24-hydroxylase (CYP46A1) synthesize 4β-hydroxycholes-
terol, 27-hydroxycholesterol (or (25R)26-hydroxycholesterol), 7α-hydroxycholesterol,
25-hydroxycholesterol, and 24-hydroxycholesterol, respectively. One oxysterol seems
only to be synthesized in the brain: 24S-hydroxycholesterol (also named cerebrosterol).
4-7
In a healthy brain, CYP46A1 and CYP27A1 are located in neurons and astrocytes.
CYP27A1 is also located in oligodendrocytes.
24S-hydroxycholesterol (24S-OHC) is the oxysterol predominantly localized and
synthetized in the brain, where it represents 80% of the total 24-hydroxycholesterol
pool (figure 4.1) [87, 88]. 24S-hydroxycholesterol is also the exclusive oxysterol that
can pass over the blood–brain barrier, in the brain/bloodstream direction at a ratio
of about 4–7 mg/24 h (table 4.1) [89, 90]. 24S-OHC is the form of cholesterol
elimination in the brain, which helps to maintain constant cholesterol levels in the
brain. 24S-hydroxycholesterol participates in the control of cholesterol homeostasis
in the brain. Indeed, increased levels of 24S-OHC in glial cells, following an increase
in its synthesis in neuronal cells, induce activation of the nuclear liver X receptor
(LXR). Once activated, LXR positively induces the cholesterol transporter ABCA1,
resulting in an increase in cholesterol flow in neural cells [89]. 24-hydroxycholesterol
was shown to have pro-oxidant activities on two cell lines representative of cells
present in the nervous system: neuronal SK-N-BE cell lines (involvement NADPH
oxidase) and an oligodendrocyte 158N cell line (production of superoxide anion and
hydrogen peroxide) [91, 92]. In excess, due to abnormal metabolism of cholesterol,
toxicity on nerve cells can be observed.
Among other oxysterols, 27-hydroxycholesterol (27-OHC), may also be identified
in the brain, where it can be synthesized as in most cells by the enzyme CYP27A1
(figure 4.1). However, Leoni et al showed the presence of 27-OHC in the
cerebrospinal fluid (CSF) from the circulation in a non-pathological population.
The rate of 27-OHC is augmented in CSF in proportion to the interference found at
the blood–brain barrier (BBB) and at the blood–CSF barrier [88]. 27-OHC travels
from the circulation to the brain at a rate of about 4–5 mg/24 h (table 4.1), 27-OHC
is then less present in the cell-rich gray matter than in the white matter (table 4.1) [89,
93]. The conversion of 27-OHC into other polar products can explain the low levels
of 27-OHC in the brain. Indeed, 7α-hydroxylase (CYP7B1), which is highly present
in the brain, sterol 27-hydroxylase (CYP27A1) and 3β-hydroxydelta-5-steroid
Table 4.1. Composition in cholesterol and oxysterols in the brain. The corresponding flow to physiological or
pathological situations are mentioned for each oxysterol.
Oxysterols Traffic flow Rate Flow conditions
24-OHC Brain–blood circulation 2–7 mg/24 h Pathological conditions

7α,25-diHCO Brain–blood circulation 0.5 mg/24 h Pathological conditions
7α,26-diHCO Brain–blood circulation 1 mg/24 h Pathological conditions
7αH,3O-CA Brain–blood circulation 2 mg/24 h Pathological conditions
7KC Brain–blood circulation 2 mg/24 h Pathological conditions
7β-OHC Brain–blood circulation 2 mg/24 h Pathological conditions
3β,5α-diHC-6O Brain–blood circulation 0.1 mg/24 h Pathological conditions
27-OHC Blood–brain circulation 4–5 mg/24 h Pathological conditions
(loss of integrity of
blood–brain barrier)
4-8
dehydrogenase, 27-OHC can participate in this conversion [93, 94]. 7α-hydroxy-3-

oxocholest-4-enoic acid (7αH,3O-CA), the major end metabolite, was found to be
exported from the brain at a rate of about 2 mg/24 h (table 4.1) [90, 95, 96]. This rate
reflects the cerebral consumption of 27-OHC. High levels of 7αH,3O-CA were
associated with the loss of integrity of the BBB. It was then suggested that the
7αH,3O-CA could be used as an indicator of BBB integrity instead of the CSF/
serum albumin ratio [96].
7β-hydroxycholesterol, 7-ketocholesterol and 3β,5α-dihydroxycholestan-6-one
(3β,5α-diHC-6O) were identified as oxysterols exported from the brain at rates of
about 2–0.1 mg/24 h [90]. These data were only observed when total oxysterols were
measured (table 4.1) [90]. Other oxysterols: 7α-hydroxy-3-oxocholest-4-enoic acid
(7αH,3O-CA), 7α,25-dihydroxycholest-4-en-3-one (7α,25-diHCO) and 7α,(25R)26-
hydroxycholest-4-en-3-one (7α,26-diHCO) were also exported from the brain
[90, 97].
A diminution of 24-OHC levels in plasma has been detected in certain neuro-
logical pathologies where a diminution in the number of neurons is observed [98].
Secretion of 24-OHC into the CSF (1% of the total secretion) can also mirror loss
of neurons and neuronal damage [99]. Levels of 24-OHC in CSF are augmented
in patients with demyelinating diseases. An increase in the concentration of
27-hydroxycholesterol (27-OHC) is detected in patients with an impaired BBB
[88, 99]. Higher levels of 24-OHC are present during MS active periods than during
times of remission. Teunissen et al showed a decrease of 24-OHC in patients with
primary progressive MS as well as in elderly patients with recurrent progressive MS
[100]. In AD and mild cognitive impairment, an accumulation of both oxysterols,
24-OHC and 27-OHC, is observed [99]. While the release of 24-OHC from the brain
decreases with age, other authors show that 24-OHC levels in plasma and
cerebrospinal fluid are increased in the early stages of AD [1, 101]. Nevertheless,
the plasma level of 24-OHC is reduced in the more advanced stages [98, 102]. This
could suggest that the amount of oxysterols reflects the number and condition of
neurons.
4.5 Oxysterols in Parkinson’s disease

An association between oxysterols and α-synuclein, that accumulates in Lewy bodies
has been shown [103–105]. Concentrations of oxidized cholesterol metabolites are
elevated in the cerebral cortex of individuals with Lewy body dementia (in
comparison with age-matched controls). In vitro, α-synuclein aggregation is accel-
erated by these oxidized cholesterol metabolites [103], thus proposing a correlation
between oxysterols and higher levels of α-synuclein aggregates. The actions of
27-OHC and 24-OHC have been evaluated on tyrosine hydroxylase, α-synuclein
expression levels, dopamine and noradrenaline levels and also cell death [104]. The
oxysterols 27-OHC and 24-OHC were cytotoxics at 25 and 50 μM (10 and
20 μg ml−1 respectively) on SH-SY5Y (neuronal cells). These data are comparable
to those of Kolsch et al [104, 106]. Tyrosine hydroxylase and phospho-tyrosine
hydroxylase levels are augmented by the 24-OHC, while 27-OHC enhances level
4-9
of α-synuclein and incites apoptosis [104]. The oxysterol 27-OHC augmentes

α-synuclein level, but not mRNA level, independently of LXR pathway. This
augmentation is due to inhibition of the proteasomal degradation of this protein and
a diminution in the heat shock protein 70 (HSP70) [107]. Concerning levels of both
neurotransmitters dopamine and noradrenaline, the use of 27-OHC or 24-OHC plus
27-OHC decreased rates of noradrenaline while 24-OHC alone had no effect.
Oxysterols can modulate neuromodulators and proteins implicated in the physi-
opathology of PD. The machinery of this regulation was determined in SH-SY5Y
cells. Two types of nuclear receptors are involved: estrogen receptors and liver X
receptors (LXR) [108]. The oxysterol 27-OHC is an endogenous ligand of LXR α/β
and was described to have tissue-specific effects and as a SERM (selective estrogen
receptor modulator) [109, 110]. We don’t know its effect on the estrogen receptor
α/βr. By increasing LXRβ-mediated transcription, 27-OHC, positively regulates
α-synuclein expression and reduces trancription of tyrosine hydroxylase mediated by
ERβ [108]. Alpha-synuclein expression is controled by the LXR signaling in
particular by 27-OHC, in SK-N-BE neuroblastoma cells and MO3.13 oligoden-
drocyte cell lines [105]. Theofilopoulos et al identified 24S,25-epoxycholesterol in the
midbrain ventral midbrain of mouse embryos and also confirmed that this molecule
was capable of promoting dopaminergic neurogenesis [111]. This oxysterol also
promotes the differentiation of embryonic stem cells. Its action would be via the
LXR receptor [111]. The authors therefore suggest that LXR ligands could be used
in new therapies for PD [111, 112]. Studies have been conducted to measure the rates
of oxysterols or their derivatives in cerebrospinal fluid and plasma in PD patients,
but have also investigated the possible links between oxysterols and Tau protein.
24-OHC and its esters have already been studied in AD and amyotrophic lateral
sclerosis (ALS) where their levels are decreasing. In 2013, the 24-OHC levels in CSF
are correlated to duration of the disease and 27-OHC levels are increased in CSF of
some patients in a study on Swedish patients [113]. In a recent study, the authors
looked at PD and showed that plasma levels of this oxysterol were decreasing and
that it could be used as a biomarker to assess the gravity of the disease but also to
monitor its progression [114]. A study published the same year compared the levels
in cerebrospinal fluid of 24-OHC, 27-OHC with those of the Tau protein in patients
with Parkinsonism [115]. PD is known to be a synucleinopathy but it is also
associated with genetic polymorphisms regulation tau expression. 24-OHC levels are
high in cerebrospinal fluid in Parkinson’s patients and there is a strong correlation
with Tau protein but there was no measurement or correlation made with
a-synuclein in this study [115]. Another clinical study shows a decrease in plasma
levels of 24-OHC [116]. The consumption of 2% cholesterol-enriched diet for
12 weeks by rabbits induces, in the subtantia nigra of these animals, an augmentation
in α-synuclein [117]. In the brains of these rabbits, no augmentation in cholesterol
content was observed, despite an increase in cholesterol levels in plasma [117]. By
correlating these observations obtained in the rabbit with the in vitro results
previously developed, it can be proposed that hypercholesterolemia associated with
oxidative stress may increase the plasma concentrations of 27-OHC followed by the
induction of LXR and an accumulation of α-synuclein [117]. In kainate-induced
4-10
neuronal injury, it has been demonstrated that this treatment induces an augmenta-
tion in cholesterol in rat brain, associated with an augmentation in 7-ketocholesterol
rates, and therefore often used as a biomarker of oxidative stress [118]. Neuronal
injury is observed when hippocampal slices are put in contact with 7-ketocholesterol,
an oxidative mechanism can control this toxicity [118]. In the work, cited above,
Cheng et al suspected the involvement of the conversion of cholesterol to oxysterols
to explain the absence of connection between cholesterol levels and those of its
precursors [48].
4.6 Potential involvement of oxysterols in mechanisms implicated

in Parkinson’s disease
Early work focused mainly on the possible relationships between the protein
identified as involved in PD and oxysterols. But it might be attractive to focus on
the involvement of oxysterols on other mechanisms or processes involved in
pathophysiology such as cell death, inflammation and oxidative stress. Cell death
is a complex phenomenon, where different types of death can be identified. The most
common types of death are apoptosis (intrinsic and extrinsic apoptosis), autophagy
and necrosis. Other types of death have been described more recently, such as
necroptosis, mitotic catastrophe, ferroptosis, pyroptosis or eryptosis. Some of these
pathways need to be defined to better understand the possible relationships between
oxysterols, cell death and PD. Cell death by apoptosis can be divided into several
categories. Extrinsic apoptosis refers to death by apoptosis induced by extracellular
stress signals that are relayed by specific receptors present in plasma membrane.
Extrinsic apoptosis could correspond to these three signaling pathways: (1) death
receptor signaling associated with activation of the caspase-8/or -10–caspase-3
cascade; (2) death receptor signaling linked to activation of the caspase-8/tBID/
caspase-9/caspase-3 pathway; or (3) ligand deprivation-induced dependence receptor
signaling in association with activation of the caspase-9/caspase-3 cascade [119]. The
intrinsic pathway of apoptosis refers to the activation of apoptosis by intracellular
stress stimuli (DNA damage, oxidative stress, excitotoxicity, ER stress). All signals
converge on the mitochondria which has a central place in this type of cell death.
Two pathways exist, caspase-dependent and caspase-independent intrinsic apoptosis
based on the type of cytoprotection possible [119]. Cell death by autophagy is
characterized by cytoplasmic vacuolation involving, among other things, ATG
proteins, lipidation of microtubule-associated protein 1 light chain 3 (better known as
LC3/Atg8), an increased degradation of autophagic substrates like sequestosome 1
(SQSTM1/p62). It can be considered as a cytoprotective response activated by dying
cells to survive stressful conditions [119].
Apoptosis is described for nucleated cells, such as neuronal cells, but there is a
particular form of cell death: eryptosis. Eryptosis is an erythrocytic form of cell
death contributing to eliminate defective erythrocytes [120]. This cell death is
characterized by the translocation of phosphatidylserine from the inner layer to
the outer leaflet of the plasma membrane, in association with cell membrane
scrambling and erythrocyte shrinkage [120]. In PD, different molecules implicated
4-11
in an abnormal coagulation/hematology system, were identified: prostaglandins,

cyclooxygenases, iron and thromboxanes that may suggest an involvement of
eryptosis in PD. The identification of eryptosis, in erythrocytes of patients has
been demonstrated in a study with 30 patients [121]. In other diseases such as
cardiovascular diseases, increased levels of oxysterols and hypercholesterolemia
are linked. In these conditions, a mixture of oxysterols (7-ketocholesterol, 7α-
hydroxycholesterol, 7β-hydroxycholesterol, 5α,6α-epoxy-cholesterol, 5β,6β-epoxy-
cholesterol cholestan-3β,5α,6β-triol) was able to lead to an oxidative stress-dependent
eryptosis on isolated human healthy erythrocytes [122]. Considering that these
oxysterols are known to induce eryptosis in cardiovascular disease, it can be
suggested that similar oxysterols could be bound with the initiation of eryptosis in
PD supporting the association of oxidative stress with this pathology. In PD patient
fibroblasts, reprogrammed and differentiated into dopaminergic neurons, there have
been highlighted autophagic/lysosomal perturbations, reticulum stress, and extrac-
ellular accumulation of α-synuclein [123]. Oxysterols are inductors of reticulum stress
and autophagy, this induction is sometimes associated with oxidative stress [91, 124,
125]. New oxysterols have been investigated in marine organisms and new oxysterols
have been identified [126]. The authors then studied their effects on Neuro2a neural
cells treated with 6-hydroxydopamine, considered a model for PD. Some of these new
compounds may boost the viability of cells treated with 6-hydroxydopamine and
reduce the formation of ROS in these cells. The only one with interesting activity is
(22E)-24-nor-cholesta-5,22-diene-3β,7β-diol for which the authors are considering use
in future studies as a drug candidate (figure 4.2) [126]. In PD, could oxysterols be
implicated in these two types of cell death? Could oxidative stress participate in the
induction?
Another parameter which could play an important role is inflammation; some
authors highlight its involvement in PD and search treatment can modulate
inflammation. In atherosclerosis, 27-OHC can generate the secretion of certain
cytokines as CCL2 or CXCL8 (IL-8) and could manage M2 polarization [127–129].
In microglia, does the 27-hydroxycholesterol play the same role and change the
development of disease? In the case of oxidative stress, it would be engaging to
determine the signaling pathways that could be induced by oxysterols such as the
Nrf-2 signaling [130].
Figure 4.2. The new marine oxysterol: (22E)-24-nor-cholesta-5,22-diene-3β,7β-diol [126].
4-12
Finally, it could be of interest to establish whether some oxysterols could make

new biomarkers of the disease since different types of oxysterols have been
associated with different steps contributing to PD. These data could permit us to
have an innovative approach and a better knowledge of PD physiopathology and to
identify a new pharmacological approach.
4.7 Conclusion
All the results presented show a potential involvement of oxysterols and cholesterol
in the physiopathogenesis of PD, both in target proteins such as α-synuclein or Tau
protein but also in cellular mechanisms (cell death, oxidative stress and inflamma-
tion). Further studies are of course necessary to confirm these data and to develop
knowledge on the relationships between oxysterols and new forms of cell death such
as ferroptosis. Other oxysterols will also have to be studied in plasma to differentiate
between different neurodegenerative diseases (24-OHC being found in plasma of
patients with AD, multiple sclerosis, ALS or PD) (figure 4.3). On the basis of these
recent results, some oxysterols may constitute PD biomarker, which supports the
possibility that oxysterols can be good biomarkers of neurodegenerative disease.
Figure 4.3. Oxysterols in various neurodegenerative diseases. Based on recent results (in vitro, in vivo or clinical
studies), oxysterols can be used as biomarkers in various neurodegenerative diseases, whether demyelinating
(multiple sclerosis, peroxisomal leukodystrophies) (green gears) or non-demyelinating (Niemann–Pick disease,
AD, PD, Huntington disease) (pink gears). The oxysterols that can be targeted as potential biomarkers are
25-hydroxycholesterol and 24-hydroxycholesterol.
4-13
References
[1] Lutjohann D et al 2000 Plasma 24S-hydroxycholesterol (cerebrosterol) is increased in
Alzheimer and vascular demented patients J. Lipid Res. 41 195–8
[2] Boenzi S, Deodato F, Taurisano R, Goffredo B M, Rizzo C and Dionisi-Vici C 2016
Evaluation of plasma cholestane-3beta,5alpha,6beta-triol and 7-ketocholesterol in inherited
disorders related to cholesterol metabolism J. Lipid Res. 57 361–7
[3] Zarrouk A et al 2015 Fatty acid profiles in demented patients: identification of hexacosa-
noic acid (C26:0) as a blood lipid biomarker of dementia J. Alzheimers Dis. 44 1349–59
[4] Astarita G et al 2011 Elevated stearoyl-CoA desaturase in brains of patients with
Alzheimer’s disease PLoS One 6 e24777
[5] Kou J et al 2011 Peroxisomal alterations in Alzheimer’s disease Acta Neuropathol. 122
271–83
[6] Martin M G, Pfrieger F and Dotti C G 2014 Cholesterol in brain disease: sometimes
determinant and frequently implicated EMBO Rep. 15 1036–52
[7] Fabelo N et al 2011 Severe alterations in lipid composition of frontal cortex lipid rafts from
Parkinson’s disease and incidental Parkinson’s disease Mol. Med. 17 1107–18
[8] Li Z, Zhang J and Sun H 2015 Increased plasma levels of phospholipid in Parkinson’s
disease with mild cognitive impairment J. Clin. Neurosci. 22 1268–71
[9] Zarrouk A et al 2014 Involvement of oxysterols in age-related diseases and ageing processes
Ageing Res. Rev. 18 148–62
[10] Zarrouk A, Hammami M, Moreau T and Lizard G 2015 Accumulation of 24S-hydroxycholesterol
in neuronal SK-N-BE cells treated with hexacosanoic acid (C26:0): argument in favor
of 24S-hydroxycholesterol as a potential biomarker of neurolipotoxicity Rev. Neurol. 171
125–9
[11] Jiang X et al 2011 A sensitive and specific LC-MS/MS method for rapid diagnosis of
Niemann-Pick C1 disease from human plasma J. Lipid Res. 52 1435–45
[12] van de Kraats C et al 2014 Oxysterols and cholesterol precursors correlate to magnetic
resonance imaging measures of neurodegeneration in multiple sclerosis Mult. Scler. 20
412–17
[13] Simuni T and Sethi K 2008 Nonmotor manifestations of Parkinson’s disease Ann. Neurol.
64 S65–80
[14] Selikhova M, Williams D R, Kempster P A, Holton J L, Revesz T and Lees A J 2009 A
clinico-pathological study of subtypes in Parkinson’s disease Brain 132 2947–57
[15] Obeso J A et al 2010 Missing pieces in the Parkinson’s disease puzzle Nat. Med. 16 653–61
[16] Dickson D W et al 2009 Neuropathological assessment of Parkinson’s disease: refining the
diagnostic criteria Lancet Neurol. 8 1150–7
[17] Dexter D T and Jenner P 2013 Parkinson disease: from pathology to molecular disease
mechanisms Free Radical Biol. Med. 62 132–44
[18] Hughes A J, Daniel S E, Ben-Shlomo Y and Lees A J 2002 The accuracy of diagnosis of
parkinsonian syndromes in a specialist movement disorder service Brain 125 861–70
[19] Fearnley J M and Lees A J 1991 Ageing and Parkinson’s disease: substantia nigra regional
selectivity Brain 114 2283–301
[20] Jellinger K A 1999 Post mortem studies in Parkinson’s disease–is it possible to detect brain
areas for specific symptoms? J. Neural Transm. Suppl. 56 1–29
[21] Karachi C et al 2010 Cholinergic mesencephalic neurons are involved in gait and postural
disorders in Parkinson disease J. Clin. Invest. 120 2745–54
4-14
[22] Owen A D, Schapira A H, Jenner P and Marsden C D 1996 Oxidative stress and
Parkinson’s disease Ann. N. Y. Acad. Sci. 786 217–23
[23] Schapira A H, Cooper J M, Dexter D, Jenner P, Clark J B and Marsden C D 1989
Mitochondrial complex I deficiency in Parkinson’s disease Lancet 1 1269
[24] Jenner P 2003 Oxidative stress in Parkinson’s disease Ann. Neurol. 53 S26–36 discussion
S36–8
[25] Moore D J, West A B, Dawson V L and Dawson T M 2005 Molecular pathophysiology of
Parkinson’s disease Annu. Rev. Neurosci. 28 57–87
[26] Hartmann A, Hunot S and Hirsch E C 2003 Inflammation and dopaminergic neuronal loss
in Parkinson’s disease: a complex matter Exp. Neurol. 184 561–4
[27] Imaizumi Y et al 2012 Mitochondrial dysfunction associated with increased oxidative stress
and alpha-synuclein accumulation in PARK2 iPSC-derived neurons and postmortem brain
tissue Mol. Brain 5 35
[28] Fukae J, Mizuno Y and Hattori N 2007 Mitochondrial dysfunction in Parkinson’s disease
Mitochondrion 7 58–62
[29] Perier C et al 2005 Complex I deficiency primes Bax-dependent neuronal apoptosis through
mitochondrial oxidative damage Proc. Natl. Acad. Sci. U. S. A. 102 19126–31
[30] Chiurchiu V, Orlacchio A and Maccarrone M 2016 Is modulation of oxidative stress an
answer? The state of the art of redox therapeutic actions in neurodegenerative diseases
Oxid. Med. Cell Longev. 2016 7909380
[31] Sutachan J J et al 2012 Cellular and molecular mechanisms of antioxidants in Parkinson’s
disease Nutr. Neurosci. 15 120–6
[32] Seidl S E, Santiago J A, Bilyk H and Potashkin J A 2014 The emerging role of nutrition in
Parkinson’s disease Front. Aging Neurosci. 6 36
[33] Li Z et al 2015 The effect of creatine and coenzyme q10 combination therapy on mild
cognitive impairment in Parkinson’s disease Eur. Neurol. 73 205–11
[34] McNaught K S and Jenner P 2001 Proteasomal function is impaired in substantia nigra in
Parkinson’s disease Neurosci. Lett. 297 191–4
[35] Olanow C W and McNaught K S 2006 Ubiquitin-proteasome system and Parkinson’s
disease Mov. Disord. 21 1806–23
[36] McNaught K S, Belizaire R, Isacson O, Jenner P and Olanow C W 2003 Altered
proteasomal function in sporadic Parkinson’s disease Exp. Neurol. 179 38–46
[37] McNaught K S, Belizaire R, Jenner P, Olanow C W and Isacson O 2002 Selective loss of
20S proteasome alpha-subunits in the substantia nigra pars compacta in Parkinson’s disease
Neurosci. Lett. 326 155–8
[38] Hirsch E C and Hunot S 2009 Neuroinflammation in Parkinson’s disease: a target for
neuroprotection? Lancet Neurol. 8 382–97
[39] Barger S W, Goodwin M E, Porter M M and Beggs M L 2007 Glutamate release from activated
microglia requires the oxidative burst and lipid peroxidation J. Neurochem. 101 1205–13
[40] Lee M 2013 Neurotransmitters and microglial-mediated neuroinflammation Curr. Protein
Pept. Sci. 14 21–32
[41] O’Sullivan S S, Williams D R, Gallagher D A, Massey L A, Silveira-Moriyama L and Lees
A J 2008 Nonmotor symptoms as presenting complaints in Parkinson’s disease: a
clinicopathological study Mov. Disord. 23 101–6
[42] Delenclos M, Jones D R, McLean P J and Uitti R J 2015 Biomarkers in Parkinson’s
disease: Advances and strategies Parkinsonism Relat. Disord.
4-15
[43] Paul R, Choudhury A and Borah A 2015 Cholesterol—A putative endogenous contributor
towards Parkinson’s disease Neurochem. Int. 90 125–33
[44] Guo X et al 2014 The serum lipid profile of Parkinson’s disease patients: a study from China
Int. J. Neurosci.
[45] Zhang L et al 2017 Circulating cholesterol levels may link to the factors influencing
Parkinson’s risk Front. Neurol. 8 501
[46] Tan L C et al 2015 Dietary cholesterol, fats and risk of Parkinson’s disease in the Singapore
Chinese Health Study J. Neurol. Neurosurg. Psychiatry
[47] Gudala K, Bansal D and Muthyala H 2013 Role of serum cholesterol in Parkinson’s
disease: a meta-analysis of evidence J. Parkinsons Dis. 3 363–70
[48] Cheng D et al 2011 Lipid pathway alterations in Parkinson’s disease primary visual cortex
PLoS One 6 e17299
[49] Bar-On P et al 2008 Statins reduce neuronal alpha-synuclein aggregation in in vitro models
of Parkinson’s disease J. Neurochem. 105 1656–67
[50] Wolozin B, Wang S W, Li N C, Lee A, Lee T A and Kazis L E 2007 Simvastatin
is associated with a reduced incidence of dementia and Parkinson’s disease BMC Med.
5 20
[51] Koob A O et al 2010 Lovastatin ameliorates alpha-synuclein accumulation and oxidation in
transgenic mouse models of alpha-synucleinopathies Exp. Neurol. 221 267–74
[52] Bar-On P, Rockenstein E, Adame A, Ho G, Hashimoto M and Masliah E 2006 Effects of the
cholesterol-lowering compound methyl-beta-cyclodextrin in models of alpha-synucleinopathy
J. Neurochem. 98 1032–45
[53] Kilpatrick K, Zeng Y, Hancock T and Segatori L 2015 Genetic and chemical activation of
TFEB mediates clearance of aggregated alpha-synuclein PLoS One 10 e0120819
[54] Coisne C, Tilloy S, Monflier E, Wils D, Fenart L and Gosselet F 2016 Cyclodextrins as
emerging therapeutic tools in the treatment of cholesterol-associated vascular and neuro-
degenerative diseases Molecules 21 1748
[55] Masuda M et al 2006 Small molecule inhibitors of alpha-synuclein filament assembly
Biochemistry 45 6085–94
[56] Gautam S, Karmakar S, Bose A and Chowdhury P K 2014 Beta-cyclodextrin and
curcumin, a potent cocktail for disaggregating and/or inhibiting amyloids: a case study
with alpha-synuclein Biochemistry 53 4081–3
[57] Fantini J, Carlus D and Yahi N 2011 The fusogenic tilted peptide (67-78) of alpha-
synuclein is a cholesterol binding domain Biochim. Biophys. Acta 1808 2343–51
[58] Jiang P, Gan M, Lin W L and Yen S H 2014 Nutrient deprivation induces alpha-synuclein
aggregation through endoplasmic reticulum stress response and SREBP2 pathway Front.
Aging Neurosci. 6 268
[59] Eriksson I, Nath S, Bornefall P, Giraldo A M and Ollinger K 2017 Impact of high
cholesterol in a Parkinson’s disease model: Prevention of lysosomal leakage versus
stimulation of alpha-synuclein aggregation Eur. J. Cell Biol. 96 99–109
[60] Arenas F, Garcia-Ruiz C and Fernandez-Checa J C 2017 Intracellular cholesterol
trafficking and impact in neurodegeneration Front. Mol. Neurosci. 10 382
[61] Liu G et al 2017 Statins may facilitate Parkinson’s disease: Insight gained from a large,
national claims database Mov. Disord. 32 913–17
[62] Rozani V et al 2017 Statin adherence and the risk of Parkinson’s disease: A population-
based cohort study PLoS One 12 e0175054
4-16
[63] Saeedi Saravi S S, Saeedi Saravi S S, Khoshbin K and Dehpour A R 2017 Current insights
into pathogenesis of Parkinson’s disease: Approach to mevalonate pathway and protective
role of statins Biomed. Pharmacother. 90 724–30
[64] Schmitt M, Dehay B, Bezard E and Garcia-Ladona F J 2017 U18666A, an activator of
sterol regulatory element binding protein pathway, modulates presynaptic dopaminergic
phenotype of SH-SY5Y neuroblastoma cells Synapse 71 e21980
[65] Saeedi Saravi S S, Saeedi Saravi S S, Arefidoust A and Dehpour A R 2017 The beneficial
effects of HMG-CoA reductase inhibitors in the processes of neurodegeneration Metab.
Brain Dis. 32 949–65
[66] Carroll J A, Race B, Phillips K, Striebel J F and Chesebro B 2017 Statins are ineffective at
reducing neuroinflammation or prolonging survival in scrapie-infected mice J. Gen. Virol.
98 2190–9
[67] Carroll C B and Wyse R K H 2017 Simvastatin as a potential disease-modifying therapy for
patients with Parkinson’s disease: rationale for clinical trial, and current progress
J. Parkinsons Dis. 7 545–68
[68] Poly T N et al 2017 Exploring the association between statin use and the risk of Parkinson’s
disease: a meta-analysis of observational studies Neuroepidemiology 49 142–51
[69] Marques N F et al 2018 Atorvastatin prevents early oxidative events and modulates
inflammatory mediators in the striatum following intranasal 1-methyl-4-phenyl-1,2,3,6-
tetrahydropyridine (MPTP) administration in rats Neurotox. Res. 33 549–59
[70] Paul R, Choudhury A, Kumar S, Giri A, Sandhir R and Borah A 2017 Cholesterol
contributes to dopamine-neuronal loss in MPTP mouse model of Parkinson’s disease:
Involvement of mitochondrial dysfunctions and oxidative stress PLoS One 12 e0171285
[71] Zhang S, Glukhova S A, Caldwell K A and Caldwell G A 2017 NCEH-1 modulates
cholesterol metabolism and protects against alpha-synuclein toxicity in a C. elegans model
of Parkinson’s disease Hum. Mol. Genet. 26 3823–36
[72] Garcia-Sanz P, Orgaz L, Fuentes J M, Vicario C and Moratalla R 2018 Cholesterol and
multilamellar bodies: Lysosomal dysfunction in GBA-Parkinson disease Autophagy 14 717–18
[73] Raju A, Jaisankar P, Borah A and Mohanakumar K P 2018 1-methyl-4-phenylpyridinium-
induced death of differentiated SH-SY5Y neurons is potentiated by cholesterol Ann.
Neurosci. 24 243–51
[74] Paul R et al 2018 Accumulation of cholesterol and homocysteine in the nigrostriatal
pathway of brain contributes to the dopaminergic neurodegeneration in mice Neuroscience
388 347–56
[75] Boselli E, Velazco V, Caboni M F and Lercker G 2001 Pressurized liquid extraction of lipids
for the determination of oxysterols in egg-containing food J. Chromatogr. A 917 239–44
[76] Przygonski K, Jelen H and Wasowicz E 2000 Determination of cholesterol oxidation
products in milk powder and infant formulas by gas chromatography and mass spectrometry
Nahrung 44 122–5
[77] Leonarduzzi G, Sottero B and Poli G 2002 Oxidized products of cholesterol: dietary and
metabolic origin, and proatherosclerotic effects (review) J. Nutr. Biochem. 13 700–10
[78] Yan P S 1999 Cholesterol oxidation products. Their occurrence and detection in our
foodstuffs Adv. Exp. Med. Biol. 459 79–98
[79] Patel R P, Diczfalusy U, Dzeletovic S, Wilson M T and Darley-Usmar V M 1996
Formation of oxysterols during oxidation of low density lipoprotein by peroxynitrite,
myoglobin, and copper J. Lipid Res. 37 2361–71
4-17
[80] Lercker G, Rodriguez-Estrada M and Champaign I A P 2002 Cholesterol oxidation

mechanisms Cholesterol and Phytosterol Oxidation Products: Analysis, Occurrence, and
Biological Effects ed F Guardiola, P C Dutta, R Codony and G P Savage (Urbana, IL: The
American Oil Chemists Society)
[81] Vejux A, Samadi M and Lizard G 2011 Contribution of cholesterol and oxysterols in the
physiopathology of cataract: implication for the development of pharmacological treat-
ments J. Ophthalmol. 2011 471947
[82] Rodriguez I R and Larrayoz I M 2010 Cholesterol oxidation in the retina: implications of
7KCh formation in chronic inflammation and age-related macular degeneration J. Lipid
Res. 51 2847–62
[83] Mutemberezi V, Guillemot-Legris O and Muccioli G G 2016 Oxysterols: From cholesterol
metabolites to key mediators Prog. Lipid Res. 64 152–69
[84] Otaegui-Arrazola A, Menendez-Carreno M, Ansorena D and Astiasaran I 2010 Oxysterols:
a world to explore Food Chem. Toxicol. 48 3289–303
[85] Iuliano L 2011 Pathways of cholesterol oxidation via non-enzymatic mechanisms Chem.
Phys. Lipids 164 457–68
[86] Kulig W, Cwiklik L, Jurkiewicz P, Rog T and Vattulainen I 2016 Cholesterol oxidation
products and their biological importance Chem. Phys. Lipids
[87] Lutjohann D 2006 Cholesterol metabolism in the brain: importance of 24S-hydroxylation
Acta Neurol. Scand. Suppl. 185 33–42
[88] Leoni V, Masterman T, Patel P, Meaney S, Diczfalusy U and Bjorkhem I 2003 Side chain
oxidized oxysterols in cerebrospinal fluid and the integrity of blood-brain and blood-
cerebrospinal fluid barriers J. Lipid Res. 44 793–9
[89] Bjorkhem I 2006 Crossing the barrier: oxysterols as cholesterol transporters and metabolic
modulators in the brain J. Intern. Med. 260 493–508
[90] Iuliano L et al 2015 Cholesterol metabolites exported from human brain Steroids 99
189–93
[91] Nury T et al 2015 Induction of oxiapoptophagy on 158N murine oligodendrocytes treated by
7-ketocholesterol-, 7beta-hydroxycholesterol-, or 24(S)-hydroxycholesterol: Protective effects
of alpha-tocopherol and docosahexaenoic acid (DHA; C22:6 n-3) Steroids 99 194–203
[92] Gamba P et al 2011 Interaction between 24-hydroxycholesterol, oxidative stress, and
amyloid-beta in amplifying neuronal damage in Alzheimer’s disease: three partners in crime
Aging Cell 10 403–17
[93] Heverin M, Meaney S, Lutjohann D, Diczfalusy U, Wahren J and Bjorkhem I 2005
Crossing the barrier: net flux of 27-hydroxycholesterol into the human brain J. Lipid Res. 46
1047–52
[94] Stapleton G et al 1995 A novel cytochrome P450 expressed primarily in brain J. Biol. Chem.
270 29739–45
[95] Meaney S et al 2007 Novel route for elimination of brain oxysterols across the blood-brain
barrier: conversion into 7alpha-hydroxy-3-oxo-4-cholestenoic acid J. Lipid Res. 48 944–51
[96] Saeed A et al 2014 7alpha-hydroxy-3-oxo-4-cholestenoic acid in cerebrospinal fluid reflects
the integrity of the blood-brain barrier J. Lipid Res. 55 313–18
[97] Crick P J et al 2015 Quantitative charge-tags for sterol and oxysterol analysis Clin. Chem.
61 400–11
[98] Bretillon L et al 2000 Plasma levels of 24S-hydroxycholesterol in patients with neurological
diseases Neurosci. Lett. 293 87–90
4-18
[99] Leoni V 2009 Oxysterols as markers of neurological disease–a review Scand. J. Clin. Lab.
Invest. 69 22–5
[100] Teunissen C E, Dijkstra C D, Polman C H, Hoogervorst E L, von Bergmann K and
Lutjohann D 2003 Decreased levels of the brain specific 24S-hydroxycholesterol
and cholesterol precursors in serum of multiple sclerosis patients Neurosci. Lett. 347
159–62
[101] Papassotiropoulos A et al 2002 24S-hydroxycholesterol in cerebrospinal fluid is elevated in
early stages of dementia J. Psychiatr. Res. 36 27–32
[102] Papassotiropoulos A et al 2000 Plasma 24S-hydroxycholesterol: a peripheral indicator of
neuronal degeneration and potential state marker for Alzheimer’s disease Neuroreport 11
1959–62
[103] Bosco D A et al 2006 Elevated levels of oxidized cholesterol metabolites in Lewy body
disease brains accelerate alpha-synuclein fibrilization Nat. Chem. Biol. 2 249–53
[104] Rantham Prabhakara J P, Feist G, Thomasson S, Thompson A, Schommer E and Ghribi O
2008 Differential effects of 24-hydroxycholesterol and 27-hydroxycholesterol on tyrosine
hydroxylase and alpha-synuclein in human neuroblastoma SH-SY5Y cells J. Neurochem.
107 1722–9
[105] Cheng D, Kim W S and Garner B 2008 Regulation of alpha-synuclein expression by liver X
receptor ligands in vitro Neuroreport 19 1685–9
[106] Kolsch H, Ludwig M, Lutjohann D and Rao M L 2001 Neurotoxicity of 24-
hydroxycholesterol, an important cholesterol elimination product of the brain, may be
prevented by vitamin E and estradiol-17beta J. Neural Transm. 108 475–88
[107] Schommer J, Marwarha G, Schommer T, Flick T, Lund J and Ghribi O 2018 27-
Hydroxycholesterol increases alpha-synuclein protein levels through proteasomal inhibition
in human dopaminergic neurons BMC Neurosci. 19 17
[108] Marwarha G, Rhen T, Schommer T and Ghribi O 2011 The oxysterol 27-hydroxycholesterol
regulates alpha-synuclein and tyrosine hydroxylase expression levels in human neuroblastoma
cells through modulation of liver X receptors and estrogen receptors–relevance to Parkinson’s
disease J. Neurochem. 119 1119–36
[109] DuSell C D, Umetani M, Shaul P W, Mangelsdorf D J and McDonnell D P 2008
27-hydroxycholesterol is an endogenous selective estrogen receptor modulator Mol.
Endocrinol. 22 65–77
[110] Umetani M et al 2007 27-Hydroxycholesterol is an endogenous SERM that inhibits the
cardiovascular effects of estrogen Nat. Med. 13 1185–92
[111] Theofilopoulos S et al 2013 Brain endogenous liver X receptor ligands selectively promote
midbrain neurogenesis Nat. Chem. Biol. 9 126–33
[112] Griffiths W J et al 2017 Cholesterolomics: An update Anal. Biochem. 524 56–67
[113] Bjorkhem I et al 2013 Oxysterols and Parkinson’s disease: evidence that levels of 24S-
hydroxycholesterol in cerebrospinal fluid correlates with the duration of the disease
Neurosci. Lett. 555 102–5
[114] Di Natale C et al 2018 The level of 24-hydroxycholesteryl esters decreases in plasma of
patients with Parkinson’s disease Neurosci. Lett. 672 108–12
[115] Bjorkhem I, Patra K, Boxer A L and Svenningsson P 2018 24S-Hydroxycholesterol
correlates with tau and is increased in cerebrospinal fluid in Parkinson’s disease and
corticobasal syndrome Front. Neurol. 9 756
[116] Huang X et al 2019 Brain cholesterol metabolism and Parkinson’s disease Mov. Disord.
4-19
[117] Marwarha G and Ghribi O 2014 Does the oxysterol 27-hydroxycholesterol underlie
Alzheimer’s disease-Parkinson’s disease overlap? Exp. Gerontol. 68 13–8
[118] Ong W Y, Goh E W, Lu X R, Farooqui A A, Patel S C and Halliwell B 2003 Increase in
cholesterol and cholesterol oxidation products, and role of cholesterol oxidation products in
kainate-induced neuronal injury Brain Pathol. 13 250–62
[119] Galluzzi L et al 2012 Molecular definitions of cell death subroutines: recommendations of
the Nomenclature Committee on Cell Death 2012 Cell Death Differ. 19 107–20
[120] Lang E and Lang F 2015 Triggers, inhibitors, mechanisms, and significance of eryptosis: the
suicidal erythrocyte death Biomed. Res. Int. 2015 513518
[121] Pretorius E, Swanepoel A C, Buys A V, Vermeulen N, Duim W and Kell D B 2014
Eryptosis as a marker of Parkinson’s disease Aging 6 788–819
[122] Tesoriere L, Attanzio A, Allegra M, Cilla A, Gentile C and Livrea M A 2014 Oxysterol
mixture in hypercholesterolemia-relevant proportion causes oxidative stress-dependent
eryptosis Cell. Physiol. Biochem. 34 1075–89
[123] Fernandes H J et al 2016 ER stress and autophagic perturbations lead to elevated
extracellular alpha-synuclein in GBA-N370S Parkinson’s iPSC-derived dopamine neurons
Stem Cell Rep. 6 342–56
[124] Nury T et al 2013 Induction of oxiapoptophagy, a mixed mode of cell death associated with
oxidative stress, apoptosis and autophagy, on 7-ketocholesterol-treated 158N murine
oligodendrocytes: Impairment by alpha-tocopherol Biochem. Biophys. Res. Commun.
[125] Sudo R, Sato F, Azechi T and Wachi H 2015 7-Ketocholesterol-induced lysosomal
dysfunction exacerbates vascular smooth muscle cell calcification via oxidative stress
Genes Cells 20 982–91
[126] Kolesnikova S A, Lyakhova E G, Kalinovsky A I, Popov R S, Yurchenko E A and Stonik V A
2018 Oxysterols from a marine sponge Inflatella sp. and their action in 6-hydroxydopamine-
induced cell model of Parkinson’s disease Mar. Drugs 16 458
[127] Marengo B et al 2016 Oxysterol mixture and, in particular, 27-hydroxycholesterol drive M2
polarization of human macrophages Biofactors 42 80–92
[128] Kim S M et al 2015 27-Oxygenated cholesterol induces expression of CXCL8 in macro-
phages via NF-kappaB and CD88 Biochem. Biophys. Res. Commun. 463 1152–8
[129] Kim S M, Lee S A, Kim B Y, Bae S S, Eo S K and Kim K 2013 27-Hydroxycholesterol
induces recruitment of monocytic cells by enhancing CCL2 production Biochem. Biophys.
Res. Commun. 442 159–64
[130] Vurusaner B et al 2016 Nrf2 antioxidant defense is involved in survival signaling elicited by
27-hydroxycholesterol in human promonocytic cells Free Radic. Biol. Med. 91 93–104
4-20
IOP Publishing

Chapter 5
Computer assisted diagnosis of gait dynamics
neurodegenerative diseases using a machine
learning approach
J Prasanna, S Thomas George and M S P Subathra
In worldwide population statistics, millions of people are affected by neurodegener-

ative disease namely Huntington’s, Parkinson’s, and amyotrophic lateral sclerosis
disease. Neurodegenerative disease is characterized by the annihilation of the
neurons in the brain which causes the influences in the person physically and
mentally. The walking pattern of the subject with neurodegenerative disease
fluctuates from healthy subjects. In the present study, the classification of individual
neurodegenerative disease and healthy subjects is investigated using gait dynamics.
The time series gait signals are used to determine stance, stride, swing and double
support phase interval. The effective feature extraction techniques, namely local
gradient pattern (LGP) and local neighbor gradient pattern (LNGP), are proposed.
The extracted histogram based features from the time series are classified using
artificial neural network (ANN) classifier. The experimental result shows that
LNGP features produce more effective classification performance than LGP
features.
5.1 Introduction
In recent research, the classification of neurological disease using gait dynamics has
been of high clinical importance. Gait signal is one of the useful records of abnormal
movements caused by the central nervous system and a glitch in the region of the
brain. Fluctuation of walking patterns affects the quality life of the patient [1]. Gait
fluctuation happens due to the specific muscular and neurological pathology [2].
Gait recordings are obtained from the subjects while they walk. Neurological
abnormality by the neurons’ death or the loss of structure of the neurons leads to
the damage in the spinal cord [3] which causes the neurodegenerative diseases, for
instance Parkinson’s, Huntington’s and amyotrophic lateral sclerosis (ALS).

Although these diseases cannot be restored to health completely, their early

detection may improve livability of the subject [4].
In literature, the classification of neurodegenerative diseases, healthy versus
Parkinson’s, healthy versus Huntington’s, healthy versus ALS, many feature
extraction and feature selection methods have been proposed. Here some of the
literature related to this study is reviewed. In [5] the four different features such as
minimum, maximum, standard deviation, and mean features are extracted from the
time series. The selected best features are classified using support vector machine
(SVM) classifier with different kernels. This method achieved better performance
using radial based kernel for the classification of different neurodegenerative classes.
In [6] the cross correlation and autocorrelation between the time series is calculated.
The best 500 features are examined by computing mutual information between the
features. Those features are classified using decision tree classifier and achieved a
classification accuracy of 88.5% for Huntington’s versus healthy, 92.3% for
Parkinson’s versus healthy and 96.2% for the ALS versus healthy. In [7] for each
stride interval the probability density function is estimated and the best features are
computed from Kullback–Leibler divergence. Those features are fed in to the least
squares support vector machine (LS-SVM) and attained an accuracy of 82.8% for
the classification of 13 ALS and 16 healthy subjects. In [8] the wavelet analysis is
performed on the time series based on gait to decompose the signal. The extracted
features that are given as an input to the SVM classifier attained a 90.32% accuracy
for the classification of healthy and Parkinson’s. In [9] for the statistical analysis,
discriminant analysis is performed for individual left and right stride, left and right
swing, left and right stance and double support interval. The study concluded that in
ALS, Parkinson’s and Huntington’s the right swing interval differed from the
healthy, so the efficient classification is provided by the right swing interval. In [10]
the gait characteristics are comprehended using Fourier analysis which assumes the
gait signal to be linear and stationary which is not certain. So, non-linear approach
recurrence quantification analysis (RQA) and statistical analysis are utilized for the
feature extraction. The optimum features are selected by the hill-climbing feature
selection method. These optimum features are classified by two different classifiers,
namely SVM and probabilistic neural network (PNN) classifier. In [11] phase
synchronization and conditional entropy is applied on gait time series signal to
analyze the fluctuation in gait rhythm. This study exposes that those with neuro-
degenerative diseases have less robust phase synchronization and maximum condi-
tional entropy than a healthy subject. In [12] the gait fluctuation for ALS subject is
measured using signal turns count method by computing swing interval turns count
and averaged stride interval. This method achieved a classification accuracy of
89.66% by LS-SVM classifier with sigmoid kernels. In [13] nine statistical based
features are extracted from the time series. The optimum features are selected by hill-
climbing feature selection method. Four different classifiers are employed to perform
the classification. They have achieved an accuracy of 96.83% for healthy versus
neurodegenerative disease. In [14] the gait dynamics of the neurodegenerative and
healthy subjects are classified using deterministic learning theory and radial basis
function. In the present study the classification task is performed between each
5-2
neurodegenerative disease and healthy control subjects based on the gait recordings.
Local gradient pattern (LGP) and local neighbor gradient pattern (LNGP)
techniques are investigated, which could be helpful in discriminating the gait
fluctuations. The block diagram of the proposed technique is presented in figure 5.1.
LGP [15] and LNGP [15, 16] proposed for the diagnosis of epileptic seizures. The
study is organized as follows: feature extraction methods are summarized in section
5.2. The classification result for individual diseases (Parkinson’s versus healthy,
Huntington’s versus healthy, ALS versus healthy) are examined and discussed in
section 5.3. Finally the conclusions are in section 5.4.
5.2 Methodology
5.2.1 Gait dataset
The benchmark dataset gait dynamics neurodegenerative disease is used in this
study. The dataset consists of 64 gait recordings from 20 patients with Huntington’s,
15 patients with Parkinson’s, 13 patients with ALS and 16 healthy patients [17]. The
recordings were obtained by asking the subjects to walk a 77 m length hall for 5 min.
Ultra-thin force sensitive switches are located under the patient’s foot to measure the
gait signal by utilizing an on-board 12 bit resolution analog-to-digital converter with
the sampling frequency of 300 Hz. The dataset comprises the recording of gait
fluctuations of left and right stride, left and right swing, left and right stance and
double support intervals. The time series gait signal waveform is shown in figure 5.2.
5.2.2 Feature extraction

5.2.2.1 Local gradient pattern (LGP)
In this study LGP technique is developed to convert the time series signal into binary
patterns [15] using thresholding condition by comparing gradient values of neigh-
borhood and average gradient values with respect to the center and neighborhood
Figure 5.1. Block diagram of the proposed method.
5-3
Parkinson Disease
Amplitude
-80
-90
-100
50 100 150 200 250 300

Time period (Sec)
Huntington Disease
Amplitude
400
300
200
50 100 150 200 250 300
Time period (Sec)
ALS Disease
-680
Amplitude
-685
-690
50 100 150 200 250 300

Time period (Sec)
Healthy control
530
Amplitude
520
510
500
50 100 150 200 250 300
Time period (Sec)
Figure 5.2. The visual representation of gait time series signals of neurodegenerative disease.
Figure 5.3. The illustration of LGP technique.
sample values of the time series value [18]. The operation of LGP is illustrated in
figure 5.3. The steps involved in LGP are given below:
Step 1: The center values (MC ) and neighboring sample values (Mi ) are
initialized.
5-4
Step 2: The neighboring samples are divided into M/2 points to the left and right
of the center value. The neighboring points towards most significant bits from
the center point (Mc ) are M0, M1, M2, M3 and the neighboring points towards
least significant bits from the center point (Mc ) are M4, M5, M6, M7 .
Step 3: For each sample value the gradient value is calculated and the average of
gradient value computed by the following formula:
1 m −1
gravg = ∑i=0 gri (5.1)
P
where gri = ∣Mi − Mc ∣.
Step 4: The comparison of each gradient value and (gri ) and average gradient
value (gravg ) is performed. If gri is greater than gravg then value 1 will be
allotted, otherwise 0 will be allotted to form a gradient binary pattern.
Step 5: The binary code is converted to decimal by multiplying 2i .
M −1
LGP = ∑i=0 s(gri − gravg )2i (5.2)
where
⎧ 0 gri < gravg
⎪
s(x ) = ⎨ (5.3)
⎩1 gri ⩾ gravg
⎪
5.2.2.2 Local neighbor gradient pattern (LNGP)

LNGP provides efficient discrimination by eliminating variation in the local and
global patterns which are a drawback of the prevailing pattern techniques by
considering the successive gradient patterns of the neighborhood sample values [16].
The histogram based features are extracted for the discrimination of the neuro-
degenerative and healthy classes. The operation of LNGP is illustrated in figure 5.4.
LNGP is computed by the following steps:
Step 1: The center values (MC ) and neighboring sample values (Mi ) are fixed.
Step 2: The gradient values are calculated for each sample by equation (5.4)
gri = ∣Mi − Mc ∣ (5.4)
Step 3: The comparison of successive gradient values of neighbor point (gri ) and
the next neighbor point (gri +1) is performed including center point. If the
gradient value gri is greater than gri +1, 1 is assigned, otherwise 0 is assigned.
Step 4: The binary code is then converted in to a decimal form my multiplying 2i
di = gri − gri +1 (5.5)
M −1
LNGP = ∑i=0 Sdi 2i (5.6)
5-5
Figure 5.4. The illustration of LNGP technique.
where
⎧ 0 di < 0
s(x ) = ⎨ (5.7)
⎩1 di ⩾ 0
The histogram based features are extracted from the patterns created by LGP and
LNGP. The probability of occurrence of a sample point is called a histogram. It
computes the features easily by using the standard descriptors. The length of the
feature depends on the length of neighboring points (N).
In this study a total of eight neighboring sample points are considered. The length
of the features (LF ) of LGP and LNGP is given as follows:
LF = 2N (5.8)
Hence 256 histogram features are extracted using LGP and LNGP. The
gait recordings are obtained from the two force sensors with the 90 000 samples
(2 × 90 000). For each sensor recording 256 features are taken. So, in total 512
features are extracted from Parkinson’s, Huntington’s, ALS, and the healthy
control for the discrimination.
5-6
Figure 5.5. Artificial neural network classifier.
5.2.3 Artificial neural network (ANN)

In biomedical signal processing, classification is a crucial mission to diagnose
subjects with the specific disorder. In the present study an ANN classifier [19],
which follows the parallel computing mechanisms by the interconnected neurons,
is used. Neurons are connected with each other with the associated weights and
learning rules [20]. The layers in the ANN are categories to input, hidden, output
layer [21]. The 512 histogram based features are given to the input layer. A single
hidden layer with 50 hidden neurons is considered for the experimentation. The
ANN classifier is trained by the conjugate gradient back propagation (traincgb)
function. The functional architecture of the ANN classifier is shown in figure 5.5.
ANN classification is performed in the following steps:
(i) Initialization of inputs, weights, target and learning rate. The histogram
features are taken as inputs.
(ii) Calculate the net value which is the summation of inputs and weights.
(iii) Compute the output using sigmoid activation function.
(iv) Stop the execution if the output value is the same as the target value.
(v) If it is not, adjust the weight then perform steps ii to iv.
(vi) The execution is continued until the output value is equal to the target.
The classification performance is calculated for individual neurodegenerative

diseases based on the classification accuracy, sensitivity, specificity, positive pre-
dictive value and negative predictive value which is given below:
TP + TN
Accuracy = * 100 (5.9)
TP + TN + FP + FN
TP
Sensitivity = * 100 (5.10)
TP + FN
5-7
TN
Specificity = * 100 (5.11)
TN + FP
TP
Positive predictive value = * 100 (5.12)
TP + FP
TN
Negative predictive value = * 100 (5.13)
TN + FN
In the above equation TP corresponds to true positive, TN is defined as true
negative, FP is referred to as false positive and FN represents false negative.
5.3 Results and discussion

We have used the time series data which are publically available in the Physionet.
The recordings from the 15 Parkinson’s, 20 Huntington’s, 13 ALS and 16 healthy
control subjects are taken for the experimentation. The time series gait signal is
derived from the stride, swing, and stance intervals. The histogram based features
are extracted from the LGP and LNGP patterns by means of neighborhood and
center sample values.
Two force sensors are placed under the feet of the patients. Each recording
obtained from the force sensor has samples of 90 000. A total of 512 features are
extracted and provided for classification. The 8 bit binary pattern of LGP and
LNGP ranges from 00000000 to 11111111, that is, H(0) to H(255). The histogram
features of LGP and LNGP are shown in figures 5.6 and 5.7, respectively. The
4 4
10 LGP features for Parkinson 10 LGP features for Huntington
3.5 3
3 2.5
2.5
2
2
1.5
1.5
1
1
0.5 0.5
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
4 4
10 LGP features for ALS 10 LGP features for Healthy
3.5 3
3 2.5
2.5
2
2
1.5
1.5
1
1
0.5 0.5
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Figure 5.6. Extracted histogram features by LGP.
5-8
4 4
10 LGP features for Parkinson 10 LGP features for Huntington
6 5
5
4
4
3
3
2
2
1
1
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
4 4
10 LGP features for ALS 10 LGP features for Healthy
6 5
5
4
4
3
3
2
2
1
1
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Figure 5.7. Extracted histogram features by LNGP.
Table 5.1. Classification results of the proposed method.
Parkinson’s Huntington’s Neurodegenerative

Performance Proposed versus versus ALS versus disease versus
metrics method healthy (%) healthy (%) healthy (%) healthy (%)
Accuracy LGP 98.67 99.01 99.31 99.37

LNGP 99.87 99.54 100 99.68
Sensitivity LGP 99 99.37 99.37 98.82
LNGP 99.41 100 100 99.37
Specificity LGP 98.37 99.5 99.23 99.58
LNGP 100 99.09 100 99.79
Positive LGP 98.37 99.37 99.37 98.75
predictive LNGP 100 98.75 100 99.37
value
Negative LGP 99 99.5 99.23 99.58
predictive LNGP 99.33 100 100 99.79
value
classification of each group of patients and the healthy control is evaluated. Ten-fold
cross validation is performed to train and test the neural network.
The diagnosis of neurodegenerative disease is performed using the discriminate
features based on histogram with ten-fold cross validation. The evaluated results for
the classification of Parkinson’s neurological healthy subject gait signal are shown in
table 5.1. The proposed method achieved a classification accuracy of 98.67% with
LGP and 99.87% with LNGP for the classification of Parkinson’s versus healthy.
5-9
For the classification of healthy versus Huntington’s the obtained accuracy is

99.01% and 99.44% using LGP and LNGP, respectively.
Various methods have been proposed for the classification of neurodegenerative
diseases using gait signal. The proposed method compared with the existing methods
Table 5.2. Performance comparison of the proposed pattern techniques with existing techniques.
Feature extraction Accuracy

method Features Classifier Classification task (%)
Statistical analysis [5] Minimum, maximum, SVM Parkinson’s versus 89.33

average, standard healthy
deviation Huntington’s versus 90.28
healthy
ALS versus healthy 96.79
Neurodegenerative 90.63
disease versus
healthy
Auto correlation [6] Mean, standard Decision Parkinson’s versus 92.3
deviation tree healthy
Huntington’s versus 88.5
healthy
ALS versus healthy 96.2
disease versus
healthy
Parzan window power Stride patterns LS-SVM ALS versus healthy 93.1
density function [7]
Wavelet Maximum, minimum, SVM Parkinson’s versus 90.32
decomposition [8] mean, variance, energy healthy
Signal turns count [12] Swing interval count and LS-SVM ALS versus healthy 89.66
average stride interval
Proposed method Histogram ANN Parkinson’s versus
LGP healthy—
LNGP LGP 98.67
LNGP 99.87
Huntington’s versus 99.01
healthy—
LGP
LNGP 99.44
ALS versus healthy— 99.31
LGP
LNGP 100
disease versus
healthy—
LGP
LNGP 99.68
5-10
to diagnose neurological disease is presented in table 5.2. The proposed method

exhibits that it attains the maximum accuracy compared to already existing
techniques and also it provides the best discriminate features. The advantage of
the proposed method is its simplicity and reliability.
5.4 Conclusion
In the present study, two feature extraction methods are proposed to classify
neurodegenerative subjects and healthy subjects. The proposed approaches, namely
LGP and LNGP, are performed on the gait signal which converts the signals into a
specific pattern. The histogram based features with high discriminated power are
extracted from the patterns. Those features are classified using ANN classifier for
the classification of four different cases. The classification results show that the
LNGP method is superior to the LGP approach. The proposed method, LGP and
ANN, attains a maximum classification accuracy of 99.37% for the classification
between healthy and a combination of the three diseases, and LNGP with ANN
achieved a classification accuracy of 100% for the classification between ALS and
healthy subjects. In the future, different pattern techniques will be used for the
automated diagnosis of neurodegenerative disease.
References
[1] Djurić-Jovičić M D, Jovičić N S, Radovanović S M, Stanković I D, Popović M B and Kostić
V S 2014 Automatic identification and classification of freezing of gait episodes in
Parkinson’s disease patients IEEE Trans. Neural Syst. Rehabil. Eng. 22 685–94
[2] Selzler R, Green J R and Goubran R 2018 Neurodegenerative disease prediction based on
gait analysis signals acquired with force-sensitive resistors IEEE Life Sciences Conf. (LSC)
pp 122–5
[3] Vipani R, Hore S, Basak S and Dutta S 2017 Gait signal classification tool utilizing Hilbert
transform based feature extraction and logistic regression based classification 2017 Third Int.
Conf. on Research in Computational Intelligence and Communication Networks (ICRCICN)
pp 57–61
[4] Dutta S, Ghosh D and Chatterjee S 2018 Multifractal detrended cross correlation analysis of
neuro-degenerative diseases—an in depth study Physica A: Stat. Mech. Appl. 491 188–98
[5] Daliri M R 2012 Automatic diagnosis of neuro-degenerative diseases using gait dynamics
Measurement 45 1729–34
[6] Gupta K, Khajuria A, Chatterjee N, Joshi P and Joshi D 2018 Rule based classification of
neurodegenerative diseases using data driven gait features Health Technol. 1–14
[7] Wu Y and Shi L 2011 Analysis of altered gait cycle duration in amyotrophic lateral sclerosis
based on nonparametric probability density function estimation Med. Eng. Phys. 33 347–55
[8] Joshi D, Khajuria A and Joshi P 2017 An automatic non-invasive method for Parkinson’s
disease classification Comput. Methods Programs Biomed. 145 135–45
[9] Khajuria A, Joshi P and Joshi D 2018 Comprehensive statistical analysis of the gait
parameters in neurodegenerative diseases Neurophysiology 50 38–51
[10] Prabhu P, Karunakar A K, Anitha H and Pradhan N 2018 Classification of gait signals into
different neurodegenerative diseases using statistical analysis and recurrence quantification
analysis Pattern Recognit. Lett.
5-11
[11] Ren P, Zhao W, Zhao Z, Bringas-Vega M L, Valdes-Sosa P A and Kendrick K M 2016

Analysis of gait rhythm fluctuations for neurodegenerative diseases by phase synchronization
and conditional entropy IEEE Trans. Neural Syst. Rehabil. Eng. 24 291–9
[12] Wu Y and Krishnan S 2009 Computer-aided analysis of gait rhythm fluctuations in
amyotrophic lateral sclerosis Med. Biol. Eng. Comput. 47 1165
[13] Xia Y, Gao Q and Ye Q 2015 Classification of gait rhythm signals between patients with
neuro-degenerative diseases and normal subjects: Experiments with statistical features and
different classification models Biomed. Signal Process. Control 18 254–62
[14] Zeng W and Wang C 2015 Classification of neurodegenerative diseases using gait dynamics
via deterministic learning Inf. Sci. 317 246–58
[15] Sairamya N J, George S T, Ponraj D N and Subathra M S P 2018 Detection of epileptic
EEG signal using improved local pattern transformation methods Circ. Syst. Signal. Process.
37 5554–75
[16] Sairamya N J, George S T, Balakrishnan R and Subathra M S P 2018 An effective approach
to classify epileptic EEG signal using local neighbor gradient pattern transformation
methods Australas. Phys. Eng. Sci. Med. 41 1029–46
[17] Hausdorff J M, Cudkowicz M E, Firtion R, Wei J Y and Goldberger A L 1998 Gait
variability and basal ganglia disorders: stride‐to‐stride variations of gait cycle timing in
Parkinson’s disease and Huntington’s disease Mov. Disord. 13 428–37
[18] Jaiswal A K and Banka H 2017 Local pattern transformation based feature extraction
techniques for classification of epileptic EEG signals Biomed. Signal Process. Control 34 81–92
[19] Singla R, Chambayil B, Khosla A and Santosh J 2011 Comparison of SVM and ANN for
classification of eye events in EEG J. Biomed. Sci. Eng. 4 62
[20] Barkana B D, Saricicek I and Yildirim B 2017 Performance analysis of descriptive statistical
features in retinal vessel segmentation via fuzzy logic, ANN, SVM, and classifier fusion
Knowl.-Based Syst. 118 165–76
[21] Sharma B and Venugopalan K 2014 Comparison of neural network training functions for
hematoma classification in brain CT images IOSR J. Comput. Eng. 16 31–5
5-12
IOP Publishing

Chapter 6
Rett syndrome
Ola H Skjeldal
Rett syndrome is a severe neurodevelopmental disorder, which mostly affects females.

Usually it starts very early when the child is between 6 and 18 months. The main clinical
symptoms and signs are impairment in cognition, motor control and communication.
Epilepsy occurs in 80%–90% of the cases. This chapter gives an overview of the clinical
features, the epidemiology and the genetic background of this condition.
6.1 Background
Rett syndrome is a severe neurodevelopmental disorder, predominantly affecting
females (Hagberg et al 1983, Kerr and Engerström 2001, Rett 1966, Smeets et al
2011). The main clinical signs are impairment in cognition, motor control and
communication skills. However, there are a substantial number of other clinical
features, which make this condition a unique clinical entity.
Rett syndrome was first described in the medical literature in 1966 by the
Austrian neuropediatrician Andreas Rett. However, anecdotal descriptions of
persons with Rett syndrome can be found in the literature many years before Rett
published his original article. If we go back to the thirties, the Norwegian author
Sigrid Undseth wrote a book about her daughter, who had a severe intellectual
disability. Based on the descriptions given in this book, it is likely that this girl
suffered from Rett syndrome. The Swedish author and journalist Greta Bolin
unknowingly gave the first written and detailed description of Rett syndrome. In
1956 she published a book with the title ‘Mitt barn är annorlunda’ (Bolin 1956). The
book was written to create knowledge, insight and hope for parents of disabled
children, and Greta Bolin gives an insightful description of her daughter’s life. Lotta
was at that time nine year old, and was severely neurologically handicapped. It is
worth noting some of the descriptions Bolin gives of Lotta:
• She is a sweet little girl with the lank body of a 9 year-old girl, who now and
then lights up in a smile or laughter.
• Her hands are held wringed together, and she puts them in her mouth.

• She is grinding her teeth constantly.

• She takes no interest in toys, although she may laugh when pictures in a book
are shown to her.
• The only time she shows a spontaneous interest is when music is played.
• Her gait is staggering and broadbased.
Many years later, in 1986, Lotta was diagnosed having Rett syndrome by Bengt
Hagberg. However, it was Andreas Rett who first recognized and systematically
described the clinical features of Rett syndrome, which many years later came to
bear his name. Rett was a neuropediatrician, working in Vienna, and as with many
of the clinicians working at that time, he was extremely systematic and thorough in
his work. What originally formed his description was that on one and the same day
two female patients who had almost identical clinical symptoms and signs consulted
him. They were seated next to each other on the lap of their mothers, and what
struck Andreas Rett was that they more or less displayed the same behavior.
Especially he noted that the stereotypic hand movements were different from what
could be seen among other disabled children. Then he remembered other girls he had
seen with similar features. All of these girls had a developmental stagnation and then
regression, gait apraxia and stereotypical hand movements. He published his
findings in a relatively unknown Austrian journal (‘Wiener Medizinische
Wochenschrift’) and called the syndrome ‘Über ein eigenartiges hirnatrophisches
syndrome bei hyperamonaemie im kindesalter’ which was a rather complicated
name (Rett 1966). Because the article was published in the German language in a
relatively unknown journal, few people read it. Thus, the syndrome was unknown
for many years.
At the same time, Bengt Hagberg, a Swedish neuropediatrician in Uppsala, was
recognizing the same constellation of clinical characteristics in females from Sweden.
He early defined this condition as a separate clinical entity and he named it ‘Vesslan
disease’ because many of the girls were found in the area of Vesslan in Sweden. This
resulted in the first widely read English language paper on Rett syndrome (Hagberg
et al 1983). In this paper Hagberg published the clinical symptoms and signs of
35 females and suddenly neuropediatricians and geneticists in the Western world
were alerted to the existence of this new and unique neurodevelopmental disorder. In
Norway the pediatric neurologists Randi Kiil and Ruth Bostad (Bostad and Kiil
1987) published a study of Norwegian girls with Rett syndrome.
In the forthcoming years a number of research groups were established and
marshaling a series of research studies. Since 1966, when Andreas Rett for the first
time described Rett syndrome, the number of scientific publications has increased
enormously (see figure 6.1).
Today we know that Rett syndrome is more common and has a broader
variability in clinical phenotype than was originally thought. The main gene
involved in this condition, MeCP2, was discovered in 1999 (Amir et al 1999).
However, for an accurate diagnosis we utilize strictly defined clusters of
6-2
800
700
600
500
400
300
200
100
0
1966-1970 1971-1975 1976-1980 1981-1985 1986-1990 1991-1995 1996-2000 2001-2005 2006-2010 2011-2013 2014-2017
Figure 6.1. The number of Rett syndrome publications in the last 50 years.
characteristic deviant features and profiles in the development of the child. These
have been suggested in a number of diagnostic criteria published.
6.2 Clinical features

Rett syndrome is a genetic neurodevelopmental disorder which primarily affects
females (Smeets et al 2011). It is one of the leading genetic causes of severe
intellectual disabilities with a prevalence rate ranging from 1/9000 to 1/15 000
(Bienvenu et al 2006, Skjeldal et al 1997). In the typical classical form the patients
have a normal postnatal development followed by a period of regression resulting in
reduction in head growth, loss of acquired motor and language skills, gait
abnormalities, cognitive impairments and autistic behaviour and after a while,
stereotypical hand movements (hand-washing movements) (Glaze et al 2010). After
some years the development more or less reaches a plateau or shows quite slow
improvement (Glaze et al 2010, Hagberg et al 2002). In this stage a variety of Rett
syndrome specific symptoms show up, for instance orthopedic complications
(scoliosis), seizures, teeth grinding and symptoms reflecting disturbances in the
autonomic nervous system. The syndrome is usually associated with comorbidities
such as reduced somatic growth, gastro-intestinal problems, osteopenia and auto-
nomic dysfunction.
Hagberg et al (1986) described the clinical development in four stages:
1. The early-onset stagnation.
2. The rapid developmental regression.
3. The pseudo-stationary stage.
4. The later stage with motor deterioration.
Chahrour and Zoghbi (2007) have visualized the developmental process of these
four stages, as shown in figure 6.2.
However, there is a wide variability in the rate of progression, age at onset,
clinical profiles, seizures and communication abilities. Thus, in addition to the
classical form of Rett syndrome, accounting for about three quarters of the cases,
there are a number of established recognizable atypical variants. These atypical
variants include congenital Rett syndrome (Ariani et al 2008, Rolando 1985), early
6-3
Figure 6.2. The four developmental stages of Rett syndrome (after Chahrour and Zoghbi 2007, Hagberg et al
1986).
seizure variant (Hanefeld 1985), a preserved speech variant (Zappella 1992), and
also the so-called male variant (Coleman 1990).
6.2.1 Congenital Rett syndrome

The congenital variant was initially described by Rolando in 1985 and followed up
by Ariani and colleagues in 2008. In this form of Rett syndrome, the girls are floppy
and show signs of delayed development from the first months of life. The most
typical clinical findings are muscular hypotonia, unresponsiveness, and irritability in
the neonatal period. There is a delay in eye contact, delayed head control, crying and
feeding difficulties. Motor development is usually severely impaired, and hand use is
almost absent. The diagnosis of Rett syndrome is usually late compared with those
girls having the classical form.
6.2.2 Early onset seizure variant (Hanefeld variant)

This phenotype was first described by Folker Hanefeld in 1985, hence the name
Hanefeld variant. As the name indicates, these girls develop seizures very early in
life, usually before three months of age. Some of the patients also suffer from
infantile spasms, which is a malignant form of epilepsy. Because of the early start of
the seizures, the diagnosis can for a long time be epileptic encephalopathy. However,
6-4
when the more typical Rett syndrome features appear, the diagnosis is usually soon
verified.
6.2.3 Preserved speech variants

The preserved speech variant, also called the Zappella variant, includes more mildly
affected females in whom some speech and also motor skills are preserved. This
variant has different subgroups, which have their own clinical characteristics
(Zappella et al 2001). However, the course of the disorder is in stages as in classical
Rett syndrome, but the clinical symptoms and signs seems to be milder.
6.2.4 Male variants

Most of the patients with Rett syndrome are females. However, during the years
male cases with Rett syndrome have been described. All are based on clinical
criteria. Mary Coleman reported already in 1990 boys who fulfilled the then
diagnostic criteria of Rett syndrome (Coleman 1990). In the following years, a
substantial number of males have been reported with typical Rett syndrome clinic
and also with mutations in MeCP2. Clinical features most commonly include the
impairment of motor abilities, regression of language and development of stereo-
typical hand movements, breathing abnormalities, seizures and scoliosis. In 2011,
Kristine Ravn and colleagues reported two families with both brothers and sisters
having Rett syndrome, and MeCP2 frameshift mutations (Ravn et al 2011).
6.3 Important clinical symptoms and signs in Rett syndrome

6.3.1 Seizures
Epilepsy is very common in Rett syndrome and occurs in over 80% of cases. Usually
it starts in the first years of life, and in some patients it can be very difficult to treat.
About 20%–30% of Rett syndrome patients have the so-called drug-resistant
epilepsy (Bao et al 2013). Very often the epilepsy correlates with the clinical severity
and also the type of MeCP2 mutation (Bao et al 2013, Glaze et al 2010, Jian et al
2007, Pintaudi et al 2010). Seizures seem to be more frequent in patients with large
deletions in the gene (Glaze et al 2010).
The age at onset of the seizures is usually about 4 years (Pintaudi et al 2010).
However, the variation is considerable. All types of seizures occur, but the most
common reported include complex partial, generalized tonic–clonic and myoclonic
seizures (Pintaudi et al 2010). Frequently used anti-epileptic drugs are broad-
spectrum drugs, such as valproate, lamotrigine, carbamazepine (Bao et al 2013).
Vagal nerve stimulator and ketogenic diet are also used (Glaze et al 2010, Wilfong
and Schultz 2006). An important point is that paroxysmal non-epileptic manifes-
tations, such as breath holding, hyperventilation, and staring can be misdiagnosed as
seizures (Jian et al 2007). Usually, prolonged video-EEG monitoring is done to
differentiate non-epileptic events from epileptic seizures (d’Orsi et al 2012). It is
common that Rett syndrome females have discharges detected on EEG. If clinical
seizures are absent, anti-epileptic medication should usually not be initiated.
6-5
6.3.2 Respiratory and cardial disturbances

Respiratory disturbances with episodic hyperventilation, hypoventilation and
apnoea are very common in girls with Rett syndrome. During sleep many girls
and women with Rett syndrome usually have a bradycardia (slow heart rhythm).
When awake, tachycardia (fast hearth rhythm) and bradycardia alternate (Rohdin
et al 2007). Julu et al (2008) categorized the cardiac and respiratory disturbances in
three groups. The first group they named forceful breathers with a low partial
pressure of CO2. The second group was mentioned as the feeble breathers with a
constant high pressure of CO2. The third was named the apneustic breath with
severe apnea tendency and accumulation of CO2. The biological background for
these respiratory patterns is most probably a brain stem dysfunction.
6.3.3 Sleep
Disturbances in the sleep pattern are common in Rett syndrome, at least in parts of
life. It is claimed that it occurs in over 80% of the cases. Roughly the sleep
disturbances are characterized by increased daytime sleep and lack of sleep in the
night (Young et al 2007). Furthermore, there are disturbances in the sleep and wake
patterns, and very often, especially in the adult population, episodic nighttime
screaming and laughing.
6.3.4 Gastrointestinal disturbances

Gastrointestinal disturbances are very common problems in Rett syndrome.
Prolonged feeding time and reflux occurs in most of the females with Rett syndrome
as well as constipation, dysphagia, chewing and swallowing difficulties.
Malnutrition, in spite of normal dietary intake occurs (Motil et al 2012) and thus
gastrostomy can be necessary in some girls for the purpose of ensuring optimal
nutrition.
6.3.5 Communication
The wide spectrum of Rett syndrome also includes girls with considerably higher
cognitive abilities than originally thought (Zappella et al 1998, 2003). There are
several reports in the literature of females with Rett syndrome who are capable of
intentional communication and verbal language skills. These atypical Rett syn-
drome cases can vary from girls who display characteristics of the syndrome from
birth, to girls who have far milder characteristics, and maintain oral speech. The
group that has some verbal skills is known as the preserved speech variant.
However, both females and boys with classical Rett syndrome are considered to
be severely affected in the areas of cognitive and language ability (Percy 2016, Pini
et al 2016). They usually do not develop the ability to communicate intentionally or
to interact socially with other people. Yet parents and teachers have reported that
the girls can use some communication skills. There are reports by parents that tell
about girls who use limited verbal speech, especial in situations where they
experience stress. Others can use gestures or eye gaze to convey meaning. Some
6-6
girls can recognize pictures or other symbols, and some enjoy stories (Koppenhaver
et al 2001).
In spite of this it is important to underline that so far no one has been able to find
strategies resulting in intentional communicative behaviors in females with Rett
syndrome. Communication interventions contribute to a responsive and
predictable environment for the females and to the establishment of a set of strategies
that may be used by parents and others in their environment (von Tetzchner 1997).
6.4 Genetic basis of Rett syndrome

In 1999 a major discovery occurred when the gene involved in causing Rett
syndrome was identified (Amir et al 1999). Rett syndrome was found to be
associated with mutations in the methyl-CpG binding protein-2 (MeCP2 gene,
located at the X chromosome; Xq28). By far the most the mutations are sporadic
and de novo, and the majority are of paternal origin. Thus, Rett syndrome is very
rarely inherited, which was claimed even before 1999. Clinicians observed that it was
most often only one member in a family with the syndrome. Today, one assumes
that less than 0.5% of the cases are inherited.
The genomic locus of the MeCP2 gene is approximately 80 kb and consists of
4 exons, from which two different isoforms of the protein (MeCP2) are transcribed.
The protein which is coded by exon 2, 3 and 4 (MeCP2) is mostly expressed in the
brain and also in the peripheral tissue (Dragich et al 2007).
Furthermore, MeCP2 mutations are not found in all Rett cases. In some of the
typical Rett cases without MeCP2 mutations, defects in other genes have been found
(Hara et al 2015, Lee et al 2016). Interestingly, mutations in the MeCP2 gene have
also been reported to be associated with a wide range of different clinical disorders in
both males and females. Thus, it is important to emphasize that Rett syndrome
remains a clinical rather than a molecular diagnosis (Glaze et al 2010).
During recent years it has become clear that persons with MeCP2 mutations
present with a much broader phenotype than originally thought. Examples of these
are neonatal encephalopathy, learning disabilities without any kind of Rett
syndrome similarities, autism, Angelman like phenotypes and variant forms of
Rett syndrome (Sanmann et al 2012). Again, it is important to underline that the
diagnosis of Rett syndrome is based on clinical observations and rests on the clinical
criteria (see table 6.1).
6.5 Conclusions
Although our knowledge of Rett syndrome has grown considerably in recent years,
many questions and problems remain. Still we don’t know the exact function of the
MeCP2 gene, apart from its action as a transcription repressor. Furthermore, we
don’t know all the genes or proteins in the central nervous system which are
influenced by mutations in MeCP2.
At least 2 out of the 4 main criteria and 5 out of 11 supportive criteria
1. Partial or complete loss of acquired purposeful hand skills.
2. Partial or complete loss of acquired spoken language.
6-7
Table 6.1. Revised diagnostic criteria for Rett syndrome (after Glaze et al 2010).
1. Consider diagnosis when postnatal deceleration of head growth is observed

A. Required for typical or classic Rett syndrome
(1) A period of regression followed by recovery or stabilization.
(2) All main criteria and all exclusion criteria.
(3) Supportive criteria are not required, although often present in typical Rett.
B. Required for atypical variant of Rett

(1) A period of regression followed by recovery or stabilization.
(2) Main, exclusion and supportive criteria.
(A) Main criteria 1 2.
(B) Exclusion criteria for typical Rett syndrome:
(1) Brain injury secondary to trauma (peri- or postnatally), neurometabolic disease, or severe
infection that causes neurological problems.
(2) Grossly abnormal psychomotor development in first 6 months of life.
(C) Supportive criteria for atypical Rett syndrome diagnosis:

(1) Breathing disturbances when awake.
(2) Bruxism when awake.
(3) Impaired sleep pattern.
(4) Abnormal muscle tone.
(5) Peripheral vasomotor disturbances.
(6) Scoliosis/kyphosis.
(7) Growth retardation.
(8) Small cold hands and feet.
(9) Inappropriate laughing/screaming spells.
(10) Diminished response to pain.
(1 1) Intense eye communication (staring).
3. Gait abnormalities: Impaired (dyspraxic) or absence of ability.

4. Stereotypic hand movements such as hand wringing/squeezing, clapping/
tapping, mouthing and washing/rubbing automatisms.
Recently, as pointed out earlier, mutations in other genes have been reported to give
a Rett syndrome phenotype. It is still unclear if these reported mutations are primary
or secondary to an unknown effect of MeCP2. Further research will hopefully lead
to more knowledge of the interplay of MeCP2, its protein MeCP2 and the target
genes and proteins.
Clinically, there are also a lot of questions to be solved. The fluctuating clinical
course, the phenotypic variability and the genotype–phenotype correlation are still
6-8
not completely understood. We do not yet know enough about how Rett syndrome
develops in the different life stages. Disabled people live longer today. This also
applies to persons with Rett syndrome. Females with Rett syndrome older than
70 years are no longer a rarity and we do not know enough about their clinical
situation and about their needs in that age group. Furthermore, new treatment
strategies have been introduced, so far without any success. However, sooner or later
we will also get a breakthrough in this area.
References
Amir R E, Van den Veyver I B, Wan M, Tran C Q, Francke U and Zoghbi H Y 1999 Rett
syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-binding
protein 2 Nat. Genet. 23 185–8
Ariani F et al 2008 FOXG1 is responsible for the congenital variant of Rett syndrome The Am. J.
Hum. Genet. 83 89–93
Bao X, Downs J, Wong K, Williams S and Leonard H 2013 Using a large international sample to
investigate epilepsy in Rett syndrome Dev. Med. Child Neurol. 55 553–8
Bienvenu T et al 2006 The incidence of Rett syndrome in France Pediatr. Neurol. 34 372–5
Bolin G 1956 Mitt barn er annorlunda (Stockholm, Sweden: Rabén and Sjögren)
Bostad R and Kiil R 1987 Rett syndrome (Et nytt sykdomsbilde) Tidsskr. Nor. Lægeforen. 107
1659–61
Chahrour M and Zoghbi H Y 2007 The story of Rett syndrome: from clinic to neurobiology
Neuron 56 422–37
Coleman M 1990 Is classical Rett syndrome ever present in males? Brain Dev. 12 31–2
d’Orsi G et al 2012 Epileptic seizures, movement disorders, and breathing disturbances in Rett
syndrome: diagnostic relevance of video-polygraphy Epilepsy Behav. 25 401–7
Dragich J M, Kim Y H, Arnold A P and Schanen N C 2007 Differential distribution of the
MeCP2 splice variants in the postnatal mouse brain J. Comp. Neurol. 501 526–42
Glaze D G et al 2010 Epilepsy and the natural history of Rett syndrome Neurology 74 909–12
Hagberg B, Aicardi J, Dias K and Ramos O 1983 A progressive syndrome of autism, dementia,
ataxia, and loss of purposeful hand use in girls: Rett’s syndrome: report of 35 cases Ann.
Neurol. 144 471–9
Hagberg B, Hanefeld F, Percy A and Skjeldal O 2002 An update on clinically applicable
diagnostic criteria in Rett syndrome: comments to Rett syndrome clinical criteria consensus
panel satellite to European Paediatric Neurology Society Meeting Baden Baden, Germany,
11 September 2001 Eur. J. Paediatr. Neurol. 6 293–7
Hagberg B, Witt-Engerström I, Opitz J M and Reynolds J F 1986 Rett syndrome: a suggested
staging system for describing impairment profile with increasing age towards adolescence
Am. J. Med. Genet. Part A 25 47–59
Hanefeld F 1985 The clinical pattern of the Rett syndrome Brain Dev. 7 320–5
Hara M, Ohba C, Yamashita Y, Saitsu H, Matsumoto N and Matsuishi T 2015 De novo
SHANK3 mutation causes Rett syndrome‐like phenotype in a female patient Am. J. Med.
Genet. Part A 167 1593–6
Jian L, Nagarajan L, de Klerk N, Ravine D, Christodoulou J and Leonard H 2007 Seizures
in Rett syndrome: an overview from a one-year calendar study Eur. J. Paediatr. Neurol. 11
310–7
Julu P O et al 2008 Cardiorespiratory challenges in Rett’s syndrome Lancet 371 1981–3
6-9
Kerr A M and Engerström I W 2001 The clinical background to the Rett disorder Rett Disorder
and the Developing Brain ed A M Kerr and W I Engerström (Oxford, UK: Oxford University
Press), pp 1–26
Koppenhaver D A, Erickson K A, Harris B, McLellan J, Skotko B G and Newton R A 2001
Storybook-based communication intervention for girls with Rett syndrome and their mothers
Disabil. Rehabil. 23 149–59
Lee J S, Yoo Y, Lim B C, Kim K J, Choi M and Chae J H 2016 SATB2-associated syndrome
presenting with Rett-like phenotypes Clin. Genet. 89 728–32
Motil K J et al 2012 Gastrointestinal and nutritional problems occur frequently throughout life in
girls and women with Rett syndrome J. Pediatr. Gastroenterol. Nutr. 55 292–8
Percy A K 2016 Progress in Rett Syndrome: from discovery to clinical trials Wien. Med.
Wochenschr. 166 325–32
Pini G et al 2016 Rett syndrome: a wide clinical and autonomic picture Orphanet J. Rare Dis. 11
132–48
Pintaudi M et al 2010 Epilepsy in Rett syndrome: clinical and genetic features Epilepsy Behav. 19
296–300
Ravn K, Roende G, Duno M, Fuglsang K, Eiklid K L, Tümer Z, Nielsen J B and Skjeldal O H
2011 Two new Rett syndrome families and review of the literature: expanding the knowledge
of MECP2 frameshift mutations Orphanet J. Rare Dis. 6 58
Rett A 1966 Uber ein eigarties hinartrophisches Syndrom bei Hyperammoniamie in Kindesalter
Wien. Med. Wochenschr. 116 723
Rohdin M, Fernell E, Eriksson M, Albåge M, Lagercrantz H and Katz-Salamon M 2007
Disturbances in cardiorespiratory function during day and night in Rett syndrome Pediatr.
Neurol. 37 338–44
Rolando S 1985 Rett syndrome: report of eight cases Brain Dev. 7 290–6
Sanmann J N, Schaefer G B, Buehle B A and Sanger W G 2012 Algorithmic approach for methyl-
CpG binding protein 2 (MECP2) Gene testing in patients with neurodevelopmental
disabilities J. Child Neurol. 27 346–54
Skjeldal O H, von Tetzchner S, Aspelund F, Herder G A and Lofterød B 1997 Rett syndrome:
geographic variation in prevalence in Norway Brain Dev. 19 258–61
Smeets E E J, Pelc K and Dan B 2011 Rett syndrome Mol. Syndromol. 2 113–27
von Tetzchner S 1997 Communication skills among females with Rett syndrome Eur. Child
Adolesc. Psychiatry 6 33–7
Wilfong A A and Schultz R J 2006 Vagus nerve stimulation for treatment of epilepsy in Rett
syndrome Dev. Med. Child Neurol. 48 683–6
Young D, Nagarajan L, de Klerk N, Jacoby P, Ellaway C and Leonard H 2007 Sleep problems in
Rett syndrome Brain Dev. 29 609–16
Zappella M 1992 The Rett girls with preserved speech Brain Dev. 14 98–101
Zappella M, Gillberg C and Ehlers S 1998 The preserved speech variant: a subgroup of the Rett
complex: a clinical report of 30 cases J. Autism Dev. Disord. 28 519–26
Zappella M et al 2003 Study of MECP2 gene in Rett syndrome variants and autistic girls Am. J.
Med. Genet. Part B: Neuropsychiatr. Genet. 119 102–7
Zappella M, Meloni I, Longo I, Hayek G and Renieri A 2001 Preserved speech variants of the
Rett syndrome: molecular and clinical analysis Am. J. Med. Genet. Part A 104 14–22
6-10
IOP Publishing

Chapter 7
Knowledge about epilepsy in university
health students
Natalia Cristina de Oliveira Vargas e Silva and Priscila de Souza
Epilepsy affects people of all ages and usually does not have an identifiable cause. It
is a very prevalent neurologic disease that affects more than 50 million people
worldwide. About three-quarters of people living with epilepsy in low and middle-
income countries do not receive proper treatment. Unavailability of antiepileptic
drugs and the lack of trained health professionals to recognize, diagnose and treat
epilepsy contribute to this scenario. The lack of knowledge is considered a major
determinant of social rejection against people with epilepsy. Stigma related to this
pathology affects people in different ways and can influence social relationships,
schooling opportunities, employment and emotional health of the person with
epilepsy. Several studies have identified a low level of knowledge about the disease
among family members, caregivers and university students. This chapter discusses
knowledge about epilepsy among health students. It is unanimous in all available
research in this field, that important issues related to the care of the epileptic patient
appear not to have been adequately addressed during university years. Educational
activities may contribute to reduce gaps in knowledge, but the routine inclusion of
discussions on this subject throughout university education will result in the
students’ familiarity with the disease and consequently will have repercussions on
their future professional attitudes. Effectively prepared professionals will certainly
contribute to increase treatment adherence and personal and professional oppor-
tunities for persons with epilepsy.
7.1 Introduction
Epilepsy is defined as an excessive disorganized electrical discharge in the brain and
can be divided into three groups of seizures: partial or focal, generalized seizures
and non-classifiable ones [1]. The diagnosis of epilepsy usually depends on the

description of the seizures by the patient, relatives or witnesses. Thus, history and
clinical examination are essential for early diagnosis.
The World Health Organization (WHO) estimates that 8 out of every 1000 people
worldwide have epilepsy, and in developing countries this prevalence is higher than
in developed nations [2]. Although epilepsy is one of the most prevalent neurological
pathologies, it is still surrounded by prejudice, stigma and lack of awareness towards
the disease [3–5].
Epilepsy affects people of all ages and generally does not have an identifiable
cause. According to the WHO [6], over 50 million people have epilepsy worldwide,
and almost 80% of them live in low and middle-income countries. About 2.4 million
people are diagnosed with epilepsy each year.
In Asia, a continent constituted of heterogeneous countries and limited resources,
there is a great variation in the prevalence of epilepsy: from 1.5 to 14/1000 people,
whereas in Africa this number can reach 15/1000 people, and in Latin America,
18/1000 people [7].
It is estimated that around three-quarters of people living with epilepsy in low and
middle-income countries do not receive the proper necessary treatment. Epilepsy can
be treated; medication is usually cheap and effective to control seizures in 70% of
cases. Unavailability of antiepileptic drugs added to the lack of trained health
professionals to recognize, diagnose and treat epilepsy leads to a low adherence to
treatment by patients [6]. Thereby, morbidity and mortality of people with epilepsy
differ across the world, and are often directly related to culture, economic develop-
ment, social aspects, health resources, discrimination and lack of knowledge of
health professionals [7, 8].
Rejection often starts in childhood and may be associated with parents’ lack of
knowledge or belief that crises could be caused by witchcraft or spells [9, 10].
Different authors mention that discrimination is motivated by the idea of assigning
patients with epilepsy to impotence, frailty and mental deficiency, as well as the fear
of witnessing and having to deal with seizures [3–5].
The lack of knowledge is considered a major determinant of social rejection
against people with epilepsy [4, 5]. In the study by Souza et al [11], almost 10% of the
study population, senior university students, believed epilepsy was contagious, and
more than 34% believed it was a mental illness.
Marriage of epileptic patients is still a subject surrounded by prejudice. In the
United States, only in 1980 did the last state ban the prohibition of marriage
involving persons with epilepsy [12].
Stigma related to this pathology affects people in different ways and can influence
social relationships, schooling opportunities, employment and emotional health of
the person with epilepsy [13].
Several studies have been carried out to identify knowledge about epilepsy among
family members, caregivers of the person with epilepsy and also among university
students [3, 4, 9–11, 14–19]. These authors have noticed that, regardless of schooling
and culture, the lack of knowledge is huge. The study by Souza and Oliveira [20], a
systematic review, observed large gaps on knowledge about the disease among
future health professionals. This is a cause of concern, as future and current
7-2
professionals are responsible for the reception of the person with epilepsy in health
services, especially in the scope of interdisciplinary and multidisciplinary attention.
A multidisciplinary team must be involved in the treatment of the person with
epilepsy. A physician is responsible for the diagnosis and prescription of drug
treatment, a nutritionist provides guidance on a patient’s diet, the social service
professional informs patients about their rights and duties, psychologists assist
cognitive evaluation and physical rehabilitation professionals aid patients in their
motor activities in daily living.
Professional preparation and training of health professionals must emphasize the
approach to epilepsy. Increasing the level of knowledge and confidence of health
professionals will reduce the obstacles faced by people living with epilepsy and their
families in the different social environments, especially regarding access to education
and health services.
According to Rahman [10], university students are a part of society with high
levels of knowledge and great potential to become role models and contribute to the
growth of the country. Not only is it essential to ensure that the ones in need are able
to have access to treatment, but access to information about epilepsy is equally
important to patients and professionals.
7.2 Cultural differences in knowledge of epilepsy

Knowledge of health professionals and students about the diagnosis and treatment
of epilepsy may result in a greater access of these patients to health services, as well
as broadening their social and professional opportunities. A recent study on
knowledge about epilepsy in students of health under graduation courses carried
out in five countries [11] revealed that students still have many misconceptions about
the disease. Data do not differ from others’ aforementioned surveys.
As advances in access to information and technologies are more available day
after day, many people and health professionals still treat individuals with epilepsy
with dated attitudes as occurred at the beginning of the last century [21].
According to the historical survey by Gomes [22] people with epilepsy of ancient
Rome were avoided for fear of contamination, and in the middle ages they were
persecuted as witches. University students also claim that epilepsy is contagious,
even in minor percentages, yet such an assertion is still present, and 42.9% of
US students still believed that epilepsy can be considered a symptom of mental
illness [11].
Additional data from the study by Souza et al [11], obtained from the authors,
indicate that, out of the 102 study participants, only 8% considered themselves to
have a good awareness about epilepsy, although about 25% of them had already
assisted someone with this condition.
The Epilepsy Knowledge Questionnaire [8] was employed by Souza et al [11] and
consists of 55 items (34 to assess knowledge of medical aspects of epilepsy and 21 to
evaluate knowledge of social aspects), and in 54 of them, participants must answer
‘true’ or ‘false’ to the statement. There is also a multiple-choice question about the
prevalence of the disease.
7-3
Regarding the medical aspects of epilepsy, more than 50% of the students
participating in the research provided wrong answers to the following questions
(correct answers are in brackets):
• Epileptic seizures are a temporary lack of oxygen to the brain (False).
• All those who start drugs for epilepsy have to take them for life (False).
• If you forget to take an antiepileptic drug for a day, it is usually ok to take
two doses together (True).
• Brain surgery is a method of preventing seizures (True).
• Most mothers taking antiepileptic drugs can breastfeed (True).
• Most seizures may result in brain damage (False).
Despite this, more than 80% of correct answers were given to questions about non-
contagion of the disease, seizures in individuals without epilepsy, electroencephalo-
gram diagnosis, duration of seizures, cerebral hemispheres affected in seizures,
description of epileptic seizures, regularity in medication use, cure of the disease,
stress and ingestion of alcoholic beverages [11].
As for the social aspects of the disease, more than 50% of the students
participating in the research responded incorrectly to the following questions:
• It is possible that a person whose seizures only happen during sleep may hold
a driver’s license (False).
• If a person has been seizure free for 10 years and has the correct license, he/
she may be allowed to drive heavy vehicles, public service vehicles or taxis
(True).
• During a seizure, you should put a hard object such as a spoon or pen in the
mouth of the patient (False).
• If a person with epilepsy has a simple, uncomplicated seizure, there is no need
to call a doctor or ambulance (True).
• Most people with epilepsy should avoid taking active part in sports (False).
• Most people with epilepsy should avoid working with open machinery (True).
• Most people with epilepsy should avoid all factory and building work (False).
• In medical terms, epilepsy is a recent phenomenon (False).
• What proportion of the population do you believe have active epilepsy?
(4–10/1000).
The only statements in this domain where students demonstrated good knowledge
(more than 80% of correct answers) related to declaring the presence of the disease to
the traffic authorities, presence of the epileptic subject in environments with blinking
lights, television and computer, water sports practice, immigration and the perform-
ance of labor activities at heights.
Considering that the research participants would soon become health professio-
nals, important issues related to the care of the epileptic patient appear not to have
been adequately addressed. Fewer than 5% of students correctly answered that, in
case of crisis, no object should be introduced into the patient’s mouth. In addition, in
two countries, not a single student provided a correct answer to this statement.
7-4
Similar data was found regarding people who have epilepsy during sleep only.
Less than 5% of the students believed that, in that case, there is no need to confiscate
the driver’s license. In addition, the lack of professional knowledge about driving by
people with epilepsy has a direct impact on the orientation received by patients. A
study with physicians found that these professionals were not able to provide this
orientation properly [18].
In the study by Souza et al [11], there were no significant differences in the
knowledge of students from developed and developing countries. Although authors
considered the general knowledge of the sample to be low, students from Brazil and
Portugal performed slightly better than the ones from Argentina and USA,
especially concerning the medical aspects of the disease.
7.3 Assessment of knowledge

A study comparing the knowledge of freshman and senior health students [3],
observed that, as expected, the latter had greater knowledge about epilepsy.
However, in the authors’ judgement, the knowledge of senior students was still
insufficient.
Likewise, a questionnaire was applied by Simonato et al [23] to obtain
information about personal attitudes and general knowledge about epilepsy with
first and second grade teachers from private and public schools and university
professors from medical and non-medical areas. The total sample was 97 people, all
from the city of São Paulo (Brazil). According to this survey, no differences were
observed among teachers from public and private schools, and university professors.
They all presented erroneous attitudes. Half of the teachers would introduce objects
into the mouth of an epileptic student in seizure. More than one third of the
professors (health area included) would follow the same procedure.
In order to fill a gap in university knowledge, an educational activity was
developed and implemented in health students [4]. One hundred and sixteen students
answered a questionnaire before and after an intervention aimed at increasing
knowledge related to epilepsy. After the 2 h workshop, the answers to the
questionnaire applied by the authors were more adequate than those previously
provided, however, many gaps in student knowledge were still observed.
A study carried out with students from a southwestern Asian country [5] revealed,
besides inadequate knowledge about the disease, prejudiced attitudes of future
professionals towards persons with epilepsy. In addition, most participants of this
study assumed that they would be afraid to treat such patients because they fear
having to deal with seizures.
Prejudice about epilepsy and the negative stereotypes of the population still
persist, and people living with epilepsy still continue to suffer from the lack of
technical information and the negative beliefs of the people around them.
7.4 Future perspectives

The World Health Assembly (WHA) has recently adopted a resolution on epilepsy
to promote a coordination action in several countries to discuss its implications for
7-5
health, social and public knowledge. Called WHA68.20, the resolution highlights
the need for the World Health Organization to help strengthen the capacities of
countries to tackle epilepsy and facilitate resource mobilization. This resolution is a
powerful tool to help countries implement effective actions to improve medical and
social services for people living with epilepsy, promote public awareness about the
disease and allocate resources for research [6].
The WHA68.20 provides general information about the disease, an overview of
current projects as examples of what can be achieved, and actions that can be taken
by specific groups. Themes and preparation of the material was conducted by
professionals with expertise in public policy, specialized professional, people living
with epilepsy and their families, non-governmental organizations and the general
public [6].
The pilot-project was initiated in four countries (Ghana, Mozambique,
Myanmar, and Vietnam) with one of the main goals being to expand the skills of
non-specialist health professionals to diagnose, treat and follow-up on people with
epilepsy.
It is straightforward and clear that treating epilepsy around the world is still very
challenging because there are wide gaps:
• it is still considered a low priority in many countries;
• health care systems have limited capacities and face inequitable distribution
of resources;
• lack or severe shortage of appropriately trained staff;
• inadequate access to affordable drugs;
• societal ignorance and misconceptions;
• poverty.
Between 2002 and 2004, China, where epilepsy prevalence is approximately 4.6/1000
people, participated in a large project entitled ‘Out of the shadows’. The project was
carried out in six provinces and aimed at testing feasibility to diagnose and treat
epilepsy in primary care, with the use of first line medication [24].
During the implementation phase of the project, 2455 patients with epilepsy were
treated (out of the 66 000 screened). Educational activities on epilepsy for the
general public, patients and their families were conducted through media channels
(television and press media) and aimed at stressing to the community that epilepsy is
a treatable condition [24].
Among the main results, 34% of the patients were seizure-free within one year and
34% had their seizures decreased by over 50%. Two years after the intervention that
included antiepileptic medication and education, the treatment gap in the project
area of the participating provinces significantly decreased from 62.6% to 49.8% (a
reduction of 12.8%).
Trained primary health care physicians are able to diagnose and treat people with
epilepsy, and this model of care can successfully reduce the epilepsy treatment gap.
In 2012, over 24 000 public health workers were trained in epilepsy management and
nearly 200 000 people were screened for epilepsy [6, 24].
7-6
The routine inclusion of discussions on this subject throughout university

education will result in the students’ familiarity with the disease and consequently
will have repercussions on their future professional attitudes. This may result in
better access for people with epilepsy to basic health services. Besides, effectively
prepared professionals will certainly contribute to increase treatment adherence and
personal and professional opportunities for persons with epilepsy.
References
[1] Commission on Classification and Terminology of the International League Against
Epilepsy (ILAE) 1981 Proposal for revised clinical and electroencephalographic classification
of epileptic seizures Epilepsy 22 489–501
[2] World Health Organization 2001 Epilepsy: Epidemiology, Etiology and Prognosis - WHO:
Factsheet 165
[3] Fonseca L C, Tedrus G M A S, Costa A C F, Luciano P Q and Costa K C 2004
Conhecimentos e atitudes sobre epilepsia entre universitários da área da saúde Arq.
Neuropsiquiatr. 52 1068–73
[4] Tedrus G M A S, Fonseca L C and Vieira A L C 2007 Knowledge and attitudes toward
epilepsy amongst students in the health area Arq. Neuropsiquiatr. 65 1181–5
[5] Hassona Y M, Mahmoud A A A, Ryalat S M and Sawair F A 2014 Dental students’
knowledge and attitudes toward patients with epilepsy Epilepsy Behav. 36 2–5
[6] World Health Organization 2015 WHA68.20 Global Burden of Epilepsy and the Need for
Coordinated Action at the Country Level to Address its Health, Social and Public Knowledge
Implications (Geneva: World Health Organization) Accessed 25 April 2019. Available from:
http://apps.who.int/gb/ebwha/pdf_files/WHA68/A68_R20-en.pdf
[7] Neni S W, Latif A Z A and Lua S Y W P L 2010 Awareness, Knowledge and attitudes
towards epilepsy among rural populations in east coast Peninsular Malaysia: A preliminary
exploration Seizure 19 280–90
[8] Jarvie S, Espie C A and Brodie M J 1993 The development of a questionnaire to assess the
knowledge of epilepsy, 1: general knowledge of epilepsy Seizure 2 179–85
[9] Kabir M, Iliyasu Z, Abubakar I S, Kabir Z S and Farinyaro A U 2005 Knowledge, attitude
and beliefs about epilepsy among adults in a northern Nigerian urban community Ann. Afr.
Med. 4 107–12
[10] Rahman A B 2005 Awareness and knowledge of epilepsy among students is a Malaysian
University Seizure 14 593–6
[11] Souza P, Portes L A, Thomas R K, Bonito J R, Rua M, Pacheco F J, Plaatjes P and Oliveira
N C 2018 Knowledge about epilepsy in university health students: a multicenter study
Epilepsy Behav. 79 112–16
[12] McLin W M and Boer H M 1995 Public perceptions about epilepsy Epilepsia 36 957–9
[13] Jacoby A 2002 Stigma, epilepsy, and quality of life Epilepsy Behav. 03 10–20
[14] Doughty J, Baker A G, Jacoby A N and Lavaud V 2003 Cross-cultural in levels of
knowledge about epilepsy Epilepsia 44 115–223
[15] Falavigna A, Tele A R, Roxo M R R, Velho M C, Silva R C, Mazzocchin T and Vedana
V M 2009 Awareness and attitudes on epilepsy among undergraduate health care students in
southern Brazil J. Epilepsy Clin. Neurophysiol. 15 19–23
7-7
[16] Goel S, Singh N, Lal V and Singh A 2013 Knowledge, attitude and practice of students
about first aid epilepsy seizures management in a northern Indian city Ann. Indian Acad.
Neurol. 16 538–43
[17] Goldstein L H, Minchin L, Stubbs P and Fenwick P B C 1997 Are what people know about
their epilepsy and what they want from an epilepsy service related Seizure 06 435–42
[18] Thomas R H and Hughes T A T 2009 ‘Can I drive, doctor?’ LEAN thinking may help us
answer the question Pract. Neurol. 9 71–9
[19] Young G B, Derry P, Hutchinson I, John V, Matijevic S, Parrent L and Wiebe S 2002 An
epilepsy questionnaire study of knowledge and attitudes in Canadian college students
[20] Souza P and Oliveira N C 2017 Conhecimento sobre epilepsia em universitários da área da
saúde Cad. Esc. Saúde 17 25–9
[21] Masia L S and Devinsky O 2000 Epilepsy and behavior: a brief history Epilepsy Behav. 1 27–36
[22] Gomes M M 2000 Doctor’s perspectives and practices regarding epilepsy Arq. Neuropsiquiat.
58 221–6
[23] Simonatto D, Dias M D, Pinto T H B B and Albuquerque M 1992 Epilepsia e Educação
Pública Arq. NeuroPsiquiat. 50 309–18
[24] World Health Organization 2009 Epilepsy Management at Primary Health Level in Rural
China: A Global Campaign Against Epilepsy Demonstration Project (Geneva: World Health
Organization)
7-8
IOP Publishing

Chapter 8
Current methods and new trends in signal
processing and pattern recognition for the
automatic assessment of motor impairments:
the case of Parkinson’s disease
Juan Rafael Orozco-Arroyave, Juan Camilo Vásquez-Correa and Elmar Nöth1
This chapter describes current methods and future trends in signal processing and
pattern recognition techniques applied to the automatic analysis of motor impair-
ments observed in patients with Parkinson’s disease (PD). A revision of the state of
the art is presented considering methods and applications that include the analysis
of speech, gait, and handwriting signals for the evaluation of PD. The suitability of
existing methods and the potential use of novel approaches are also presented in the
context of the automatic diagnosis and monitoring of the disease progression. Apart
from the classical clinical scales used to assess the neurological state of PD patients,
specific scales designed to evaluate the speech impairments of the patients are also
reviewed. Additionally, the suitability of a modified version of the Frenchay
Dysarthria Assessment scale (m-FDA) is evaluated. Besides the classical signal
processing methods based on the analysis in the time and frequency domains, other
methods that consider non-linear features and also deep neural networks are
explored in several experiments. We believe that the methods presented here are a
step forward in the development of automatic systems that will assist doctors, in the
near future, in the process of diagnosing and monitoring of PD patients.
1
Juan Rafael Orozco-Arroyave and Juan Camilo Váquez-Correa are with GITA Lab, Faculty of Egineering,
University of Antioquia, Medellín-Colombia and with Pattern Recognition Lab, University of Erlangen-
Nürnberg, Erlangen-Germany.
Elmar Nöth is with Pattern Recognition Lab, University of Erlangen-Nürnberg, Erlangen-Germany.

8.1 Introduction
PD was first described by Dr James Parkinson in his well known publication:
An essay on the shaking palsy [1]. Dr Parkinson defined the disease as ‘Involuntary
tremulous motion, with lessened muscular power, in parts not in action and even
when supported; with a propensity to bend the trunk forward, and to pass from a
walking to a running pace: the senses and intellects being uninjured’ [1]. Although
the description was done more than two centuries ago, it was accurate, especially in
the description of the motor abnormalities (non-motor affections started to be
documented a few decades ago), and covered most of what people commonly
describe today as PD. Since its first description, many scientists and clinicians have
studied the disease and have tried to find its cure, however, in spite of the huge
efforts made globally, there is still no way of stopping or preventing it. Humanity
has made a lot of progress in understanding and treating PD, but still it seems like a
long way has to be traveled before finding a cure. Thanks to the fast development of
sensing and modeling methodologies, the last decade has shown new alternatives for
patients, care-givers and clinicians to evaluate and monitor the symptoms. This
chapter will show current developments and future technological trends in the
automatic evaluation of PD. Methods based on the modeling of speech, gait, and
handwriting signals will be presented along with several analyses of their suitability
to support the diagnosis. Methods to help in the automatic monitoring of the disease
progression will also be introduced. Although non-motor symptoms have been
recently documented as part of the disease manifestation and progression, this topic
is out of the scope of this chapter. We suggest readers interested in knowing about
non-motor signs in PD should review the literature of emerging topics of
neuroscience.
8.2 Clinical assessment of the disease

8.2.1 General symptoms and common tools for neurological evaluation of PD
Several cardinal features describe PD including resting tremor, rigidity, akinesia or
bradykinesia, postural instability, flexed posture, and freezing [2]. A typical
examination of a PD patient in clinics may include the evaluation of several
features/symptoms which are briefly described below [3]:
• Bradykinesia: this is probably the most disabling symptom in PD and affects
all of the motor skills of the patients. Amplitude, precision, and rhythm of
movements are typically affected.
• Tremor: this is the symptom that the people associate the most with PD. It
consists of 4–5 Hz trembling of limbs (typically the arms) when they are fully
relaxed. It can also be easily observed when the patient is walking.
• Rigidity: it is manifested as involuntary resistance to movements (passive and
active) and it is observed as a ‘cogwheel movement’. The typical way of
diagnosing it is known as Froment sign (first described by James Froment)
and consists of observing an increase in tone of the limb during contralateral
active movements.
8-2
• Postural instability and deformity: postural instability in PD appears mainly

due to postural loss of reflexes. It is manifested as difficulties when rising from a
chair or pivoting when turning around (during walking). Patients with PD
typically take more steps (short ones) to turn around and take more than one
step to recover their position when the doctor pulls them from the shoulders
while standing behind them. Postural deformity appears in the more advanced
stages of the disease and typically results in flexed neck and trunk posture.
• Gait disorders: symptoms related with abnormal gait typically appear at early
stages of the disease. They are manifested as slow and shuffled gait and
sometimes with loss of arm swing. Short steps are also common in PD
patients especially when they are starting or stopping the gait.
Although most of the motor symptoms of PD are well documented, its progression
differs among patients, therefore it is also necessary to evaluate impairments (motor
and non-motor) in the context of each patient [2]. The evaluation of PD in clinical
practice is mainly based on neurological scales which are designed to guide the
expert clinician during the assessment. There exist several scales but the most
common are the Hoehn and Yahr (H&Y) and the Unified Parkinson’s Disease
Rating Scale (UPDRS).
The H&Y scale was first introduced in 1967 by Dr Margaret M Hoehn and Dr
Melvin D Yahr [4]. The main aim was to provide a gross assessment of the disease
progression. The first version of the scale is composed by integer numbers ranging
from 0 (no signs of disease) to 5 (wheelchair bound). The scale was later revised in
2004 by the Movement Disorder Society Task Force on Rating Scales for
Parkinson’s Disease. The resulting updated scale currently comprises seven disease
stages from 1 to 5 in integer steps except for the additional intermediate stages in 1.5
and 2.5 [5]. The H&Y is mainly used by clinicians to have a brief but accurate
impression about the disease progression. However, when more detailed information
is required, this scale is not suitable and a more detailed evaluation is necessary.
With the aim to provide a comprehensive but efficient evaluation, the UPDRS was
introduced in 1987 by Dr Fahn, Dr Elton, and members of the UPDRS program [6].
The scale was designed to include clinical information already included in other
existing scales and to provide more sensitivity to the clinical assessment. Such a scale
incorporated most of the clinical observations documented by neurologist experts
around the world by that time. The total scale ranges between 0 and 199 and has
55 items and it is composed by four sub-scales: (i) mentation, behavior, mood
(4 items); (ii) activities of daily living (13 items); (iii) motor examination (27 items);
and (iv) complications of therapy (11 items).
The UPDRS became the most widely used clinical scale for PD, however, the
Movement Disorder Society (MDS) decided to revise it in 2003 with the aim of
finding possible ambiguities, weaknesses, and areas where current scientific develop-
ments needed to be included [7]. The result of such a revision was published in 2008
and it is the currently used MDS-UPDRS scale [8]. The updated scale has a total of
65 items and it is also distributed into four parts or sub-sections: (i) non-motor
aspects of experiences of daily living (13 items); (ii) motor aspects of experiences of
8-3
daily living (13 items); (iii) motor examination (33 items); and (iv) motor compli-
cations (6 items). The maximum score of this scale is 272.
The aforementioned scales have been considered the global standard for the
clinical evaluation of PD. Those scales were designed and validated by expert
neurologists which is a very strong argument for being accepted globally. However,
the scales are mainly focused on motor symptoms related with the movement of the
arms, hands, legs, neck, etc. The communication capabilities of the patients are only
covered within one item, therefore speech disorders that appear due to the
progression of the disease (like the hypokinetic dysarthria) are under-represented.
There exist several scales for the evaluation of speech impairments related with PD.
Three of them are briefly described in the next subsection.
8.2.2 Clinical evaluation of dysarthria

There are therapies that clinicians have designed to improve oral communication
skills of PD patients. For instance the Lee Silverman Voice Treatment (LSVT) was
introduced in 1987 by Dr Loraine Ramig [9]. This is an auto-therapy protocol
designed to guide the patient while executing the exercises. The treatment includes
five tasks: production of the sustained vowel /ah/, sustained phonation of the vowel
/ah/ while decreasing and increasing the tone, reading of ten isolated words, and
finally reading of a sentence with ten words2. Another therapy called Pitch Limiting
Voice Treatment (PLVT) was presented in 2003 by Dr Bert J M de Swart et al in
[10]. It includes several exercises to improve posture and relaxation in patients. The
main aim is to reduce stress in the jaw and larynx. The therapy consists in asking the
patient to speak loudly and slowly. Prosody therapy is also included by pronouncing
words syllable by syllable. According to the authors, since the PLVT ‘teaches’ the
patients how to speak with a low voice, it prevents them from increasing the
laryngeal muscle tone, which results in lower laryngeal resistance and more relaxed
vocal cords. The third scale that we want to mention in this chapter is the Frenchay
Dysarthria Assessment (FDA). It was introduced in 1983 and later revised in 2008
by Dr P M Enderby and Dr R Palmer [11]. As its name suggests, this scale is focused
on the evaluation of dysarthria which is the condition that groups most of the speech
impairments developed by PD patients. The FDA scale covers a wide range of
aspects including reflexes, respiration, lips movement, palate movement, laryngeal
capacity, tongue posture/movement, intelligibility, and swallowing. To evaluate
swallowing the administration of the scale requests the patient to drink different
kinds of beverages like water and yogurt before speaking, therefore the evaluation
requires the patient to be with the examiner during the assessment. In many cases the
transportation from home to the clinic is not possible, especially for PD patients in
intermediate or advanced stages who typically have a reduced mobility.
Additionally, the scale is also not suitable for patients who live in remote rural
areas where there is almost no clinical expert. To overcome these issues we recently
proposed a modified version of the FDA scale, namely m-FDA [12]. This version
2
Lee Silverman voice treatment: http://www.lsvtglobal.com.
8-4
can be administered considering only speech recordings of the patients. Of course

swallowing aspects are not covered in this version of the scale, however most of the
speech aspects included in the original FDA scale are included in the modified
version. Details are presented in the next subsection.
8.2.3 The modified Frenchay dysarthria assessment (m-FDA)

The m-FDA scale consists of 13 items and evaluates seven aspects of the speech
including respiration, lips movement, palate/velum movement, laryngeal movement,
intelligibility, and monotonicity. Each item ranges from 0 (completely healthy) to
4 (very impaired). The total score ranges from 0 to 52 [12]. Preliminary evaluations
of the validity and suitability of the scale were performed with speech recordings of
the PC-GITA corpus [13]. This database contains recordings of 100 participants, 50
with PD and 50 age- and gender-matched healthy controls. The labeling process
with the m-FDA was performed by three phoniatricians who first agreed on the
evaluations of ten speakers (five per group and randomly chosen). These initial
evaluations allowed the experts to standardize the evaluation criteria. Afterwards,
the experts evaluated the speech of the other 90 speakers independently. The inter-
rater reliability among the phoniatricians was 0.75, which was computed calculating
the average Spearman’s correlation between all possible pairs of raters [12].
Table 8.1 summarizes the speech aspects and items included in the scale. The total
scale evaluates 13 items divided into seven aspects of the speech production. Each
item is evaluated in a range from 0 to 4 (integer values), thus the total scale ranges
from 0 (healthy speech) to 52 (completely dysarthric).
The m-FDA scale has not been clinically validated yet, however, it can be used to
get information about the progression of symptoms related with speech.
Additionally, it can be administered based on speech recordings, i.e. patients can
stay at home to do the exercises on their own or following the instructions given by
Table 8.1. Aspects and items included in the m-FDA scale.
Aspect m-FDA items
Breathing (1) Duration of respiration

(2) Respiratory capacity
Lips (3) Strength of closing the lips
(4) General control the lips
Palate/Velum (5) Nasal escape
(6) Velar movement
Laryngeal (7) Phonatory capacity in vowels
(8) Phonatory capacity in continuous speech
(9) Effort to produce speech
Tongue (10) Velocity to move the tongue in /pa-ta-ka/
(11) Velocity to move the tongue in /ta/
Intelligibility (12) General intelligibility
Monotonicity (13) Monotonicity and intonation
8-5
the doctor who is in the clinic or even given by a virtual agent. This is a step towards
the automatic administration of speech and language therapy for PD patients. We
believe that in the near future these therapies will help clinicians to better monitor
the disease progression and the impact of the medication intake.
With the aim of showing the suitability of the scale to differentiate between PD
patients and healthy speakers, figure 8.1 shows histograms and fitted probability
density functions of the scores assigned by the phoniatricians to the speakers of the
extended multimodal PC-GITA corpus, which is described in section 8.5. The
median among the three experts was used for comparison purposes. The total scores
of the scale and the seven aspects separately are included. The Mann–Whitney
U-test was performed to reject the null hypothesis that the distribution of the
m-FDA scores assigned for HC and PD speaker are equal. For all cases the null
hypothesis was rejected ( p ≪ 0.005). Although the two groups are not perfectly
discriminated, the scale shows itself to be clinically useful to support the diagnosis
and follow-up processes. Section 8.7 shows the results obtained in different experi-
ments where the suitability of the m-FDA scale is evaluated.
8.3 Automated analysis of the disease

8.3.1 Motor signals for the study of PD
During its progression, PD affects all of the sub-systems involved in motor activities
like speech production, walking, handwriting, and others. Non-motor activities are
(a) (b) (c) (d)
(e) (f) (g) (h)
Figure 8.1. Fitted probability density functions and histograms of the m-FDA scores assigned for HC and PD
patients. (a) total m-FDA; (b) m-FDA, breathing aspect; (c) m-FDA, lips movement; (d) m-FDA, palate/
velum movement; (e) m-FDA, larynx movement; (f) m-FDA, tongue movement; (g) m-FDA, monotonicity;
and (h) m-FDA, intelligibility. U indicates the Mann–Whitney U statistic and p is the resulting p-value.
8-6
also affected but they are out of the scope of this chapter. The degree of impairment
caused by each motor sub-system defers from one patient to another and highly
depends on the medication intake, therefore it is necessary to have methods to
quantify to which extent a motor skill is affected and also to know in which stages of
the disease it is potentially required to provide permanent or partial assistance to the
patient. The next subsections present a summary of methods and techniques typically
used in the state-of-the-art to quantify the degree of impairment developed in three
different motor skills: speech production, gait, and handwriting. The description
includes classical and also new methods like those based on deep neural networks [14].
8.3.1.1 Speech production

Speech impairments observed in PD patients are typically grouped and called
hypokinetic dysarthria. This includes features such as rigidity of the vocal folds,
bradykinesia, and reduced muscular control of the larynx and other organs involved
in speech production. The effect of dysarthria in speech includes increased acoustic
noise [15], reduced intensity [16], harsh and breathy voice quality [17], increased
voice nasality [18], monopitch, monoludness, and speech rate disturbances [19],
imprecise articulation of consonants [20], and involuntary introduction of pauses
[21]. It is documented that speech impairments are among the first to appear, even
before clinicians diagnose the disease [22, 23] and their progression highly affects
social life of patients [24]. Several studies have described speech impairments
developed by PD patients in terms of four different dimensions: phonation,
articulation, prosody, and intelligibility [22, 25–28].
Phonation symptoms are related to the stability and periodicity of the vocal fold
vibration. Different phonation deficits have been observed in PD patients, including
differences in glottal noise compared to healthy speakers, incomplete vocal fold
closure, and vocal folds bowing. These characteristics are typically evaluated with
measures such as noise-to-harmonics ratio (NHR), glottal-to-noise excitation ratio
(GNE), harmonics-to-noise ratio (HNR), and voice turbulent index (VTI) [29, 30]. On
the other hand, phonation symptoms can also be evaluated in terms of perturbation
measures such as jitter, shimmer, amplitude perturbation quotient (APQ), pitch
perturbation quotient (PPQ), and non-linear dynamics (NLD) measures [12, 25, 30,
31]. These measures are thought to quantify abnormal and/or unstable vibration of
vocal folds during the production of vocal sounds. Additional phonation features may
include the use of time–frequency representations such as modulation spectra to assess
tremor in the speech of the patients [32], features computed from the reconstruction of
the glottal source signal [33, 34], or the use of the wavelet transform to evaluate
fluctuations in the contour of the fundamental frequency F0 [35].
Articulation symptoms are related to the modification of position, stress, and
shape of several limbs and muscles to produce speech. These symptoms have been
classically modeled with features based on the formant frequencies, vowel space area
(VSA), formant centralization ratio (FCR) [36], and vowel articulation index (VAI)
[37]. Besides these measures, mainly based on the production of sustained vowel
phonations, clinicians have observed the imprecise production of stop consonants
such as /p/, /t/, /k/, /b/, /d/, and /g/ during the production of continuous speech signals
8-7
[20, 38, 39]. The literature also reported shallower formant frequencies in PD
patients than in healthy speakers, which may indicate incomplete tongue/jaw
movements [40, 41]. Other articulation symptoms include reduced duration of
vocalic segments and transitions, and increased voiced onset time (VOT) [42].
Other articulation features include those extracted in [43], where the authors
modeled six different articulatory deficits in PD: vowel quality, coordination of
laryngeal and supra-laryngeal activity, precision of consonant articulation, tongue
movement, occlusion weakening, and speech timing. Besides, another articulation
model was proposed in [26], where the author modeled the difficulty of PD patients
to start/stop the vocal fold vibration in continuous speech. The model was based on
the energy content in the transitions from unvoiced to voiced and from voiced to
unvoiced segments. In [44] the authors proposed articulation features based on the
VOT segments, including the VOT duration, the VOT ratio, and the vowel
variability quotient. In [45] the authors modeled the dynamics of the amplitude
envelope of DDK exercises using several NLD features. Another articulation model
was proposed in [46], where the authors considered a forced alignment strategy to
segment the different phonetic units in the speech utterances. The phonemes were
segmented and grouped to train a Gaussian mixture model (GMM) for each
phoneme unit. The posterior probabilities of the GMMs obtained per phoneme
were used to model the articulation capabilities of the patients.
Prosody deficits in PD are manifested as monotonocity, monoloudness, reduced
stress, and changes in speech rate and pauses [19]. In addition, bradykinesia and
freezing of movement sometimes cause difficulty in the initiation of voluntary speech
and inappropriate long silences. The most common features used to model prosody
are based on the F0 and energy contours, duration, and pitch periods [12, 27, 47, 48].
Finally, intelligibility allows one to measure how comprehensible is the speech of a
person. In other words, how much of what a person is saying can be understood
during a conversation. This speech dimension has been classically evaluated by means
of perceptual tests [49]. This approach is expensive, time consuming and varies among
different listeners. With the aim to reduce these sources of error and ambiguity, the
research community started to use automatic speech recognition (ASR) systems to
automatically assess intelligibility. Typical evaluations are based on the word error
rate (WER), which is the most common measure of intelligibility and recent studies
have revealed that it is higher in PD patients than in healthy controls [12, 50, 51].
The most recent approaches consider the aforementioned classical dimensions
(phonation, articulation, prosody, and intelligibility) to train speaker models to
represent specific traits in speakers. For instance, in [52] the authors considered
phonation, articulation, and prosody features to train speaker models based on
i-vectors [53]. In [28] the authors combined phonation and articulation features with
state-of-the-art speaker recognition techniques such as GMM-universal background
models (GMM-UBM) and i-vectors. The second approach was also considered in
[54] where the dysarthria severity of PD patients was modeled. The authors in [55]
also considered speaker models based on GMM-UBM systems and i-vectors to
monitor the disease progression of PD patients in a longitudinal study. Those
models were trained with phonation, articulation, and prosody features to evaluate
8-8
the contribution of each speech dimension to the total dysarthria level according to
the m-FDA score.
Besides ‘hand-crafted’ feature extraction models, there is a growing interest in the
research community to consider deep learning models to assess the speech of PD
patients. The ‘2015 computational paralinguistic challenge (ComParE)’ [56] had one
of the sub-challenges on the automatic estimation of the neurological state of PD
patients. The ground-truth was given according to the part III of the MDS-UPDRS
scale. The winners of the challenge [57] reported a Spearman’s correlation coefficient
of 0.65 when the speech tasks were grouped automatically per speaker and Gaussian
processes with deep neural networks (DNNs) were used to perform the prediction of
the clinical score. Later, a deep learning based articulation model was proposed in
[58] and improved in [59]. The focus of those recent studies was to model difficulties
of PD patients to stop/start the vibration of the vocal folds while speaking [26].
Onset and offset transitions were modeled with time–frequency representations,
which were then used as input for a convolutional neural network (CNN). The CNN
learned the most discriminative features from the time–frequency representation to
classify the speech of PD patients and HC speakers. The dysarthria severity
according to the m-FDA score was also evaluated. In [60] the authors proposed a
deep learning model to predict the dysarthria severity adding an intermediate
interpretable hidden layer with four perceptual dimensions: nasality, vocal quality,
articulatory precision, and prosody.
8.3.1.2 Gait
One of the major manifestations of PD appears in gait, and they typically cause
disability of patients in the most advanced stage of the disease. In the earliest stages
bradykinesia is reflected in smaller arm swing, slower turns and reductions in step
length [61]. With the disease progression, gait becomes more unstable, freezing of
gait (FoG) episodes occur, and falls are frequently reported [62]. Particularly, one of
the most common symptoms in gait is the FoG, which is defined as an absence or
marked reduction of forward progression of the feet despite the intention to walk
[63]. Patients describe FoG as a feeling of having the feet glued to the ground and
being temporarily unable to re-initiate gait. FoG is context-dependent, i.e. it triggers
when patients walk through narrow spaces, when they initiate or end gait, when an
obstacle impedes patients from following their gait trajectory, or when turning
around [64]. From the signal processing perspective, harmonics in acceleration
signals between 3 and 8 Hz have been observed when FoG occurs [65]. Another
revealing symptom is the tremor, which is defined as a rapid back-and-forth
movement of a body segment [66]. Tremor in PD patients appears mainly at rest,
and tends to disappear during posture or movement [67]. However, in severe stages
of the disease, it may remain present during hand posture or movement which is
called kinetic tremor [68]. The frequency associated with resting tremor typically
ranges between 3.5 and 7.5 Hz [69], and the frequency for kinetic tremor ranges from
4 to 12 Hz [67].
The research community has shown a growing interest in the automatic analysis
of gait in PD patients. The research objectives have been mainly focused to
8-9
discriminate between PD patients and HC subjects, to evaluate the neurological

state of the patients, and to detect specific gait impairments (like FoG or tremor)
during the walk [70]. Models have been commonly created with signals collected
from inertial sensors, e.g. accelerometers and gyroscopes attached to the body of the
patients [71–75], and with force-sensitive sensors placed inside the shoes [76, 77].
Most of the studies have considered kinematic features based on the duration and
velocity of the steps [75, 77–82]. In [77] the authors computed several kinematic
features such as stride, swing, and stance duration to discriminate between PD
patients and HC subjects. In [79] the authors aimed to find the most accurate set of
kinematic features to classify PD patients and HC subjects. The selected features
included the stride length and time, swing time, and step time asymmetry. In [75] the
authors proposed new features to assess gait impairments in PD patients. The set of
new features includes the peak forward acceleration in the loading phase and peak
vertical acceleration around heel-strike. According to the authors, the first feature
encodes the engagement in stride initiation and the second one the hardness of the
impact at heel-strike. In [81] the authors proposed a model to quantify the capability
of PD patients to turn around. The proposed model was based on the computation
of four kinematic features: number of steps, turning time, number of continuous
steps, and number of hesitations. Those features were used to train a fuzzy inference
system, which maps the kinematic features into a continuous scale related with the
turning capabilities of the patients. In [82] the same authors proposed a model to
quantify the leg agility of PD patients combining kinematic features and a fuzzy
inference system. The authors computed features related to the heel tapping exercise
such as the amplitude when raising the leg, number of hesitations, amplitude trend,
and speed trend.
Other studies have considered spectral features to evaluate gait impairments [71,
83–88]. In [71] the authors computed several spectral features such as the dominant
frequency, the energy in different frequency bands, and the entropy of the spectrum
to evaluate the severity of tremor in PD patients. In [72] the authors computed
several spectral and statistical features extracted from gait signals, including the
energy content in different frequency bands, the variance, the root-mean square
energy, and others. Other spectral features include those based on the peak
frequency, spectral centroids, bandwidth of the signal, and energy content distrib-
uted into several bands according to the discrete wavelet transform [89].
Additionally, the Freezing index was proposed in [80] to detect FoG episodes. The
index was defined as the power in the ‘freeze’ frequency band (3–8 Hz), divided by
the power in the normal loco-motor frequency band (0.5–3 Hz) [90]. A modified
version of such an index, based on the continuous wavelet transform, was proposed
in [85]. Finally, in a recent study the authors considered spectral features based on
the discrete wavelet transform and several statistical features to predict different sub-
scores of the MDS-UPDRS-III for lower limbs, including the bradykinesia sub-
score (section III, item 31) and the sum of the scores for leg agility (section III, item
26), arising from a chair (section III, item 27), and gait (section III, item 29) [88].
Apart from the majority of studies where kinematic, spectral and statistical
features are used to model gait impairments of PD patients, there are other works
8-10
that consider non-linearities that appear during the walking process [76, 89, 91–94].
For instance, higher complexity and randomness have been observed in the
acceleration in the sagittal axis of gait in healthy individuals compared to PD
patients [89, 95]. Conversely, in the frontal and transversal planes, higher complexity
and chaotic behavior has been observed in the gait of PD patients due to the tremor
condition and FoG episodes [89]. In [89] the authors considered the largest
Lyapunov exponent, Lempel–Ziv complexity, and several entropy measures to
classify PD patients and HC subjects. Other nonlinear measures like the fuzzy
entropy and the Teager–Kaisser energy have been extracted from signals captured
with force-sensitive switches placed inside the shoes. Finally, features based on
recurrence quantification analysis were also considered in [94] to detect FoG
episodes in PD patients.
As in the case of speech analysis, there are studies that have considered raw data
as input to machine learning pipelines mainly based on sequence learning strategies
like hidden Markov models (HMMs) [96], or the novel deep learning algorithms
[97, 98]. In [97] the authors aimed to detect FoG episodes in PD patients using
CNNs. The inputs to the networks were formed by stacking spectral representations
of consecutive time intervals. CNNs were also considered in [99] to detect FoG
events in PD patients using the raw data of inertial sensors. The authors in [98]
aimed to detect the time instance before a FoG event using long short-term memory
(LSTM) networks trained with raw signals obtained from inertial sensors. Although
this is a newly explored approach, preliminary experiments show promising results,
which is motivating many other researchers to use them.
8.3.1.3 Handwriting
The symptoms of PD that can be evaluated through handwriting include micro-
graphia, bradykinesia, and tremor [100]. Micrographia is related to the reduction of
the size in handwriting. Bradykinesia causes time during handwriting to be longer
than usual. Tremor is related to involuntary movements and the resulting irregular
shapes in drawings. The complete handwriting impairments in PD patients have
been grouped and called PD dysgraphia, which is related to difficulties in controlling
fine motor movements required during the handwriting process [101].
Handwriting assessment can be divided into online and offline, depending on how
the signals are acquired. Online handwriting signals are captured using digital
tablets, and contain information related to the dynamics of the handwriting process,
e.g. position and pressure of the pen, azimuth and altitude angles. Some tablets also
allow the capture of information from the in-air movement before the patient
supports the pen on the tablet’s surface. On the other hand, offline handwriting
signals can be collected using also tablets or using a regular pen and paper. Offline
handwriting analysis typically includes only spatial attributes from drawings. This
chapter is focused on signals and phenomena observed in online handwriting.
There are several studies that have considered handwriting assessment to classify
PD patients and HC subjects. Most of those studies are based on kinematic features
that consider the dynamics of the velocity, acceleration, and jerk of the strokes [102–
110]. These kinematic features have been commonly combined with features based
8-11
on the pressure of the pen and with features extracted from the azimuth or altitude
angles [102, 104, 108, 110]. Recent studies have shown that the in-air movement is
also important to characterize handwriting impairments of PD patients [102, 104,
107, 110]. New kinematic features were introduced in [111] to characterize the
velocity contour of the handwriting. The proposed features were based on a Sigma-
lognormal model, which considers the state of the neuromuscular system to perform
the handwriting process [112].
Other studies have shown the importance of other groups of features based on
geometric, spectral, and NLD analyses to characterize the handwriting impairments
of PD patients. For instance, in [113] the authors proposed features related to the
power spectral density of the speed stroke. Additional spectral features were
considered in [110], where the authors modeled the trajectory of an Archimedean
spiral as an amplitude modulated signal and computed features from the spectral
components of the resulting signal. Other geometric features were proposed in [114],
where the authors compute the difference between strokes drawn by the participants
and a template of an Archimedean spiral. Similar geometric features based on the
error between the Archimedean spiral drawn by the patient and a template were
proposed in [110]. On the other hand, NLD features were considered in [105], where
the authors compute features based on the entropy of the horizontal and vertical
movements. Other NLD features such as correlation dimension, Hurst exponent,
largest Lyapunov exponent, and others have also been used to characterize the
handwriting process [110].
Recent deep learning approaches have also been used to classify the handwriting
of PD and HC subjects. For instance the authors in [115] classified reconstructed
images of Archimedean spirals using a CNN. In [116] the authors proposed a model
called deep echo state network to classify PD patients and HC subjects who draw
Archimedean spirals. The deep learning model was based on recurrent neural
networks (RNNs) that processed the time-series of horizontal and vertical move-
ments, the grip angle, and the pressure of the pen when the patients draw the spirals.
A combination of NLD analysis and CNNs was proposed in [117] to classify
handwriting samples of HC subjects and PD patients. The authors considered a
spatial representation based on recurrent plots to visualize the temporal dynamics of
the handwriting samples. The images obtained from the recurrent plots were
characterized and classified using a CNN. In [118] the authors proposed the use
of parallel CNNs to extract different features from hand-drawn shapes. The
architecture of the CNNs include a pre-trained version of AlexNet [119] trained
with the ImageNet dataset. A pre-trained version of AlexNet was also considered in
[120] where the authors applied a transfer learning strategy on CNNs initially
trained with the ImageNet and MNIST databases.
8.4 Revealing features for the automatic diagnosis and monitoring

of Parkinson’s disease
The next subsection includes the description of different features that are extracted
from each bio-signal. In general, the dynamics of each signal is modeled considering
8-12
aspects like temporal variation, energy content, spectral and geometric information,
non-linear behavior, and others. Although details of the models and algorithms are
provided here, we recommend the reader to have a look of the references for further
details, especially those related with the implementation process.
8.4.1 Speech
8.4.1.1 Phonation features
They are used to model abnormal patterns in the vocal fold vibration and are
extracted from the voiced segments, where there is vibration of the vocal folds. They
can be computed in sustained vowels and in continuous speech signals. Phonation
features include seven descriptors computed for short-time frames of the speech
signal.
Jitter and shimmer describe temporal perturbations in frequency and amplitude,
respectively. Jitter is computed according to equation (8.1), where N is the number
of frames in the speech utterance, Mf is the maximum of the fundamental frequency,
and F0 corresponds to the fundamental frequency computed on the kth frame.
Shimmer is computed using equation (8.2), where Ma is the maximum amplitude of
the signal, and A(k ) corresponds to the amplitude on the kth frame.
N
100
Jitter (%) = ∑ ∣F0(k ) − Mf ∣ (8.1)
N · Mf k=1
N
100
Shimmer (%) = ∑ ∣A(k ) − Ma∣ (8.2)
N · Ma k = 1
APQ measures the long-term variability of the peak-to-peak amplitude of the speech
signal. The computation includes a smoothing factor of 11 voiced periods. Similarly,
PPQ measures the long-term variability of the fundamental frequency, with a
smoothing factor of five periods. Both APQ and PPQ are computed as the absolute
average difference between the amplitude or frequency values (for APQ or PPQ,
respectively) of each frame and the average of its neighbors, divided by the average
values for the complete signal. Both perturbation quotients are computed using
equation (8.3), where L = N − (k − 1), D(i ) is the pitch period sequence when
computing the PPQ or the pitch amplitude sequence when computing the APQ. N is
the number of frames, k is the length of the moving average (11 for APQ or 5 for
PPQ), and m = (k − 1)/2.
L 1 k
1 ∑ j =1D(i + j − 1) − D(i + m)
PQ = ∑ k 1 N
(8.3)
L i=1 ∑n=1D(i )
N
Additionally, the first and second derivatives of F0 are included in the model, along
with the energy content of the signal. Further details of the methods can be found in
[121]. Four statistical functionals are calculated per feature (mean, standard
deviation, skewness, and kurtosis), forming a 28-dimensional feature vector per
8-13
Figure 8.2. Sustained phonations of vowel /a/ and their corresponding F0 contour for a 71 year old healthy
speaker with m-FDA = 0 (left), and a 77 year old PD patient with m-FDA = 41 and MDS-UPDRS-III = 92
(right).
utterance. The implementation of the phonation features described here is available

online to be used by the research community3. Figure 8.2 shows an example with
two speech signals and their F0 contour. Recordings of an HC speaker (left) and a
PD patient (right) are included. Note the stability of the F0 contour for the HC
subject compared to the one obtained for the PD patient.
8.4.1.2 Articulation features

The main aim of the articulation features is to model aspects related to the
modification of position, stress, and shape of several limbs and muscles involved
in the speech production process. Articulation capabilities of the patients are
evaluated considering the energy content in the transition from unvoiced to voiced
segments (onset) and from voiced to unvoiced segments (offset). These features were
introduced in [26] and have shown to be accurate in modeling different speech
impairments of PD patients [12, 51, 122]. The detection of voiced/unvoiced segments
is based on the computation of F0. Those segments where there are F0 values are
considered as voiced, conversely segments where no F0 values are found, are labeled
as unvoiced. Once the border between voiced and unvoiced sounds are detected,
40 ms of the signal are taken to the left and to the right, forming a segment with
80 ms length. Figure 8.3 shows an example of the resulting spectrograms of an onset
transition for an HC speaker (left) and a PD patient (right). Note that the transition
for the HC subject is well defined, while the transition for the PD patient is not
clearly identified.
The spectrum of the transitions is distributed into 22 critical bands according to
the Bark scale. The Bark-band energies (BBE) are calculated according to [123]. For
frequencies below 500 Hz the bandwidths of the critical bands are constant at
100 Hz, while for medium and high frequencies the increment is proportional to the
logarithm of frequency. Equation (8.4) indicates the frequency distribution accord-
ing to the Bark scale. arctan(·) is measured in [radians] and f in [Hz].
⎛ f ⎞2
Bark(f ) = 13 · arctan(0.00076 f ) + 3.5 arctan ⎜ ⎟ (8.4)
⎝ 7500 ⎠
3
https://github.com/jcvasquezc/DisVoice/tree/master/phonation.
8-14
Figure 8.3. Speech signals and spectrograms of an onset transition for a 71 year old healthy speaker with
m-FDA = 0 (left), and a 77 year old PD patient with m-FDA = 41 and MDS-UPDRS-III = 92 (right).
In addition to the BBE computed upon the transitions, 12 Mel-frequency cepstral

coefficients (MFCCs) and their first two derivatives are also calculated per segment
to obtain a smooth representation of the voice spectrum in the transitions. The result
of such a procedure takes into account the human auditory perception, which makes
it more suitable to model speech signals. The articulation feature set are comple-
mented with the computation of the first and second formant frequencies, and their
first two derivatives. These additional features allow one to represent resonances in
the vocal tract and the capacity of the speaker to keep the tongue in a certain
position while producing voiced sounds. The implementation of the articulation
features described here is also available online and can be used by the research
community4.
8.4.1.3 Prosody features

This speech dimension is included to model timing, intonation, and loudness during
the production of natural speech. Prosody has been typically modeled by measures
related with duration and the contour of F0 and energy. Prosody features allow one
to model the monotonicity and monoloudness aspects in PD patients. A list of
features that are commonly computed to assess prosody in PD patients is shown in
table 8.2. The implementation of the prosody features is also available online5.
8.4.1.4 Phonological features

These are represented by a vector with interpretable information about the mode
and manner of articulation. These aspects are commonly known by clinicians
because they are related with specific movements in the articulators of the vocal
tract.
For this study, the different phonemes of the Spanish language are grouped into
18 phonological classes defined according to table 8.3, which are based on the
4
https://github.com/jcvasquezc/DisVoice/tree/master/articulation.
5
https://github.com/jcvasquezc/DisVoice/tree/master/prosody.
8-15
Table 8.2. Description of prosody features.
Num. Feature Description
Features based on F0
1–6 F0-contour Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
7–12 Tilt of a linear estimation of F0 for each voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
13–18 MSE of a linear estimation of F0 for each voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
19–24 F0 on the first voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
25–30 F0 on the last voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
Features based on energy
31–34 Energy-contour for voiced segments Average, Standard deviation, Skewness, Kurtosis
35–38 Tilt of a linear estimation of energy contour for each Average, Standard deviation, Skewness, Kurtosis
voiced segment
39–42 MSE of a linear estimation of energy contour for each Average, Standard deviation, Skewness, Kurtosis
voiced segment
43–48 Energy on the first voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
8-16
49–54 Energy on the last voiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
55–58 Energy-contour for unvoiced segments Average, Standard deviation, Skewness, Kurtosis
59–62 Tilt of a linear estimation of energy contour for each Average, Standard deviation, Skewness, Kurtosis
unvoiced segment
63–66 MSE of a linear estimation of energy contour for each Average, Standard deviation, Skewness, Kurtosis
unvoiced segment
67–72 Energy on the first unvoiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
73–78 Energy on the last unvoiced segment Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
Features based on duration
79 Voiced rate Number of voiced segments per second
80–85 Duration of Voiced Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
86–91 Duration of Unvoiced Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
92–97 Duration of Pauses Average, Standard deviation, Maximum, Minimum, Skewness, Kurtosis
98–103 Duration ratios Pause/(Voiced+Unvoiced), Voiced/Pause, Pause/Unvoiced, Unvoiced/
(Voiced+Unvoiced), Voiced/(Voiced+Unvoiced), Unvoiced/Pause
Table 8.3. Distribution of the different Spanish phonemes into phonological classes.
Phonological class List of phonemes Phonological class List of phonemes
Vocalic /a/, /e/, /i/, /o/, /u/ Consonantal /b/, /tʃ/, /d/, /f/, /g/, /x/, /k/, /l/, /ʎ/, /m/, /n/,
/p/, /ɾ/, /r/, /s/, /t/
Back /a/, /o/, /u/ Anterior /e/, /i/
Open /a/, /e/, /o/ Close /i/, /u/
Nasal /m/, /n/ Stop /p/, /b/, /t/, /k/, /g/, /tʃ/, /d/
Lateral /l/ Continuant /f/, /b/, /tʃ/, /d/, /s/, /g/, /ʎ/, /x/
Flap /ɾ/ Voice /a/, /e/, /i/, /o/, /u/, /b/, /d/, /l/, /m/, /n/, /r/,
/g/, /ʎ/
Trill /r/ Strident /f/, /s/, /tʃ/

Labial /m/, /p/, /b/, /f/ Dental /t/, /d/
Velar /k/, /g/, /x/ Pause /sil/
Figure 8.4. Phonological posteriors estimated for a 71 year old healthy speaker with m-FDA = 0 (left), and a
77 year old PD patient with m-FDA = 41 and MDS-UPDRS-III = 92 (right).
movement of different articulators in the vocal tract. ‘Pauses’ are considered as a

separate phonological class. The phonological posteriors are the conditional
posterior probability of a speech frame to belong to one or more phonological
classes. The phonological posteriors will be computed with a bank of parallel RNNs,
which estimate the probability of occurrence of a specific phonological class,
according to the method recently proposed in [124]. This methodology and the
source code is also available online to be used for research purposes6.
Figure 8.4 shows the difference between the phonological posteriors estimated for
an HC subject (left) and a PD patient (right) when they pronounce the Spanish
sentence mi casa tiene (my house has). Phonological posteriors for vocalic, stop,
nasal, and strident are included. Note that the tool estimates all phonological classes
for the HC subject, but it is not able to accurately differentiate several classes for the
PD patient, for instance nasal, stop, and vocalic posteriors are overlapped.
Different features can be computed from the contour of different phonological
posteriors. Some of them include the average posterior for each phonological class
6
https://github.com/jcvasquezc/phonet.
8-17
Table 8.4. List of features computed from the phonological posteriors.
Num. Phonological feature
1–18 Average of the posteriors

19–36 Average of the active posteriors (>0.5)
37–54 Average duration of the active posteriors
55–72 Standard deviation of the posteriors
73–90 Standard deviation of the active posteriors (>0.5)
91–108 Standard deviation of the duration of the active posteriors
Figure 8.5. Different phases of the walking process.
and the regularity of the posterior probabilities, which give information about the
capability of a speaker to pronounce different sounds like plosive, nasal, labial, and
others. The average duration and the variability of the duration of speech segments
labeled with specific phonological posteriors can also be computed. Table 8.4
summarizes the features computed from each of the 18 phonological posteriors to
assess the speech of PD patients.
8.4.2 Gait
8.4.2.1 Kinematic features
This set consists of measurements to model different properties in the strides
including time, distance, and velocity. At the same time, each stride can be divided
into two phases that can be analyzed individually: the stance phase when the foot is
on the ground and the swing phase when the foot is in the air. The toe-off angle and
the heel-strike angle can also be analyzed. Figure 8.5 shows the main kinematic
aspects that are considered to be analyzed during walking. The segmentation process
is based on a dynamic time warping (DTW) algorithm, where the gait signals are
compared with a template generated from healthy subjects [125].
Several kinematic features can be computed according to the described aspects of
the gait process. The feature set includes the stride, stance and swing times, the stride
length, the velocity of each stride, the toe-off angle, or the heel-strike angle. The
average and standard deviation can be computed per each walking task. In addition,
these features can be computed per foot (left and right) with the aim of evaluating
the contra-laterality effect [126], i.e. right handed patients are more affected in the
8-18
Figure 8.6. Phase space representation from gait signals captured with a gyroscope in the transversal (Z) plane
of a male HC subject, 66 years old (left) and a PD patient, female, 74 years old and MDS-UPDRS = 64 (right).
left lower limbs, while left handed patients may exhibit more impairments in the
right parts of the body.
8.4.2.2 Non-linear features

Gait is a complex and non-linear activity that can be modeled with non traditional
features like the non-linear ones. The first step to extract those features is the phase
space reconstruction. This process is performed according to Taken’s theorem [127],
which indicates that there exists a dimension m such that a time series xt can be
represented in a multidimensional space Xt, known as phase space or attractor. It
can be created using equation (8.5), where the time-delay τ is computed by the first
minimum of the mutual information function, and the embedding dimension m is
found using the false neighbors method [128].
X t = {xt , xt−τ , … , xt−(m−1)τ} (8.5)
Figure 8.6 shows the phase space reconstructed from gait signals corresponding to
20 m walking with one stop every 10 m. The signals were captured from a gyroscope
to measure angular movements in the transverse plane (z-axis) for an HC subject
(left) and for a PD patient (right). Note that the phase space of the HC subject
exhibits well defined trajectories with a clear recurrence. Conversely, the trajectories
for the patient are more dispersed and it is difficult to identify a clear trajectory in the
attractor. Different features can be extracted from the reconstructed phase space to
assess and compare the complexity and stability of the walking process performed by
a healthy subject and a PD patient. The features that can be extracted from those
embedded attractors are briefly described as follows.
Correlation dimension (CD)

This feature is a measure of the dimensionality of the phase space where the attractor
is embedded [129]. To estimate CD the correlation sum C(ε ) is defined according to
equation (8.6), where Θ is the Heaviside step function. C(ε ) can be interpreted as the
probability to have pairs of points in a trajectory of the phase space inside the same
8-19
sphere of radius ε. In [129], the authors demonstrated that C(ε ) represents a volume
measure, hence CD can be defined by equation (8.7).
n n
1
C (ε ) = lim ∑ ∑ Θ(ε − s[i ] − s[j ] ) (8.6)
n →∞ n(n − 1) i=1 j=i+1
log(C (ε ))
CD = lim (8.7)
ε→ 0 log(ε )
Largest Lyapunov exponent (LLE)

This feature measures the sensitivity of the system to changes in the initial conditions
of the signal according to the rate at which the nearby trajectories of the phase space
converge or diverge. This feature gives information about stability properties of the
time-series, which implies that a small perturbation introduced to the system at any
period, makes its behavior unpredictable The estimation of the LLE follows the
algorithm introduced in [130]. After the reconstruction of the phase space, the
nearest neighbor of every point in the trajectory is estimated. The LLE is estimated
as the average separation rate of those neighbors in the phase space.
Hurst exponent (HE)

This feature measures the long-term dependence of a time series. It is defined
according to the asymptotic behavior of the re-scaled range of a time series as a
function of a time interval [131]. The estimation process consists in dividing the time
series into intervals of size L and calculating the average ratio between the range R
and the standard deviation σ of the time series. HE is computed as the slope of the
curve obtained from equation (8.8).
R
LHE = (8.8)
σ
Lempel–Ziv complexity (LZC)

This feature measures the degree of disorder of spatio-temporal patterns in a time
series [132]. In the computation process the signal is transformed into binary
sequences according to the difference between consecutive samples, and the LZC
reflects the rate of new patterns in the sequence. It ranges from 0 (deterministic
sequence) to 1 (random sequence). Further details of the computation process can be
found in [133, 134].
Sample entropy (SampEn)

This feature is a regularity statistic that measures the average conditional informa-
tion generated by diverging points on trajectories in the phase space. Signals with
several repetitive patterns have smaller SampEn than signals with a complex
dynamics. This feature does not include self comparisons of points in the phase
8-20
space, which makes its computation more efficient than the computation of the
approximate entropy, which requires the comparison of all points in the phase space.
Detrended fluctuation analysis (DFA)

Similar to the HE, DFA evaluates the long-term dependency of a time-series, except
that the DFA can be applied when the signal exhibits a non-stationary behavior. The
DFA is used to estimate the stochastic component of the gait process by looking at
trends over time intervals in the signal.
8.4.3 Handwriting
8.4.3.1 Kinematic features
According to the literature, there are several kinematic features that can be extracted
from online handwriting data including the trajectory of the strokes, velocity of the
trajectory in the horizontal and vertical axes, acceleration of the strokes, features
based on the pressure of the pen and their derivatives, features based on the in-air
movement before the participant to put the pen on the tablet’s surface, and those
based on the azimuth and altitude angles. Different statistical functionals can be
estimated from the extracted features, such as mean, standard deviation, skewness,
kurtosis, maximum, and minimum. These features can be extracted over different
tasks.
8.4.3.2 Geometric features

A novel feature set based on geometrical and spectral analysis of the Archimedean
spiral was introduced recently in [110]. The trajectory of the Archimedean spiral is
modeled as an amplitude-modulated signal r̂ (t ) defined in equation (8.9).
rˆ(t ) = (a3t 3 + a2t 2 + a1t + a 0) · sin(2πft ) (8.9)
The real trajectory r(t ) is modeled as a sinusoidal signal with increasing amplitude
and frequency f. The amplitude coefficients (ai) of the modeled trajectory values are
estimated with a third-order polynomial regression based on the maximum peaks of
the real trajectory. f is the fundamental frequency of the trajectory and it is estimated
using the Fourier transform. Different features based on the geometry from the
trajectory can be extracted from the model: the MSE between the real and modeled
trajectories, the coefficients of the third order polynomial used for the model
(ai , i ∈ {0, 1, 2, 3}), and the location of the peaks in the trajectory. In addition,
several spectral features can also be computed from the trajectory, including the
fundamental frequency, the amplitude of the first spectral components of the
trajectory.
Figure 8.7 shows an example of the difference in the handwriting of a PD patient
and an HC subject. Figure 8.7(a) and (b) shows the Archimedean spiral drawn by an
HC subject and a PD patient, respectively. Note the tremor exhibited in the spiral
drawn by the patient. Information extracted from the pen when the participants
draw the spirals is shown in figure 8.7(c) and (d) for the HC subject and the PD
patient, respectively. Note the difference in the stability of the pressure of the pen,
8-21
(a) (b)
(c) (d)
(e) (f)
Figure 8.7. Different information extracted from handwriting signals captured from an HC subject and a PD
patient. (a) Archimedean spiral drawn by a 71 year old HC subject. (b) Archimedean spiral drawn by a
73 years old PD patient with MDS-UPDRS-III = 65. (c) Signals extracted from the pen while the HC subject
was drawing the spiral. (d) Signals extracted from the pen while the patient was drawing the spiral. (e) Real and
modeled trajectory for the HC subject. (f) Real and modeled trajectory for the PD patient.
the difference in the trajectory (r) followed by the patient and by the HC subject.
Note also that the number of pen-up and pen-down movements (z) performed by the
PD patient is larger than the number exhibited by the HC. Finally, figure 8.7(e) and
(f) displays the model of the trajectory described by equation (8.9) for a HC subject
and a PD patient, respectively. Note that the trajectory of the PD patient is more
irregular than the one observed in the HC subject. Note also that there is a high
difference in the MSE values between the modeled and real trajectories for the HC
subject (MSE = 0.33) and the PD patient (MSE = 2.12).
8.5 Existing data

Pattern recognition applications need data to be trained and tested. This is a
standard and mandatory procedure that allows scientists to know whether the tested
approach is suitable and accurate for the addressed task. The medical field is perhaps
the most difficult one in which to obtain suitable data that will allow the
construction of robust, stable, and reliable models. The main aim of the following
subsections is to provide the reader with information about existing databases (some
8-22
of them public) that could be used to start or deepen the study of motor impairments
in PD patients. Speech, gait, handwriting, and multimodal studies are reviewed. In
some cases the data can be freely accessible. For cases where a website is not
provided to download the data, we encourage the reader to contact the authors of
each study and ask whether it is possible to get the access.
8.5.1 Speech
One of the first scientific approaches for the quantitative analysis of PD speech was
the Parkinson’s Voice Initiative (PVI)7. During that project sustained phonations of
the vowel /ah/ pronounced by about 50 patients were recorded. Although the
recordings are not publicly available, the main contribution of this initiative was to
capture the attention of many people around the world and to motivate other
researchers to address this problem. A few years after the PVI, Professor Sabine
Skodda in [135] presented a study where a total of 73 PD patients and 43 healthy
controls (German native speakers) were recorded. The participants were asked to do
several speech tasks including the sustained phonation of vowel /ah/, rapid repetition
of the syllables /pa-ta-ka/, the reading of a text with 81 words, and a monologue.
This corpus continued to be enhanced and its most updated versions were further
used in several studies including [136] and [26]. In the same year, Dr Jan Rusz
introduced a corpus with 20 newly diagnosed PD patients and 16 healthy control
subjects (Czech native speakers) [22]. The corpus included recordings of the
sustained vowel /i/, rapid repetition of the syllables /pa-ta-ka/, twelve isolated words,
three sentences, a read text with 80 words and a monologue. Since its release, the
database has been continuously updated and now the authors have recorded a total
of 50 PD patients and an equal number of healthy speakers [137]. In 2013 a Turkish
database was released [138]. The corpus contains recordings of 20 patients with PD
and 20 healthy subjects (all of them Turkish native speakers). The speech tasks
included sustained vowels, isolated words, digits, and sentences. One year later, in
2014 the corpus PC-GITA was released [13]. This database contains recordings of 50
speakers with PD and 50 age and gender balanced healthy control subjects. All of
the subjects are Colombian Spanish native speakers. The recording process was done
in noise-controlled conditions, and using a professional audio setting. The partic-
ipants were requested to perform several tasks including: sustained phonation of the
five Spanish vowels, six diadochokinetic exercises (rapid repetition of the syllables
/pa-ta-ka/, /pe-ta-ka/, /pa-ka-ta/, /pa/, /ta/, /ka/), a set with 45 isolated words, 10
sentences, a reading text with 36 words (containing all of the sounds and phonemes
of the Colombian Spanish), and a monologue. Thanks to the generosity of the
participants and the people in the Parkinson’s foundation in Medellín, Colombia
(Fundalianza Parkinson Colombia8,9), this corpus is available upon request by
contacting the first author of the paper where the data was released [13]. In 2015
7
Parkinson’s voice initiative (PVI) http://www.parkinsonsvoice.org/.
8
Fundalianza Parkinson Colombia https://es-la.facebook.com/fundalianzaparkinsoncolombia/.
9
Fundalianza Parkinson Colombia https://www.fundalianzaparkinson.org/.
8-23
another corpus was presented in [139]. It includes recordings of 168 PD patients, all
of them English native speakers. No healthy control people participated in the study.
The recordings include the sustained phonation of vowel /ah/, rapid repetition of the
syllables /pa-ta-ka/, and a reading passage. There is also a recent initiative that
pushed the study of Parkinson’s speech. It is led by the mPower consortium and
contains recordings of more than 2000 speakers including PD patients and HC
subjects [140, 141]. The corpus includes recordings of the sustained vowel /ah/
collected using smartphones. Apart from the databases mentioned previously, there
is a set available in the UCI-ML repository, which contains acoustic feature vectors
extracted from recordings of sustained vowels produced by 31 subjects (23 with
PD)10. Unfortunately only the feature vectors are available.
With the fast development of portable and wearable sensors, gait and handwriting
signals started to become popular and accessible several years ago. Some of the
studies where data of these two bio-signals are used to evaluate motor skills of PD
patients are mentioned below.
8.5.2 Gait
There are several studies where gait signals of PD patients are used, and there are also
repositories with data available to work on this topic. For instance around 2005 the
PhysioNet repository11 was released. This database contains measures of gait from 93
PD patients and 73 HC subjects. The data is thought to mainly address questions
about reaction force during walking. The task consisted of asking the participants to
walk for about 2 min on level ground. Later, in 2010, the authors in [142] presented
and released the Daphnet Freezing of Gait Data Set, which aimed to evaluate FoG
events. The recording system used on-body acceleration sensors to measure move-
ments of the patients. About 8 h of recording signals were captured from 10 PD
patients and FoG events were observed in 8 of them. These data are also available in
the UCI-ML repository12. In [143] the authors presented a database with a total of 92
PD patients and 81 HC subjects. Gait signals were collected using the eGaIT system13,
which consists of accelerometers and gyroscopes attached to the lateral heel of the
shoes to record motion signals. The study was extended and now the database includes
recordings of 190 patients and 101 healthy subjects [144]. The typical tasks recorded
include a 20 m walk and a 40 m walk with a pause every 10 m. Toe–heel tapping and
time up and go tests are also included in many cases. Most of the recent studies of gait
are based on wearable sensors attached to the body or to the shoes. In addition, there
exists another way of collecting walking signals using a walkway. In [145] the authors
presented a database with signals recorded from 310 PD patients while walking on a
walkway. Different gait parameters can be extracted from this recording device,
however note that its use is restricted to clinical environments.
10
Parkinsons Data Set. https://archive.ics.uci.edu/ml/datasets/Parkinsons.
11
PhysioNet: the research resource for complex physiologic signals. https://physionet.org.
12
Daphnet Freezing of Gait Data Set. https://archive.ics.uci.edu/ml/datasets/Daphnet+Freezing+of+Gait.
13
https://www.astrum-it.de/healthcare-medizintechnik/forschungsprojekte/sensorbasierte-bewegungsanalyse.
html.
8-24
In general terms the trend in the community shows that the most
suitable approach consists of recording signals from wearables unobtrusively, such
that the patient does not notice that is being monitored. Although this approach is
still under development, preliminary experiments have shown that it is promising
and will potentially be used to monitor elderly people and ambulatory patients, not
only those with PD but also with other clinical conditions.
8.5.3 Handwriting
The automatic analysis of online handwriting exercises performed by PD patients
has also increased in recent years. There are several studies in this field and a couple
of databases are publicly available. For instance, the PahaW database was released
in 2016 and it is publicly available14. The corpus contains recordings of different
handwriting tasks including the Archimedean spiral, the cursive letter ‘l’, the bigram
‘le’, a set of words, and one sentence. A total of 37 PD patients and 38 HC subjects
participated in the study. Several signals were collected including the on-surface
movement, in-air movement, pressure, and position. Further information about the
corpus can be found in [104]. There is also a repository in the UCI-ML website15.
This database was released in 2014 [146] and contains drawings of the Archimedean
spiral performed by 25 PD patients and 15 HC subjects. Besides the aforementioned
databases, there is a corpus in the Kaggle repository16 which contains handwriting
recordings of 62 PD patients and 15 healthy subjects. The database was collected in
2009 and it is publicly available. The recorded tasks include the Archimedean spiral
(with and without template to guide the participant) and circles drawn around a red
point displayed on the tablet.
8.5.4 The extended multimodal PC-GITA corpus

This database is an extended version of the PC-GITA corpus [13], and corresponds
to the data used for the experiments and results described in section 8.7. This
extended version contains recordings of speech, handwriting, and gait, collected
from 106 PD patients and 87 HC subjects. All of the subjects are Colombian Spanish
native speakers. None of the participants in the HC group has history of symptoms
related to PD or any other kind of movement disorder. In addition, some of the PD
patients were recorded in up to 10 different sessions from 2012 to 2019. This
longitudinal data allow the study of the disease progression considering different
modalities: speech, gait, and handwriting. There is a total of 225 recording samples
performed by the PD patients. The HC subjects were recorded only in one session.
The three bio-signals were captured in the same session during 1 h, distributed as
follows: 15 min for speech, 30 min for gait, and 15 min for handwriting.
Unfortunately, not all bio-signals are available for all recorded sessions.
14
https://bdalab.utko.feec.vutbr.cz/.
15
https://archive.ics.uci.edu/ml/datasets/Parkinson+Disease+Spiral+Drawings+Using+Digitized+Graphics+
tablet.
16
https://www.kaggle.com/team-ai/parkinson-disease-spiral-drawings.
8-25
Table 8.5. Clinical and demographic information of the subjects.
PD patients HC subjects Patients versus controls
Gender [F/M] 49/57 43/44 p = 0.99a

Age [F/M] 60.9(11.2)/64.7(9.4) 61.4(9.8)/64.9(10.5) p = 0.60b
Time since diagnosis [F/M] 15.5(14.5)/8.1(5.9) —
MDS-UPDRS-III [F/M] 36.2(18.1)/36.3(18.9) —
m-FDA total [F/M] 26.3(7.9)/26.4(7.7) 6.6(7.0)/9.0(8.2) p ≪ 0.005b
Time since diagnosis and age are given in years. [F/M]: Female/Male. Average(Standard deviation).
a
p-value calculated through Chi-square test.
b
p-value calculated through Mann–Whitney U-test.
Information of the three bio-signals is available for 39 of the HC subjects and for 6
recording sessions of the PD patients, which make multimodal data available for 125
of the 225 recording samples.
The patients were evaluated by a neurologist expert and labeled according to the
MDS-UPDRS-III scale in 6 of the 10 recording sessions. Most of the patients are in
the initial or intermediate state of the disease. Additionally, the speech recordings
were labeled by expert phoniatricians according to the m-FDA scale [12] described
in section 8.2. Table 8.5 summarizes clinical and demographic aspects of the
participants included in the corpus. Statistical tests are included in the caption of
the table to validate the balance in gender and age, and the significant difference that
exists between the m-FDA scores assigned to PD patients and HC subjects. The
distributions of age, MDS-UPDRS-III, total score of the m-FDA scale, and the time
post the diagnosis are shown in figure 8.8. The distributions of each sub-item of the
m-FDA scale can be found in figure 8.1.
The speech of the participants was recorded with a sampling frequency of 16 kHz
and 16 bit resolution. The same speech tasks recorded in the PC-GITA corpus [13],
except for the isolated words, are included in this extended version. Detailed
information about the 10 sentences is included in table 8.6.
Handwriting data consist of online drawings captured with a tablet Wacom cintiq
13-HD17 with a sampling frequency of 180 Hz. The tablet captures six different
signals: x-position, y-position, in-air movement, azimuth, altitude, and pressure. The
subjects performed a total of 14 exercises divided into writing and drawing tasks (see
table 8.7). Additional information about the handwriting exercises can be found in
[110, 147, 148].
Gait signals were captured with the eGaIT system18, which consists of a 3D-
accelerometer (range ±6g ) and a 3D gyroscope (range ± 500°/s) attached to the
external side (at the ankle level) of the shoes [125]. Data from both feet were
captured at a sampling rate of 100 Hz and 12 bit resolution. The exercises included
17
Cintiq 13HD Graphic pen tablet for drawing http://www.wacom.com/en-us/products/pen-displays/cintiq-13-
hd.
18
https://www.astrum-it.de/healthcare-medizintechnik/forschungsprojekte/sensorbasierte-bewegungsanalyse.
html.
8-26
Figure 8.8. Distribution of age, MDS-UPDRS-III values, total score of the m-FDA scale, and time post PD
diagnosis.
Table 8.6. Details of the sentences included in the corpus.
Sentence Duration # words # phonemes
1 Mi casa tiene tres cuartos 1.9(0.4) 5 20

2 Omar, que vive cerca, trajo miel 2.6(0.7) 6 22
3 Laura sube al tren que pasa 2.2(0.5) 6 19
4 Los libros nuevos no caben en la mesa de la oficina 3.4(1.0) 11 39
5 Rosita Niño, que pinta bien, donó sus cuadros ayer 4.3(1.2) 9 37
6 Luisa Rey compra el colchón duro que tanto le gusta 4.0(1.2) 10 38
7 Viste las noticias? Yo vi ganar la medalla de plata en 7.9(2.2) 17 67
pesas, Ese muchacho tiene mucha fuerza!
8 Juan se rompió una pierna cuando iba en la moto 3.2(0.9) 10 34
9 Estoy muy triste, ayer vi morir a un amigo 3.3(0.8) 9 32
10 Estoy muy preocupado, cada vez me es más difícil 4.3(1.1) 10 41
hablar!
Duration is given in seconds. Average (Standard deviation).
Table 8.7. Handwriting tasks performed by the participants.
Writing tasks Drawing tasks
Alphabet Circle
Free sentence Guided circle
Name Cube
Digits Repetition of the cursive letter l
Signature Repetition of the cursive letter m
Rectangles
The Rey–Osterrieth figure [149]
Free Archimedean spiral
Archimedean spiral with template
8-27
20 m walking with a stop after 10 m (2 × 10 walk), and 40 m walking with a stop

every 10 m (4 × 10 walk). Additional information about the gait data can be found
in [93].
8.6 Automatic analysis of the extracted features

The suitability of the features extracted from speech, handwriting, and gait signals is
evaluated in two scenarios: (1) to discriminate between healthy subjects and PD
patients, and (2) to predict the disease severity of the patients. Different learning
algorithms are used for each scenario.
The classification of PD patients and HC subjects is evaluated with a support
vector machine (SVM), which assigns the feature vector x ∈ d with d number of
features to one of the classes i.e. PD or HC. The main aim of the SVM algorithm is
to find the optimal separating hyper-plane to maximize the separability between the
classes. The decision function of the SVM is expressed according to equation (8.10).
The term ξi is a slack variable that penalizes the amount of errors allowed in the
optimization process. yi ∈ {−1, +1} are the class labels, ϕ(x) is a kernel function
that transforms the feature space x into a higher dimensional space where a linear
solution exists. The weight vector w and the bias value b are the terms that define the
separating hyper-plane.
yi · (w⊤ϕ(xi ) + b) ⩾ 1 − ξi , i = 1, 2, 3, … , N (8.10)
The optimization problem to find the separating hyper-plane is given by equation

(8.11), where the hyper-parameter C controls the trade-off between ξi and the width
of the margin. The samples where the condition yi · (w⊤xi + b ) = 1 − ξi holds, i.e.
those vectors that lie on the hyper-planes, are called support vectors.
1 N
minimize ∥w∥2 + C ∑ ξi
w, b 2 i =1
(8.11)
subject to y · (w⊤xi + b) ⩾ 1 − ξi
ξn ⩾ 0
We consider a Gaussian kernel function ϕ(x), which is given by equation (8.12),
where the hyper-parameter γ is the bandwidth of the kernel function.
2∥x 2
ϕ(x) = e−γ i−xj∥ ∀i ,j (8.12)
The prediction of the disease severity of the patients is performed with a regression
algorithm. In this case it is considered a support vector regression (SVR). The SVR
formulation is similar to that addressed for the SVM. For this case the labels yi are
real variables (yi ∈  ), like the scores of the m-FDA or the MDS-UPDRS-III scales.
The SVR introduces an ε-insensitive loss-function to predict the state the patients
yî ∈  . The regression function is defined as yî = w⊤xi + b, and the optimization
problem can be formulated as in equation (8.13). The term ∣yî − yi ∣ ⩽ ε is the error in
the prediction, and ε is the insensitive hyper-parameter which makes the loss-
function zero for errors smaller than ε.
8-28
1 N
minimize ∥w∥2 + C ∑ ξi
w, b 2 i =1
(8.13)
subject to ∣yî − yi ∣ ⩽ ε + ξi
ξi ⩾ 0
The hyper-parameters of the learning algorithms have to be optimized, e.g. C and γ
for the SVM, and C, γ, and ε for the SVR. The optimization process is performed
following a cross-validation strategy, which consists of dividing the feature set
extracted from all subjects into three partitions, one used to train the algorithms, one
used to evaluate and optimize the hyper-parameters, i.e. development set, and the
last one is used for test. These partitions are performed K times with different
samples in the test set in order to guarantee that all samples are tested at least once.
This process is called K-fold cross-validation.
The optimization of the hyper-parameters of the learning algorithms is performed
in a randomized search strategy, as follows: the values of the hyper-parameters C, γ,
and ε are modeled with an exponential probability density function, which generates
values for each hyper-parameter to be evaluated according to the performance in the
development set. After several iterations with different generated values from the
probability functions, the hyper-parameters that produced the highest accuracy are
stored. After K-folds, the optimal hyper-parameters are found based on the median
of the values of the hyper-parameters obtained for each fold. Finally, the K-fold
cross-validation is repeated but only with the train and test set in order to guarantee
that all test samples are evaluated with the optimal hyper-parameters, which leads to
more realistic and stable results. However, despite this optimization process, the
most realistic results are those obtained by evaluating the trained algorithm with a
completely independent test set that never participated in the optimization process.
This test set would behave like a group of patients who visit the doctor and request a
medical screening, without having participated in the creation of the screening tool.
Although this is the most realistic scenario, lack of data to create the models in
clinical applications limits the scientists to perform experiments with independent
test sets. A good example of experiments performed considering a realistic scenario
can be found in [110].
8.6.1 Performance metrics

The performance of the learning algorithms is evaluated with different metrics that
have been proposed to evaluate performance in classification and regression
experiments. For classification algorithms, the most common evaluation metrics
are derived from the Confusion matrix, which is a table that allows one to visualize
the percentage of miss-classified samples per class. An example of a confusion matrix
derived from the two-class problem of discriminating between PD and HC subjects
is presented in table 8.8. The following concepts can be defined from the table:
• True positives (TP) represent the number of patients correctly classified as PD
patients.
8-29
Table 8.8. Example of a confusion matrix for the classification of PD and

HC subjects.
Predicted classes
PD HC
Real PD TP FN
classes HC FP TN
• True negatives (TN) represent the number of healthy subjects correctly

classified as healthy.
• False positives (FP) are the number of healthy subjects classified as patients.
• False negatives (FN) are the number of patients subjects classified as healthy
subjects.
Different performance metrics can be computed from the confusion matrix. The
most common are accuracy (ACC), sensitivity (SENS), and specificity (SPEC).
These metrics are defined in equation (8.14), (8.15), and (8.16), respectively. ACC
represents the capability of the system to correctly classify samples of both classes;
SENS indicates the performance of the system to correctly detect samples of the
target class e.g. PD patients; and SPEC represents the capability of the system to
correctly classify samples in the non-target class e.g. HC subjects.
TP + TN
ACC = × 100% (8.14)
TP + TN + FP + FN
TP
SENS = × 100% (8.15)
TP + FN
TN
SPEC = × 100% (8.16)
TN + FP
There are several cases where the accuracy may not reveal the real performance of
the classification algorithm. For instance when the number of samples in one class is
higher than the number of samples in the other one, which is known as unbalanced
classes. For those cases the unweighted average recall (UAR) is defined as a
performance metric to deal with the unbalance of the samples in the test set. The
UAR is defined as the average between the percentage of correct samples classified
per class. For a two-class problem, the UAR is equivalent to the average between the
specificity and the sensitivity.
In addition to the described metrics, the receiver operating characteristic (ROC)
curve is a graphical representation of the performance of a classification algorithm.
It is based on the trade-off between the true positive rate (TPR) or sensitivity and the
false positive rate (FPR) or (1-specificity) [150]. An example of an ROC curve is
shown in figure 8.9, where the TPR defines the number of correct positive results
8-30
Figure 8.9. Example of the scores and the decision threshold of a classification algorithm (left), and an ROC
curve (right).
when a decision threshold is shifted towards the decision space, denoted as the
vertical line from the left part of figure 8.9. On the other hand, the FPR defines the
number of miss-classified positive samples that appear while the decision threshold is
shifted. The doted line in the ROC curve in the right part of figure 8.9 represents the
random decision. The area under the ROC curve (AUC) is a more robust perform-
ance metric of a classification algorithm, and ranges from 0.5 (chance) to 1.0 (perfect
classification) [150].
Regarding the performance metrics of a regression algorithm, it is defined as the
Pearson correlation coefficient (r) between the real labels yi and the predicted ones ŷi
according to equation (8.17), where σ refers to the standard deviation and cov is the
covariance between the predicted and real labels. r is a real number that ranges
between −1 and 1 and indicates the linear relation between both labels. Note that
r = 1 indicates total positive linear correlation, r = 0 means no linear correlation, and
r = −1 indicates total negative linear correlation which means an opposite behavior
between the real and predicted labels.
cov(yi , yî )
r= (8.17)
σ (yî ) · σ (yi )
As the correlation between the real and predicted labels will not be always linear,
there exists the Spearman’s correlation coefficient (ρ), which allows one to measure
the non-linear relation between both variables. ρ can be defined as the Pearson’s
correlation coefficient between the rank of the real labels rg(yi ) and the rank of the
predicted labels rg(ŷi ). It can be defined according to equation (8.18), where di is the
difference between the two ranks of each sample, and N is the number of samples.
Table 8.9 indicates a common way to classify different Spearman’s correlation
intervals and their corresponding criteria to interpret results.
2
cov(rg(yi ), rg(yî )) 6∑d i
ρ= =1− (8.18)
σ (rg(yî )) · σ (rg(yi )) N (N 2 − 1)
8.7 Experiments and results

The following subsections introduce different experiments performed with speech,
gait, and handwriting signals separately, and also experiments performed with the
8-31
Table 8.9. Different correlation intervals and their corresponding

interpretation criteria.
Correlation interval Criteria
0.00 < ∣ρ∣ < 0.20 Very weak

0.20 ⩽ ∣ρ∣ < 0.40 Weak
0.40 ⩽ ∣ρ∣ < 0.60 Moderate
0.60 ⩽ ∣ρ∣ < 0.80 Strong
0.80 ⩽ ∣ρ∣ ⩽ 1.00 Very strong
combination of the three of them. Results of classification and regression experi-

ments are included.
8.7.1 Experiments with speech signals

The features described in section 8.4.1 are used to model speech signals recorded
from the speakers in the multimodal extended PC-GITA corpus, described in section
8.5. Different speech tasks are considered to perform the classification of PD
patients and HC subjects. Besides, the dysarthria severity according to the m-FDA
score is evaluated with regression experiments. Different algorithms are trained for
each speech dimension: phonation, articulation, prosody, and phonological, and for
each speech task: six DDK tasks (DDK1: /pa-ta-ka/, DDK2: /pa-ka-ta/, DDK3: /pe-
ta-ka/, DDK4: /pa/, DDK5: /ta/, DDK6: /ka/), read text, monologue, and read
sentences according to the list in table 8.6. Finally, the models for each feature set
and exercise are combined to improve the results in the prediction. Details of each
experiment are presented below.
8.7.1.1 Classification of PD versus HC

Results classifying PD patients and HC subjects are shown in table 8.10. The results
are shown in terms of the AUC because, as was described above, it is a more robust
metric to evaluate two-class classification systems. The fusion was performed based
on the most voted decision according to the local decisions made by the classifiers on
each feature set and speech task. The highest AUC per task is highlighted in bold.
Note that the fusion of the speech tasks and feature sets highly improves the
results of the classification. The highest AUC value is obtained by combining
information of all speech tasks and all feature sets (0.927). When considering
different models individually, the highest AUC values (above 0.80) are obtained
with articulation features, which indicates the suitability of the approach to
accurately model articulation deficits of PD patients. It also confirms previous
observations extensively documented in the literature where articulation impair-
ments are highlighted as the most prevalent and sensitive in the speech of these
patients. It is also worth stressing the fact DDK tasks show good results (around
0.80) which is very positive, especially considering that those tasks are very easy to
produce and are potentially useful to evaluate the speech of patients in almost every
8-32
Table 8.10. AUC values for the classification of PD versus HC subjects.
Feature set
Exercise Phonation Articulation Prosody Phonological Fusion
DDK1 0.719 0.818 0.686 0.794 0.829

DDK2 0.711 0.866 0.693 0.717 0.815
DDK3 0.718 0.834 0.623 0.808 0.819
DDK4 0.689 0.778 0.645 0.716 0.784
DDK5 0.671 0.830 0.644 0.750 0.820
DDK6 0.723 0.797 0.661 0.755 0.824
Read text 0.691 0.888 0.786 0.764 0.872
Monologue 0.672 0.859 0.743 0.808 0.858
Sentence 1 0.638 0.831 0.710 0.680 0.806
Sentence 2 0.701 0.820 0.746 0.737 0.829
Sentence 3 0.631 0.830 0.668 0.731 0.806
Sentence 4 0.708 0.827 0.719 0.752 0.860
Sentence 5 0.634 0.819 0.710 0.772 0.827
Sentence 6 0.752 0.819 0.682 0.768 0.836
Sentence 7 0.695 0.840 0.691 0.774 0.819
Sentence 8 0.708 0.827 0.786 0.695 0.864
Sentence 9 0.710 0.888 0.762 0.838 0.880
Sentence 10 0.737 0.863 0.711 0.748 0.846
Fusion 0.845 0.908 0.881 0.845 0.927
DDK: diadochokinetic exercise.
language. This is mainly because the production of these DDK tasks requires the
speaker to move several articulators including lips, tongue, and velum. The results
obtained with phonation, prosody and phonological features are around 0.70. This
is likely because although phonation and prosody are affected in PD patients, these
are not the most sensitive aspects to evaluate dysarthric speech signals. In general
terms, monologue and DDK tasks exhibit the highest AUC values when phono-
logical features are used. This is because monologues are rich in sounds and
phonemes, and DDK tasks require the production of stop sounds like /p/, /t/, and
/k/ that need specific energy and timing to be correctly pronounced. Other
phonological classes like nasals and strident are included in the model which allows
the evaluation of specific impairments in the speech production. In summary, each
feature set and speech task is useful to characterize different aspects related to speech
production. The improvements observed when models and tasks are combined
indicate that the information is complementary and suitable to be used to assess the
speech of PD patients.
Details of the results obtained with the fusion of speech tasks and feature sets are
presented in table 8.11. Note the balance between sensitivity and specificity. This is a
good indicator because it shows that the system is robust and equally good in
detecting PD patients and detecting HC speakers. The ROC curve resulting from
8-33
Table 8.11. Details of the results when all features and speech tasks are combined to classify PD versus HC
subjects.
ACC UAR SENS SPEC AUC
Fusion of features and tasks 84.6% 85.5% 84.1% 86.8% 0.927

ACC: accuracy. UAR: unweighted average recall. SENS: sensitivity.
SPEC: specificity. AUC: area under the ROC curve.
(a) (b)
Figure 8.10. (a) ROC curve of the best classification result with speech signals. (b) Distribution of the
classification scores.
these experiments is presented in figure 8.10(a). Additionally, since in this case we

are using an SVM to perform the classification, the scores computed by the classifier
to assign each sample to one of the two classes correspond to the distance of each
sample to the separating hyper-plane. Figure 8.10(b) shows the result of plotting
histograms of those scores and their corresponding fitted probability density
functions.
8.7.1.2 Prediction of the dysarthria level of the speakers based on the m-FDA
The same feature sets considered in the classification experiments are also considered
to predict the m-FDA scores assigned by expert phoniatricians. Spearman’s
correlation coefficients (ρ) are computed between the total m-FDA score and each
extracted feature from the speech exercises with the aim to evaluate the suitability of
the features to predict the dysarthria severity of the speakers. The features that
exhibited the highest correlation coefficients are listed in table 8.12. Only
Spearman’s correlations larger than 0.4 (moderate) were considered for the ranking.
Note that none of the prosody features are part of the list. This fact can be explained
because the m-FDA scale is mainly focused on evaluating articulation deficits, which
are the most prevalent and sensitive in the speech of PD patients. The ranking of
8-34
Table 8.12. Speech features with the highest Spearman’s correlation between real and predicted m-FDA scores.
Phonation Articulation Phonological
Feature Task ρ Feature Task ρ Feature Task ρ
avg. PPQ DDK5 0.444 std. BBE8_on Sent. 6 −0.526 avg. back_post_act Read text −0.515
avg. APQ Sent. 7 −0.437 std. BBE9_on Sent. 6 −0.526 std. vocalic_post Read text −0.482
avg. Jitter DDK5 0.432 std. BBE16_on Read text −0.524 avg. back_post_act Sent. 7 −0.482
std. F0 DDK5 0.424 std. BBE15_on Read text −0.523 std. vocalic_post Sent. 7 −0.468
std APQ DDK5 0.423 std. BBE7_on Sent. 6 −0.521 std. back_post DDK3 −0.467
avg. APQ Sent. 8 −0.419 std. BBE11_on Sent. 6 −0.518 avg. anterior_post DDK4 0.466
avg. APQ Sent. 10 −0.403 std. BBE12_on Sent. 6 −0.511 std. back_post Sent. 7 −0.461
std. Shimmer Sent. 8 −0.403 std. BBE16_on Sent. 6 −0.507 avg. dental_post_act Sent. 6 −0.457
std. BBE14_on Sent. 6 −0.504 avg. back_post_act DDK3 −0.456
std. BBE10_on Sent. 6 −0.502 std. anterior_post DDK2 0.455
std. BBE15_on Sent. 6 −0.499 std. consonantal_post Sent. 6 −0.454
8-35
std. BBE9_on Read text −0.496 avg. anterior_post DDK2 0.454
std. BBE14_on Read text −0.494 std. back_post Sent. 4 −0.451
std. BBE13_on Sent. 7 −0.493 std. vocalic_post Sent. 6 −0.449
std. BBE14_on Sent. 10 −0.493 avg. vocalic_post_act Sent. 6 −0.448
std. BBE14_on Sent. 7 −0.493 avg. vocalic_post_act Sent. 7 −0.446
std. BBE10_on DDK3 −0.488 std. back_post Read text −0.441

std. BBE11_on Read text −0.487 avg. stop_post_act Read text −0.438
avg: average; std: standard deviation.
phonation features shows that long-term perturbation features such as APQ and
PPQ are the most correlated with the m-FDA scale, especially those extracted from
the DDK exercises and the read sentences. The ranking of the articulation features
corresponds to the standard deviation of different Bark-band energies extracted
from the onset transitions in the read sentences and the read text. Negative
correlations indicate that low variability of energy in the transitions corresponds
to high m-FDA scores of the speakers. This fact confirms the existing correlation
between the variability of the energy distribution when patients start the vibration of
the vocal folds with the dysarthria severity, which was explored in related studies
[12, 25, 136]. Besides articulation features, the phonological posteriors show high
correlation with the m-FDA scores, especially those related to the pronunciation of
the back vowels. Other phonological features correlated with the m-FDA scores
include those based on anterior, dental, and stop sounds, which have also been
highlighted as relevant aspects to evaluate speech deficits of PD patients [20].
The same models created for phonation, articulation, prosody, and phonological,
are used to train an SVR algorithm to predict the m-FDA scores of the speakers.
Results are shown in table 8.13, and include the Spearman’s correlation coefficient
obtained between the real and predicted scores. The predicted values obtained with
the Fusion correspond to the median of the predictions obtained with regressors
Table 8.13. Spearman’s correlation between real and predicted m-FDA scores.
Feature sets
Exercise Phonation Articulation Prosody Phonological Fusion
DDK1 0.418 0.406 0.434 0.501 0.545

DDK2 0.462 0.515 0.365 0.540 0.579
DDK3 0.343 0.464 0.146 0.479 0.510
DDK4 0.427 0.487 0.269 0.396 0.527
DDK5 0.392 0.585 0.296 0.460 0.610
DDK6 0.315 0.465 0.400 0.504 0.554
Read text 0.370 0.428 0.341 0.432 0.531
Monologue 0.294 0.496 0.051 0.478 0.514
Sentence 1 0.278 0.451 0.279 0.353 0.436
Sentence 2 0.348 0.409 0.413 0.309 0.468
Sentence 3 0.263 0.427 0.486 0.429 0.562
Sentence 4 0.218 0.367 0.341 0.348 0.412
Sentence 5 0.193 0.446 0.180 0.413 0.457
Sentence 6 0.444 0.520 0.376 0.406 0.537
Sentence 7 0.335 0.431 0.284 0.447 0.489
Sentence 8 0.381 0.330 0.360 0.427 0.503
Sentence 9 0.461 0.422 0.352 0.518 0.577
Sentence 10 0.385 0.404 0.328 0.326 0.456
Fusion 0.563 0.557 0.530 0.567 0.602
DDK: diadohokinetic task.
8-36
trained with each feature set and speech task separately. The results obtained per
task show that the combination of the different feature sets is the most accurate to
predict m-FDA scores. At the same time, the combination of the predictions per task
shows also to improve the correlations. The highest correlation is obtained with the
combination of all feature sets and speech tasks (ρ = 0.602), which makes sense
because in such a configuration all of the available information is used. On the other
hand, the individual analysis per feature set and task shows that the most
suitable speech dimensions to predict the dysarthria level of the patients are
Articulation and Phonological.
Figure 8.11 shows the best result in a real versus predicted plot. The displayed
result corresponds to the fusion of all feature sets and speech tasks. The
figure indicates the Spearman’s and Pearson’s correlation values and the median
absolute error (MAE). Although the result is satisfactory, other regression strategies
can be considered to improve the correlation values. Additionally, speaker models
based on i-vectors or Gaussian mixture models–universal background models
(GMM–UBM) can also be considered to improve the results, as is shown in other
studies [12, 55].
8.7.2 Experiments with gait signals

The features sets introduced in section 8.4.2 are used to perform the automatic
classification between PD patients and HC subjects and to predict the neurological
state of the patients according to MDS-UPDRS-III score. The participants included
in the extended version of the PC-GITA corpus are considered for the experiments.
Figure 8.11. Best results displayed in a real versus predicted plot. r: Pearson’s correlation coefficient,
ρ : Spearman’s correlation coefficient, MAE: Median absolute error.
8-37

The results classifying PD patients and HC subjects using kinematic and non-linear
features are shown in table 8.14. The kinematic features are shown to be more
accurate in the 4 × 10 than in the 2 × 10 exercise, which was expected because it is a
longer exercise with more steps. Conversely, the results obtained with the non-linear
features do not differ much between the two tasks. Note also that the non-linear
features outperformed the results obtained with the kinematic features in all
scenarios. This can be explained because of the difficulties in segmenting the steps
when kinematic features are used. The fusion strategy did not provide high
improvements in the results. We believe that they could be improved with additional
feature sets included in the model. Other strategy to improve results could be to
consider others like the heel–toe tapping or the time up and go test.
Details of accuracy, sensitivity, specificity and UAR obtained with the fusion of
non-linear features are included in table 8.15. Note that the AUC value is relatively
high, however, when observing the details, the specificity is very low. This indicates
that additional feature sets and tasks need to be included in the analysis to improve
the results. Figure 8.12 shows the ROC curve and histograms with the fitted
probability density function resulting from the scores assigned by the classifier.
8.7.2.2 Prediction of the neurological state of the patients based on the

MDS-UPDRS-III score
The ranking of the kinematic and non-linear features with the highest correlation
coefficients is shown in table 8.16. Only features with correlations larger than 0.4 are
included. Note that similar results are obtained with kinematic and non-linear
features. The highest correlation is obtained with the standard deviation of the
duration of strides (ρ = 0.542). Other relevant kinematic features include average
Table 8.14. AUC values resulting from the classification of PD versus HC

subjects.
Feature set
Exercise Kinematic Non-linear Fusion
2 × 10 0.715 0.804 0.782

4 × 10 0.748 0.803 0.760
Fusion 0.719 0.810 0.795
Table 8.15. Best classification results obtained with non-linear features.
Fusion of non-linear features and tasks 76.1% 70.5% 89.1% 51.9% 0.810

8-38
(a) (b)
Figure 8.12. (a) ROC curve of the best classification result with gait signals; (b) distribution of the classification
scores.
Table 8.16. Gait features with the highest Spearman’s correlation between real and predicted MDS-UPDRS-
III scores.
Kinematic Non-linear
Feature Task ρ Feature Task ρ
1 std. stride duration 4 × 10 0.542 HE Gyr. transversal plane 2 × 10 −0.514

2 avg. velocity 2 × 10 −0.508 HE Gyr. transversal plane 4 × 10 0.491
3 avg. stride distance 2 × 10 −0.506 DFA Gyr. sagittal plane 2 × 10 −0.456
4 avg. stance duration 2 × 10 0.492 DFA Gyr. frontal plane 2 × 10 −0.455
5 avg. velocity 4 × 10 −0.451 DFA Acc. frontal plane 2 × 10 −0.446
6 avg. stride distance 4 × 10 −0.443 SE Acc. sagittal plane 4 × 10 −0.421
7 avg. stride duration 2 × 10 0.424 HE Gyr. frontal plane 2 × 10 −0.417
8 std. stride duration 2 × 10 0.401 LLE Acc. sagittal plane 4 × 10 0.416
avg: average; std: standard deviation; Acc: accelerometer; Gyr: gyroscope.
velocity, average stance duration, and average distance of strides. In the non-linear
feature set the highest correlation is obtained with the HE of the gyroscope in the
transversal plane (ρ = −0.514). Other relevant non-linear features that exhibit high
correlations are DFA extracted from several axis, HE computed also in several axis,
and SE.
Kinematic and non-linear features were also combined to train the SVR to predict
the MDS-UPDRS-III score of the patients. Results are shown in table 8.17. The
highest correlation is obtained with the 4 × 10 task (ρ = 0.559), which indicates that
the information extracted from both feature sets is complementary and allows one to
improve the accuracy to evaluate the neurological state of the patients.
8-39
Table 8.17. Spearman’s correlation between real and predicted MDS-UPDRS-III scores.
Feature set
Task Kinematic Non-linear Fusion
2 × 10 0.376 0.374 0.425

4 × 10 0.296 0.539 0.559
Fusion 0.343 0.511 0.504
Figure 8.13. Best results displayed in a real versus predicted plot. r: Pearson’s correlation coefficient,
ρ: Spearman’s correlation coefficient, MAE: Median absolute error.
The real versus predicted plot is shown in figure 8.13. This result corresponds to
the fusion of all feature sets extracted from the 4 × 10 task. Other regression
strategies could be considered to improve the correlation values. Gait features can
also be used to predict only those items of the MDS-UPDRS-III score related to the
movement of lower limbs, but this approach is out of the scope of this chapter.
8.7.3 Experiments with handwriting signals

The feature sets described in section 8.4.3 are used to classify PD patients and HC
subjects and to predict the neurological state of the patients according to the MDS-
UPDRS-III score assigned by the expert neurologist.

The results are presented in terms of AUC values. The kinematic features were
extracted from signals of all tasks but the geometric ones were only extracted from
the guided spiral. Table 8.18 includes the AUC values per task, feature set and also
the fusion, which was performed based on the most voted decision per task. Note
that the fusion of the different tasks highly improves the results (up to
AUC = 0.934). Kinematic features exhibited the best results when the tasks are
8-40
Table 8.18. AUC values resulting from the classification of PD versus HC subjects.
Feature set
Task Kinematic Geometric Fusion
Alphabet 0.834 — —
Circle 0.807 — —
Guided circle 0.800 — —
Cube 0.834 — —
Free writing 0.734 — —
Cursive letter l 0.757 — —
Cursive letter m 0.738 — —
Name 0.772 — —
Digits 0.806 — —
Rectangles 0.804 — —
Rey–Osterrieth 0.806 — —
Signature 0.711 — —
Free spiral 0.726 — —
Guided spiral 0.738 0.559 0.641
Fusion 0.934 — 0.930
Table 8.19. Best classification results obtained with kinematic features.
Fusion of kinematic features and tasks 88.7% 88.1% 88.9% 87.5% 0.934
considered separately. The most relevant handwriting tasks seem to be alphabet,

cube, digits, and the Rey–Osterrieth. Finally, note that the geometric features
proposed in [110] are not accurate individually to model the presence of the disease.
However, they seem to be complementary to the kinematic features. This observa-
tion was also highlighted in [110].
Details of the result obtained with the fusion of tasks and using the kinematic
features are shown in table 8.19. Accuracy, UAR, sensitivity, and specificity are also
included. Note that the UAR is 2.6% above the one obtained with speech signals (see
table 8.11), which might be expected because handwriting symptoms are among the
most visible in Parkinson’s patients. At the same time, the ROC curve in
figure 8.14(a) shows the ROC curve, and the histograms with the fitted probability
density functions in figure 8.14(b) show the scores assigned by the classifier to predict
each sample of the database.
8-41
(a) (b)
Figure 8.14. (a) ROC curve of the best classification result with handwriting signals; (b) distribution of the
classification scores.
8.7.3.2 Prediction of the neurological state of the patients based on the

MDS-UPDRS-III score
Spearman’s correlation coefficients between real and predicted MDS-UPDRS-III
scores were computed considering models obtained with kinematic features. The
ranking of features with ρ values larger than 0.4 are shown in table 8.20. Note that
largest correlation coefficients are obtained with the cursive letters l and m, which
confirms previous findings reported in the literature [104]. Additionally, the feature
with the largest correlation value is the skewness of the velocity of the in-air
trajectory (the distance between the pen and the tablet’s surface). This result
highlights the relevance of in-air movements as a revealing biomarker of PD
dysgraphia, which has been explored in other studies [104].
The feature sets were combined and used to train an SVR algorithm to predict the
MDS-UPDRS-III score. The results are shown in table 8.21. Note that better results
are obtained with individual features in specific handwriting tasks. This likely
indicates that the features considered here are not complementary to create models
for the automatic evaluation of the neurological state of the patients. The results
obtained with the cursive letters m, l, and the guided spiral, with individual or
combined feature sets, suggest that those tasks are the most suitable to model
handwriting impairments in PD patients. Additional feature sets and regression
strategies should be explored in further studies to improve the results.
8.7.4 Fusion of modalities

With the aim of exploring multi-modal strategies to perform the automatic
classification of PD and HC subjects, this section presents the results obtained in
experiments where a late-fusion approach is applied. This approach consists in
considering the scores of the classification experiments performed with each
8-42
Table 8.20. Handwriting features with the highest Spearman’s correlation between real and predicted MDS-
UPDRS-III scores.
Kinematic
Feature Task ρ
skw. vel. in-air Cursive letter m −0.541

std. acc. azimuth Cursive letter l −0.489
skw. trajectory Name −0.460
std. vel. azimuth Cursive letter l −0.441
skw. vel. in-air Cursive letter l −0.437
kurtosis x-axis Cursive letter l 0.433
kurtosis vel. trajectory Cursive letter m 0.430
kurtosis y-axis Cursive letter m 0.428
skw. trajectory Cursive letter l −0.423
std. y-axis Cursive letter l −0.423
skw. pressure derivative Guided circle 0.423
avg. vel. trajectory Digits 0.420
std. x-axis Cursive letter l −0.408
kurtosis x-axis Cursive letter m 0.406
kurtosis angle trajectory Alphabet 0.402
kurtosis acc. x-axis Cursive letter l −0.400
avg: average; std: standard deviation; skw: skewness; vel: velocity; acc: acceleration.
Table 8.21. Spearman’s correlations between real and predicted MDS-UPDRS-III scores.
Feature set
Task Kinematic Geometric Fusion
Alphabet −0.096 — —
Guided circle 0.044 — —
Cube 0.164 — —
Free writing 0.156 — —
Cursive letter l 0.323 — —
Cursive letter m 0.356 — —
Name −0.260 — —
Digits 0.219 — —
Rectangles 0.223 — —
Rey–Osterrieth 0.059 — —
Signature −0.230 — —
Free spiral 0.215 — —
Guided spiral 0.346 0.175 0.341
Fusion 0.222 — 0.236
8-43
Table 8.22. Classification of PD and HC subjects when information of speech, handwriting, and gait is
combined.
Multi-modal approach 87.5% 88.0% 87.1% 88.7% 0.943

(a) (b)
Figure 8.15. (a) ROC curve of the best classification results with speech, handwriting, and gait; (b) distribution
of the classification scores.
bio-signal separately and combining them into a single model. The results of the
fusion are shown in table 8.22. Note that the results are better than those obtained
with each modality individually. For instance, the AUC value increased in 1.7%
compared to the one obtained with speech features (see table 8.11), 1.0% compared
to the one obtained with handwriting (see table 8.19), and 15.3% compared to the
one obtained with gait (see table 8.15). It is worth highlighting the fact that these
results are robust and stable, which means there is almost no difference between
sensitivity and specificity, therefore the model is equally robust at detecting healthy
people and Parkinson’s patients.
The ROC curve and the scores of the predictions are shown in figure 8.15. Note
that the scores are more separated than those displayed when individual modalities
were considered. Finally, the three-dimensional scatter plot included in figure 8.16
shows the separability of the classification scores obtained with each modality. Note
that despite the overlap shown in figure 8.15, the scatter-plot shows a clear
separation between most of the HC subjects and PD patients.
8-44
Figure 8.16. Discrimination between PD and HC subjects combining information from speech, handwriting
and gait.
8.8 Discussion
This chapter provided the current state of technological advances of methods and
tools to assess and evaluate motor impairments of PD patients. Three bio-signals are
considered: speech, gait, and handwriting. Three main experiments were performed:
(1) classification of PD patients and HC subjects, considering information extracted
from speech, gait, and handwriting individually, and the combination of them;
(2) the assessment of the dysarthria severity of the patients based on the recently
introduced m-FDA scale; and (3) the prediction of the neurological state of the
patients based on the MDS-UPDRS-III scale considering information extracted
from handwriting and gait. We showed the relevance of different modalities and
their combination to improve the diagnosis and evaluation of the neurological state
of the patients. Different results with different bio-signals were expected because the
disease manifestation and progression differ from one patient to another. For
instance, a patient may have problems in the lower and upper limbs, but speak as a
completely healthy person. Conversely, one patient with normal handwriting could
exhibit deficits in speech and/or gait. We are aware of the fact that further research,
with a larger sample of patients, is necessary to lead to more conclusive results.
However, we think that the results presented here are a step forward to the
development of non-intrusive methods, useful in clinical practice, to diagnose and
monitor the state of PD patients. Additional fusion strategies to allow a more robust
classification and monitoring of PD patients should also be considered in further
research [148, 151, 152].
The proposed methods for speech assessment are suitable and accurate to
discriminate between PD patients and HC subjects, and to predict their dysarthria
severity. The combination of features extracted from different dimensions of speech
such as phonation, articulation, prosody, and phonological, was necessary to
improve the results and to obtain better estimates of the dysarthria level of the
patients. This can be explained because, as in the case of how the limbs and muscles
are affected, the speech of each PD patient is affected differently. There might be
8-45
patients with severe articulation deficits but without phonation or prosody impair-
ments. On the other hand, there might be patients without articulation deficits, but
with prosody impairments that may be caused by non-motor symptoms like
depression, which also affects PD patients. Apart from the combined analysis, the
results indicate that the most accurate individual feature set to evaluate the speech
deficits of PD patients were those that model the articulation and the phonological
aspects of speech production. These results confirm that one of the most relevant
speech defects in PD patients is the imprecise articulation and the mispronunciation
of consonants.
Regarding the assessment of the handwriting impairments of PD patients, the
results indicate that the kinematic analysis is the most suitable and accurate to
discriminate between PD and HC subjects. The relevance analysis of features,
according to the correlation coefficients obtained between MDS-UPDRS-III score
and the extracted features also indicated that measures related to speed, pressure,
and acceleration of the strokes are the most suitable. The relevance of the in-air
movement analysis was also shown in the results. We think that further research
should consider more detailed analyses with this feature to improve the results.
Additionally, when evaluating the suitability of the handwriting tasks performed by
the patients, the Archimedean spiral and the cursive letters m and l are the best ones
to assess handwriting impairments of patients. The evaluation of motor deficits with
cursive letters seem to be the most accurate and reliable because it involves the
writing of the same character (which means less cognitive load) scaled in amplitude,
and requires accurate control in the strokes.
The results obtained with gait signals showed that non-linear features are better
than the kinematic ones to classify PD patients and HC subjects and to predict the
neurological state of the patients. This fact may be explained due to algorithmic
errors in the segmentation of individual steps. This aspect is highlighted in the
literature as a current challenge in automatic gait assessment and requires the
development of more robust segmentation strategies to guarantee a reliable
estimation of kinematic features [153].
Finally, the results obtained with the combination of information extracted from
the three bio-signals show that this approach is the most convenient. We are aware
of the difficulties that this approach could mean during the recording sessions,
however, the fast development of wearable sensors will make this process much
easier; the patients will not even notice that they are being evaluated and monitored
in the near future. For instance, mobile applications connected with wearable
sensors seem to be the easiest and most convenient approach.
8.9 Future trends in PD analysis

The methods and techniques described in this chapter can be extended to other
applications. Due to the variation of the disease manifestation and progression
among patients, it is necessary to monitor them individually and continuously.
Speech production, and the movement of upper and lower limbs should be included
in the analyses. Novel strategies, based on user-specific models should be proposed
8-46
and evaluated in further research for a continuous monitoring of the disease

progression. These models may help clinicians to make more accurate and timely
decisions about the medication and therapy. Our team is currently making progress
in this aspect with the release of Apkinson [154], which is an Android application for
the continuous assessment of PD patients. It is designed to capture speech and
movement signals using a smartphone. Patient’s activities are recorded unobtru-
sively and on a daliy basis and the patients receive feedback about their current state.
The described methods could also be potentially used to detect prodromal stages
of the disease, which would benefit the development of future neuroprotective
therapies [155]. There is evidence showing that the detection of prodromal stages of
PD is possible from speech [23] and gait [156]. The main difficulty of these kinds of
studies is to find patients in pre-clinical stages, i.e. before the disease appears or is at
least observable by clinicians. Once the target group is found, it is required to start
the monitoring of patients over time in order to understand which are the patterns
that become abnormal when early signs of the disease appear. Given the fact that the
north of Antioquia (a state in Colombia, south America) is perhaps the area with the
highest prevalence of genetic Parkinson’s disease [157]. Our research team in
Medellín, Colombia has been working on the recording of different bio-signals
from pre-clinical genetic subjects (people who have a gene mutation responsible for
producing PD but with no clinical signs of the disease). We hope to find promising
results in the future. Another population that can be considered to study prodromal
stages of the disease includes patients suffering from rapid eye movement (REM)
sleep behavior disorders (RBD). According to recent studies, there is a high
probability for these people to develop Parkinson’s disease within their next twelve
years of life [23, 158]. Another potential application of the methods described in this
chapter could be the discrimination of PD among other neurological disorders with
similar symptoms such as Huntington’s disease or essential tremor. There is evidence
that supports the convenience and necessity of these kinds of applications, for
instance with speech [159] and gait signals [160].
On the other hand, specific future trends related with the analysis of speech in PD
patients include a more detailed evaluation of phonological and phonetic cues with
the aim to model specific speech deficits of the patients. For instance, it is clinically
useful to detect whether the patients have more problems in pronouncing plosives,
fricatives, or nasal sounds. This would help the expert phoniatrician to select a more
specific and focused therapy for the PD patients. In addition to the articulation
features described in this chapter, other features might be explored with more detail
in further studies, for instance those based on the VOT, which have shown to be
suitable to model specific speech impairments exhibited by PD patients [44]. Another
important aspect to be considered in further research is the evaluation of speech
deficits in different languages. For instance, to evaluate whether the production of
consonantal sounds is affected differently depending on the language spoken by the
patient. This is because there is theoretical evidence suggesting that native speakers
of languages like German and Czech might be more capable of producing stop
consonants and fricatives than Spanish or French native speakers [161].
8-47
Regarding handwriting analysis, one important aspect to be considered is that

most of the studies only include kinematic and pressure features, which limits the
scope of the models. Additional features should be proposed to assess other aspects
such as fluency or size. For instance, neuromotor features based on the Sigma
lognormal model could be suitable to complement kinematic features, and to
improve the results. On the other hand, most of the studies are focused on the
classification of PD patients versus HC subjects. There are few studies that cover
other aspects like the prediction of the neurological state, the assessment of upper-
limbs impairments, and the evaluation of effects of the dopaminergic medication.
There is also a need for longitudinal studies that help in understanding and
monitoring the disease progression. Finally, most of the studies consider only data
from online handwriting, where the patients perform exercises in digital tablets. This
could be uncomfortable and non-natural for patients, and may introduce some bias
in the results. Additional studies that compare online and offline handwriting should
be proposed.
Regarding gait analysis of PD patients, there are several challenges that need to
be addressed. Most of the proposed methods and systems are highly variable among
them. There are no standard procedures indicating for instance the place where the
wearable sensors have to be attached, the tasks to be performed by the patients, or
the features to be extracted. Additionally, some of the systems considered in the
literature require multiple sensors located on different parts of the body, which is
impractical for many clinical, and especially home-based applications [162]. On the
feature extraction aspect, except for the freezing index [80], most of the current
approaches are focused on extracting features based only on the kinematic analysis
of gait, e.g. amplitude, frequency components, cadence, step length, among others,
which are not specifically designed for PD. This fact makes it still unclear which gait
features are the most informative to assess gait in PD [162]. Additional feature sets
should be explored to improve results in classification and regression. For instance,
the freezing index and other harmonic features derived from the energy content in
different frequency bands, should be further explored because they could be
complementary to the kinematic and non-linear features described in this chapter.
On the other hand, the use of modern learning strategies like those based on deep
learning methods have not been sufficiently explored so far in the context of PD.
Besides the classical feature extraction strategies addressed in this chapter, the use
of novel deep learning approaches should be explored with more detail in further
research to improve the results in the detection and monitoring of PD patients. For
instance, the use of transfer learning strategies, which consists in the use of a neural
network trained for one task to be used in a different domain could improve the
accuracy of the models. In general, the assessment of motor impairments in PD
patients may benefit from the current advances in computer vision and speech
processing.
Finally, although this chapter only covered the assessment of motor impairments
of the patients, the evaluation of non-motor symptoms such as depression, anxiety,
or cognitive decline is also important and can be automated by using state-of-the-art
technologies of speech recognition and natural language processing. These
8-48
techniques have been successfully applied in patients with other neurodegenerative

diseases such as Alzheimer’s or other types of dementia [163], and have been recently
considered to evaluate language aspects of PD patients [164, 165].
Acknowledgments
This project received funding from CODI from University of Antioquia by Grant
No. 2017–15530, and Juan Camilo Vásquez-Correa is under grants of the EU
Horizon 2020 research and innovation programme under the Marie Sklodowska-
Curie Grant Agreement No. 766287.
References
[1] Parkinson J 1817 An Essay on the Shaking Palsy (London, UK: Whittingham and
Rowland)
[2] Jankovic J 2008 Parkinson’s disease: clinical features and diagnosis J. Neurol. Neurosurg.
Psychiatry 79 368–76
[3] Vidailhet M, Bourdain F and Trocello J 2005 Medical history and physical examination in
Parkinsonian syndromes Current Clinical Neurology: Atypical Parkinsonian Disorders ed I
Litvan (New Jersey: Humana) ch 10
[4] Hoehn M and Yahr M 1967 Parkinsonism: onset, progression, and mortality Neurology 17
427–42
[5] Goetz C et al 2004 Movement Disorder Society Task Force Report on the Hoehn and Yahr
staging scale: status and recommendations. The movement disorder society task force on
rating scales for Parkinson’s disease Mov. Disord. 199 1020–8
[6] Fahn S, Elton R and UPDRS program members 1987 Recent Developments in Parkinson’s
Disease (Florham Park, NJ: Macmillan Healthcare Information)
[7] Movement Disorder Society Task Force on Rating Scales for Parkinson’s Disease 2003 The
Unified Parkinson’s Disease Rating Scale (UPDRS): status and recommendations Mov.
Disord. 187 738–50
[8] Goetz C et al 2008 Movement Disorder Society-sponsored revision of the Unified
Parkinson’s Disease Rating Scale (MDS-UPDRS): scale presentation and clinimetric
testing results Mov. Disord. 23 2129–70
[9] Ramig L, Mead C, Scherer Y, Larson K and Kohler D 1988 Voice therapy and Parkinson’s
disease: a longitudinal study of efficacy Proc. of the Clinical Dysarthria Conf.
[10] de Swart B J, Willemse S C, Maassen B and Horstink M W 2003 Improvement of voicing in
patients with Parkinson’s disease by speech therapy Neurology 60 498–500
[11] Enderby P M and Palmer R 2008 FDA-2: Frenchay Dysarthria Assessment: Examiner’s
Manual (Austin, TX: Pro-ed)
[12] Vásquez-Correa J C, Orozco-Arroyave J R, Bocklet T and Nöth E 2018 Towards
an automatic evaluation of the dysarthria level of patients with Parkinson’s disease
J. Commun. Disord. 76 21–36
[13] Orozco-Arroyave J R, Arias-Londoño J D, Vargas-Bonilla J F, Gonzalez-Rátiva M C and
Nöth E 2014 New Spanish speech corpus database for the analysis of people suffering from
Parkinson’s disease Language Resources and Evaluation Conf. (LREC) pp 342–7
[14] LeCun Y, Bengio Y and Hinton G 2015 Deep learning Nature 521 436
[15] Hornykiewicz O 1998 Biochemical aspects of Parkinson’s disease Neurology 51 S2–9
8-49
[16] Baker K K, Ramig L O, Luschei E S and Smith M E 1998 Thyroarytenoid muscle activity
associated with hypophonia in Parkinson disease and aging Neurology 516 1592–8
[17] Tsanas A, Little M A, McSharry P E and Ramig L O 2010 Accurate telemonitoring of
Parkinson’s disease progression by noninvasive speech tests IEEE Trans. Biomed. Eng. 57 884–93
[18] Spencer K A and Rogers M A 2005 Speech motor programming in hypokinetic and ataxic
dysarthria Brain Lang. 94 347–66
[19] Skodda S, Grönheit W and Schlegel U 2011 Gender-related patterns of dysprosody in
Parkinson disease and correlation between speech variables and motor symptoms J. Voice
25 76–82
[20] Tykalova T, Rusz J, Klempir J, Cmejla R and Ruzicka E 2017 Distinct patterns of
imprecise consonant articulation among Parkinson’s disease, progressive supranuclear
palsy and multiple system atrophy Brain Lang. 165 1–9
[21] Moretti R et al 2003 Speech Initiation Hesitation’ following subthalamic nucleus stim-
ulation in a patient with Parkinson’s disease Eur. Neurol. 49 251–3
[22] Rusz J, Cmejla R, Ruzickova H and Ruzicka E 2011 Quantitative acoustic measurements
for characterization of speech and voice disorders in early untreated Parkinson’s disease
J. Acoust. Soc. Am. 129 350–67
[23] Hlavnicka J, Cmejla R, Tykalova T, Sonka K, Ruzicka E and Rusz J 2017 Automated
analysis of connected speech reveals early biomarkers of Parkinson’s disease in patients
with rapid eye movement sleep behaviour disorder Sci. Rep. 7 1–13
[24] Pell M D, Cheang H S and Leonard C L 2006 The impact of Parkinson’s disease on vocal-
prosodic communication from the perspective of listeners Brain Lang. 97 123–34
[25] Orozco-Arroyave J R and Vásquez-Correa J C et al 2018 NeuroSpeech: An open-source
software for Parkinson’s speech analysis Digit. Signal Process. 77 207–21
[26] Orozco-Arroyave J 2016 Analysis of Speech of People with Parkinson’s Disease 1st edn
(Berlin, Germany: Logos-Verlag)
[27] Bocklet T, Steidl S, Nöth E and Skodda S 2013 Automatic evaluation of Parkinson’s
speech—acoustic, prosodic and voice related cues Proc. of INTERSPEECH pp 1149–53
[28] Moro-Velázquez L et al 2018 Analysis of speaker recognition methodologies and the
influence of kinetic changes to automatically detect Parkinson’s Disease Appl. Soft Comput.
62 649–66
[29] Tanaka Y, Nishio M and Niimi S 2011 Vocal acoustic characteristics of patients with
Parkinson’s disease Folia Phoniatr. logop. 63 223–30
[30] Galaz Z et al 2018 Changes in phonation and their relations with progress of Parkinson’s
disease Appl. Sci. 8 1–18
[31] Naranjo L, Pérez C J, Campos-Roca Y and Martn J 2016 Addressing voice recording
replications for Parkinson’s disease detection Expert Syst. Appl. 46 286–92
[32] Villa-Cañas T, Arias-Londoño J D, Orozco-Arroyave J R, Vargas-Bonilla J F and Nöth E
2015 Low-frequency components analysis in running speech for the automatic detection of
Parkinson’s disease Proc. of INTERSPEECH 100–4
[33] Belalcázar-Bolaños E A and Orozco-Arroyave J R et al 2016 Glottal flow patterns analyses
for Parkinson’s disease detection: acoustic and nonlinear approaches Int. Conf. on Text,
Speech, and Dialogue 400–7
[34] Tu M, Berisha V and Liss J 2017 Objective assessment of pathological speech using
distribution regression IEEE Int. Conf. on Acoustics, Speech and Signal Processing
(ICASSP) (Piscataway, NJ: IEEE), pp 5050–4
8-50
[35] Sakar C O et al 2019 A comparative analysis of speech signal processing algorithms for
Parkinson’s disease classification and the use of the tunable Q-factor wavelet transform
Appl. Soft Comput. 74 255–63
[36] Sapir S, Ramig L, Spielman J and Fox C 2010 Formant Centralization Ratio (FCR): A
proposal for a new acoustic measure of dysarthric speech J. Speech Lang. Hear. Res. 53 114–33
[37] Rusz J and Cmejla R et al 2013 Imprecise vowel articulation as a potential early marker of
Parkinson’s disease: Effect of speaking task J. Acoust. Soc. Am. 134 2171–81
[38] Logemann J A, Fisher H B, Boshes B and Blonsky E R 1978 Frequency and cooccurrence
of vocal tract dysfunctions in the speech of a large sample of Parkinson patients J. Speech
Hear. Disord. 43 47–57
[39] Ackermann H and Ziegler W 1991 Articulatory deficits in Parkinsonian dysarthria: an
acoustic analysis J. Neurol. Neurosurg. Psychiatry 54 1093–8
[40] Walsh B and Smith A 2012 Basic parameters of articulatory movements and acoustics in
individuals with Parkinson’s disease Mov. Disord. 27 843–50
[41] Kim Y, Weismer G, Kent R D and Duffy J R 2009 Statistical models of F2 slope in relation
to severity of dysarthria Folia Phoniatr. Logop. 61 329–35
[42] Forrest K, Weismer G and Turner G S 1989 Kinematic, acoustic, and perceptual analyses
of connected speech produced by Parkinsonian and normal geriatric adults J. Acoust. Soc.
Am. 85 2608–22
[43] Novotn M and Rusz J et al 2014 Automatic evaluation of articulatory disorders in
Parkinson’s disease IEEE/ACM Trans. Audio Speech Lang. Process. 22 1366–78
[44] Montaña D, Campos-Roca Y and Pérez C J 2018 A Diadochokinesis-based expert system
considering articulatory features of plosive consonants for early detection of Parkinson’s
disease Comput. Methods Prog. Biomed. 154 89–97
[45] Godino-Llorente J I, Shattuck-Hufnagel S, Choi J Y, Moro-Velázquez L and Gómez-
Garca J A 2017 Towards the identification of Idiopathic Parkinson’s disease from the
speech: new articulatory kinetic biomarkers PLoS One 12 e0189583
[46] Moro-Velazquez L et al 2019 A forced gaussians based methodology for the differential
evaluation of Parkinson’s disease by means of speech processing Biomed. Signal Process.
Control 48 205–20
[47] Galaz Z and Mekyska J et al 2016 Prosodic analysis of neutral, stress-modified and rhymed
speech in patients with Parkinson’s disease Comput. Methods Prog. Biomed. 127 301–17
[48] Smith K M, Williamson J R and Quatieri T F 2017 Vocal markers of motor, cognitive, and
depressive symptoms in Parkinson’s disease Int. Conf. on Affective Computing and
Intelligent Interaction (ACII) (Piscataway, NJ: IEEE), 71–8
[49] Barnish M S and Whibley D et al 2016 Roles of cognitive status and intelligibility in
everyday communication in people with Parkinson’s disease: A systematic review J.
Parkinson’s Dis. 6 453–62
[50] Dimauro G and Di-Nicola V et al 2017 Assessment of speech intelligibility in Parkinson’s
disease using a speech-to-text system IEEE Access 5 22199–208
[51] Orozco-Arroyave J R and Vásquez-Correa J C et al 2016 Towards an automatic
monitoring of the neurological state of the Parkinson’s patients from speech IEEE Int.
Conf. on Acoustics, Speech and Signal Processing (ICASSP) 6490–4
[52] Garcia N, Orozco-Arroyave J R, D’Haro L F, Dehak N and Nöth E 2017 Evaluation of
the neurological state of people with Parkinson’s disease using i-vectors Proc. of
INTERSPEECH 299–303
8-51
[53] Dehak N, Kenny P, Dehak R, Dumouchel P and Ouellet P 2011 Front–end factor analysis
for speaker verification IEEE Trans. Audio Speech Lang. Process. 19 788–98
[54] Laaridh I, Kheder W B, Fredouille C and Meunier C 2017 Automatic prediction of speech
evaluation metrics for dysarthric speech Proc. of INTERSPEECH pp 1834–8
[55] Arias-Vergara T, Vasquez-Correa J C, Orozco-Arroyave J R and Nöth E 2018 Speaker
models for monitoring Parkinson’s disease progression considering different communica-
tion channels and acoustic conditions Speech Commun. 101 11–25
[56] Schuller B, Steidl S, Batliner A, Hantke S, Hönig F, Orozco-Arroyave J R, Nöth E, Zhang
Y and Weninger F 2015 The INTERSPEECH 2015 computational paralinguistics
challenge: nativeness, Parkinson’s & eating condition Proc. of INTERSPEECH pp 478–82
[57] Grósz T and Busa-Fekete R et al 2015 Assessing the degree of nativeness and Parkinson’s
condition using Gaussian processes and deep rectifier neural networks Proc. of
INTERSPEECH pp 1339–43
[58] Vásquez-Correa J C, Orozco-Arroyave J R and Nöth E 2017 Convolutional neural network
to model articulation impairments in patients with Parkinson’s disease Proc. of
[59] Vásquez-Correa J C, Arias-Vergara T, Orozco-Arroyave J R and Nöth E 2018 A multitask
learning approach to assess the dysarthria severity in patients with Parkinson’s disease Proc.
of INTERSPEECH pp 456–60
[60] Tu M, Berisha V and Liss J 2017 Interpretable objective assessment of dysarthric speech
based on deep neural networks Proc. of INTERSPEECH pp 1849–53
[61] Yang Y, Lee Y, Cheng S, Lin P and Wang R 2008 Relationships between gait and dynamic
balance in early Parkinson’s disease Gait Posture 27 611–5
[62] Galna B, Lord S, Burn D J and Rochester L 2015 Progression of gait dysfunction in
incident Parkinson’s disease: impact of medication and phenotype Mov. Disord. 30
359–67
[63] Nutt J G, Bloem B R, Giladi N, Hallett M, Horak F B and Nieuwboer A 2011 Freezing of
gait: moving forward on a mysterious clinical phenomenon Lancet Neurol. 10 734–44
[64] Schaafsma J D et al 2003 Characterization of freezing of gait subtypes and the response of
each to levodopa in Parkinson’s disease Eur. J. Neurol. 10 391–8
[65] Moore S T, MacDougall H G and Ondo W G 2008 Ambulatory monitoring of freezing of
gait in Parkinson’s disease J. Neurosci. Methods 167 340–8
[66] Bacher M, Scholz E and Diener H C 1989 24 h continuous tremor quantification based on
EMG recording Electroencephalogr. Clin. Neurophysiol. 72 176–83
[67] Deuschl G, Krack P, Lauk M and Timmer J 1996 Clinical neurophysiology of tremor
J. Clin. Neurophysiol. 13 110–21
[68] Wenzelburger R et al 2000 Kinetic tremor in a reach-to-grasp movement in Parkinson’s
disease Mov. Disord. 15 1084–94
[69] Cleeves L, Findley L J and Gresty M 1987 Assessment of rest tremor in Parkinson’s disease
Adv. Neurol. 45 349–52
[70] Silva-De-Lima A L et al 2017 Freezing of gait and fall detection in Parkinson’s disease
using wearable sensors: a systematic review J. Neurol. 264 1642–54
[71] Rigas G et al 2012 Assessment of tremor activity in the Parkinson’s disease using a set of
wearable sensors IEEE Trans. Inf. Technol. Biomed. 16 478–87
[72] Klucken J and Barth J et al 2013 Unbiased and mobile gait analysis detects motor
impairment in Parkinson’s disease PLoS One 8 1–9
8-52
[73] Shull P B, Jirattigalachote W, Hunt M A, Cutkosky M R and Delp S L 2014 Quantified self
and human movement: a review on the clinical impact of wearable sensing and feedback for
gait analysis and intervention Gait Posture 40 11–9
[74] Oung Q W and Muthusamy H et al 2015 Technologies for assessment of motor disorders in
Parkinson’s disease: a review Sensors 15 21710–45
[75] Hannink J and Gaßner H et al 2017 Inertial sensor-based estimation of peak accelerations
during heel-strike and loading as markers of impaired gait patterns in PD patients Basal
Ganglia 8 1
[76] Xia Y, Gao Q and Ye Q 2015 Classification of gait rhythm signals between patients with
neuro-degenerative diseases and normal subjects: experiments with statistical features and
different classification models Biomed. Signal Process. Control 18 254–62
[77] Ren P et al 2016 Analysis of gait rhythm fluctuations for neurodegenerative diseases by phase
synchronization and conditional entropy IEEE Trans. Neural Syst. Rehabil. Eng. 24 291–9
[78] Parisi F and Ferrari G et al 2015 Body-sensor-network-based kinematic characterization
and comparative outlook of UPDRS scoring in leg agility, sit-to-stand, and gait tasks in
Parkinson’s disease IEEE J. Biomed. Health Inform. 19 1777–93
[79] Djurić-Jovičić M et al 2017 Selection of gait parameters for differential diagnostics of
patients with de novo Parkinson’s disease Neurol. Res. 39 853–61
[80] Kim H et al 2015 Unconstrained detection of freezing of Gait in Parkinson’s disease
patients using smartphone Int. Conf. of the IEEE Engineering in Medicine and Biology
Society (EMBC) (Piscataway, NJ: IEEE), pp 3751–4
[81] Ornelas-Vences C et al 2017 Fuzzy inference model evaluating turn for Parkinson’s disease
patients Comput. Biol. Med. 89 379–88
[82] Ornelas-Vences C et al 2019 Computer model for leg agility quantification and assessment
for Parkinson’s disease patients Med. Biol. Eng. Comput. 57 463–76
[83] Mazilu S, Hardegger M, Zhu Z, Roggen D, Troster G, Plotnik M and Hausdorff J M 2012
Online detection of freezing of gait with smartphones and machine learning techniques
Int. Conf. on Pervasive Computing Technologies for Healthcare (PervasiveHealth) pp 123–30
[84] Das S, Amoedo B, De la Torre F and Hodgins J 2012 Detecting Parkinson’s symptoms in
uncontrolled home environments: A multiple instance learning approach Int. Conf. of
the IEEE Engineering in Medicine and Biology Society (EMBC) (Piscataway, NJ: IEEE),
pp 3688–91
[85] Rezvanian S and Lockhart T 2016 Towards real-time detection of freezing of gait using
wavelet transform on wireless accelerometer data Sensors 16 1–11
[86] Sanchez-Perez L A et al 2018 Rest tremor quantification based on fuzzy inference systems
and wearable sensors Int. J. Med. Inform. 114 6–17
[87] Samà A et al 2018 Determining the optimal features in freezing of gait detection through a
single waist accelerometer in home environments Pattern Recogn. Lett. 105 135–43
[88] Aghanavesi S, Bergquist F, Nyholm D, Senek M and Memedi M 2020 Motion sensor-
based assessment of Parkinson’s disease motor symptoms during leg agility tests: results
from levodopa challenge IEEE J. Biomed. Health Inform. 24 111–9
[89] Sejdić E, Lowry K A, Bellanca J, Redfern M S and Brach J S 2014 A comprehensive
assessment of gait accelerometry signals in time, frequency and time-frequency domains
IEEE Trans. Neural Syst. Rehabil. Eng. 22 603–12
[90] Zach H et al 2015 Identifying freezing of gait in Parkinson’s disease during freezing
provoking tasks using waist-mounted accelerometry Parkinsonism Relat. Disord. 21 1362–6
8-53
[91] Prabhu P, Karunakar A, Anitha H and Pradhan N 2018 Classification of gait signals into
different neurodegenerative diseases using statistical analysis and recurrence quantification
analysis Pattern Recogn. Lett. https://doi.org/10.1016/j.patrec.2018.05.006
[92] Tripoliti E E et al 2013 Automatic detection of freezing of gait events in patients with
Parkinson’s disease Comput. Methods Programs Biomed. 110 12–26
[93] Pérez-Toro P A et al 2018 A non-linear dynamics approach to classify gait signals of
patients with Parkinson’s disease Workshop on Engineering Applications (Berlin: Springer),
pp 268–78
[94] Chomiak T, Xian W, Pei Z and Hu B 2019 A novel single-sensor-based method for the
detection of gait-cycle breakdown and freezing of gait in Parkinson’s disease J. Neural
Transm. 126 1029–36
[95] Bauby C E and Kuo A D 2000 Active control of lateral balance in human walking
J. Biomech. 33 1433–40
[96] Cuzzolin F and Sapienza M et al 2017 Metric learning for Parkinsonian identification from
IMU gait measurements Gait Posture 54 127–32
[97] Camps J et al 2018 Deep learning for freezing of gait detection in Parkinson’s disease
patients in their homes using a waist-worn inertial measurement unit Knowl.-Based Syst.
139 119–31
[98] Torvi V G, Bhattacharya A and Chakraborty S 2018 Deep domain adaptation to predict
freezing of gait in patients with Parkinson’s disease IEEE Int. Conf. on Machine Learning
and Applications (ICMLA) (Piscataway, NJ: IEEE), pp 1001–6
[99] Xia Y et al 2018 Evaluation of deep convolutional neural networks for detection of freezing
of gait in Parkinson’s disease patients Biomed. Signal Process. Control 46 221–30
[100] Smits E J et al 2014 Standardized handwriting to assess bradykinesia, micrographia and
tremor in Parkinson’s disease PLoS One 9 1–8
[101] Letanneux A, Danna J, Velay J, Viallet F and Pinto S 2014 From micrographia to
Parkinson’s disease dysgraphia Mov. Disord. 29 1467–75
[102] Rosenblum S, Samuel M, Zlotnik S, Erikh I and Schlesinger I 2013 Handwriting as an
objective tool for Parkinson’s disease diagnosis J. Neurol. 260 2357–61
[103] Drotár P, Mekyska J, Rektorová I, Masarová L, Smékal Z and Faundez-Zanuy M 2014
Analysis of in-air movement in handwriting: A novel marker for Parkinson’s disease
Comput. Methods Programs Biomed. 117 405–11
[104] Drotár P, Mekyska J, Rektorová I, Masarová L, Smékal Z and Faundez-Zanuy M 2016
Evaluation of handwriting kinematics and pressure for differential diagnosis of Parkinson’s
disease Artif. Intell. Med. 67 39–46
[105] Kotsavasiloglou C et al 2017 Machine learning-based classification of simple drawing
movements in Parkinson’s disease Biomed. Signal Process. Control 31 174–80
[106] Mucha J et al 2018 Fractional derivatives of online handwriting: a new approach of
Parkinsonic dysgraphia analysis 41st Int. Conf. on Telecommunications and Signal
Processing (TSP) (Piscataway, NJ: IEEE), pp 1–4
[107] Jerkovic V et al 2018 Analysis of on-surface and in-air movement in handwriting of subjects
with Parkinson’s disease and atypical parkinsonism Biomed. Eng./Biomed. Tech. 64 1–8
[108] Zham P, Arjunan S P, Raghav S and Kumar D K 2018 Efficacy of guided spiral drawing in
the classification of Parkinson’s disease IEEE J. Biomed. Health Inform. 22 1648–52
[109] Mucha J et al 2018 Identification and monitoring of Parkinson’s disease dysgraphia based
on fractional-order derivatives of online handwriting Appl. Sci. 8 1–18
8-54
[110] Rios-Urrego C, Vásquez-Correa J, Vargas-Bonilla J, Nöth E, Lopera F and Orozco-Arroyave

J 2019 Analysis and evaluation of handwriting in patients with Parkinson’s disease using
kinematic, geometrical, and non-linear features Comput. Methods Prog. Biomed. 173 43–52
[111] Impedovo D 2019 Velocity-based signal features for the assessment of Parkinsonian
handwriting IEEE Signal Process Lett. 26 632–6
[112] O’Reilly C and Plamondon R 2009 Development of a Sigma–Lognormal representation for
on-line signatures Pattern Recogn. 42 3324–37
[113] Sarbaz Y, Towhidkhah F, Mosavari V, Janani A and Soltanzadeh A 2013 Separating
Parkinsonian patients from normal persons using handwriting features J. Mech. Med. Biol.
13 1–8
[114] Pereira C R et al 2016 A new computer vision-based approach to aid the diagnosis of
Parkinson’s disease Comput. Methods Programs Biomed. 136 79–88
[115] Pereira C R et al 2018 Handwritten dynamics assessment through convolutional neural
networks: An application to Parkinson’s disease identification Artif. Intell. Med. 87 67–77
[116] Gallicchio C, Micheli A and Pedrelli L 2018 Deep echo state networks for diagnosis of
Parkinson’s disease 26th Eur. Symp. on Artificial Neural Networks (Belgium: i6doc.com),
pp 397–402
[117] Afonso L et al 2019 A recurrence plot-based approach for Parkinson’s disease identification
Future Gener. Comput. Syst. 94 282–92
[118] Moetesum M, Siddiqi I, Vincent N and Cloppet F 2019 Assessing visual attributes of
handwriting for prediction of neurological disorders—A case study on Parkinson’s disease
Pattern Recogn. Lett. 121 19–27
[119] Krizhevsky A, Sutskever I and Hinton G E 2012 Imagenet classification with deep
convolutional neural networks Advances in Neural Information Processing Systems
(Cambridge, MA: MIT), pp 1097–105
[120] Naseer A, Rani M, Naz S, Imran Razzak M, Imran M and Xu G 2020 Refining
Parkinson’s neurological disorder identification through deep transfer learning Neural
Comput. Appl. 32 839–54
[121] Orozco-Arroyave J R and Belalcázar-Bolaños E A et al 2015 Characterization methods for
the detection of multiple voice disorders: neurological, functional, and Laryngeal diseases
IEEE J. Biomed. Health Inform. 19 1820–28
[122] Vásquez-Correa J, Serra J, Orozco-Arroyave J, Vargas-Bonilla J and Nöth E 2017 Effect of
acoustic conditions on algorithms to detect Parkinson’s disease from speech IEEE Int. Conf. on
Acoustics, Speech and Signal Processing (ICASSP) (Piscataway, NJ: IEEE), pp 5065–9
[123] Zwicker E and Terhardt E 1980 Analytical expressions for critical-band rate and critical
bandwidth as a function of frequency J. Acoust. Soc. Am. 68 1523–5
[124] Vásquez-Correa J C, Klumpp P, Orozco-Arroyave J R and Nöth E 2019 Phonet: a tool
based on gated recurrent neural networks to extract phonological posteriors from speech
Proc. of INTERSPEECH pp 549–53
[125] Barth J et al 2011 Biometric and mobile gait analysis for early diagnosis and therapy
monitoring in Parkinson’s disease Int. Conf. of the IEEE Engineering in Medicine and
Biology Society 868–71
[126] Sadeghi H et al 2000 Symmetry and limb dominance in able-bodied gait: a review Gait
Posture 12 34–45
[127] Takens F et al 1981 Detecting strange attractors in turbulence Lect. Notes Math. 898
366–81
8-55
[128] Kennel M B, Brown R and Abarbanel H D 1992 Determining embedding dimension for
phase-space reconstruction using a geometrical construction Phys. Rev. A 45 3403–11
[129] Grassberger P and Procaccia I 2004 Measuring the strangeness of strange attractors The
Theory of Chaotic Attractors (Berlin: Springer), pp 170–89
[130] Rosenstein M T, Collins J J and De Luca C J 1993 A practical method for calculating
largest Lyapunov exponents from small data sets Physica D 65 117–34
[131] Hurst H E 1965 Long Term Storage: An Experimental Study (London: Constable)
[132] Lempel A and Ziv J 1976 On the complexity of finite sequences IEEE Trans. Inf. Theory 22
75–81
[133] Kaspar F and Schuster H G 1987 Easily calculable measure for the complexity of
spatiotemporal patterns Phys. Rev. A 36 842–8
[134] Travieso C, Alonso J, Orozco-Arroyave J R, Vargas-Bonilla J, Nöth E and Ravelo-Garcia
A G 2017 Detection of different voice diseases based on the nonlinear characterization of
speech signals Expert Syst. Appl. 82 184–95
[135] Skodda S, Visser W and Schlegel U 2011 Vowel articulation in Parkinson’s disease J. Voice
25 467–72
Skodda S, Visser W and Schlegel U 2012 Vowel articulation in Parkinson’s disease J. Voice
25 267–8 (Erratum)
[136] Orozco-Arroyave J R, Arias-Londoño J D, Vargas-Bonilla J F, Skodda S, Rusz J, Hönig F
and Nöth E 2016 Automatic detection of Parkinson’s disease in continuous speech spoken
in three different languages J. Acoust. Soc. Am. 139 481–500
[137] Rusz J 2018 Detecting speech disorders in early Parkinson’s disease by acoustic analysis
Habilitation Thesis Czech Technical University in Prague
[138] Sakar B E et al 2013 Collection and analysis of a Parkinson speech dataset with multiple
types of sound recordings IEEE J. Biomed. Health Inform. 17 828–34
[139] Bayestehtashk A, Asgari M, Shafran I and McNames J 2015 Fully automated assessment
of the severity of Parkinson’s disease from speech Comput. Speech Lang. 29 172–85
[140] Bot B M et al 2016 The mPower study, Parkinson disease mobile data collected using
ResearchKit Sci. Data 3 1–24
[141] Bot B M et al 2016 Synapse http://doi.org/10.7303/syn5511444.1
[142] Bachlin M et al 2010 Wearable assistant for Parkinson’s disease patients with the freezing of
gait symptom IEEE Trans. Inf. Technol. Biomed. 14 436–46
[143] Klucken J et al 2013 Unbiased and mobile gait analysis detects motor impairment in
Parkinson’s disease PLoS One 8 1–9
[144] Schlachetzki J C M et al 2017 Wearable sensors objectively measure gait parameters in
Parkinson’s disease PLoS One 12 1–18
[145] Hass C J, Malczak P, Nocera J, Stegemöller E L, Shukala A, Malaty I, Jacobson C E IV,
Okun M S and McFarland N 2012 Quantitative normative gait data in a large cohort of
ambulatory persons with Parkinson’s disease PLoS One 7 1–5
[146] Isenkul M E, Sakar B E and Kursun O 2014 Improved spiral test using digitized graphics
tablet for monitoring Parkinson’s disease Proc. of the 2nd Int. Conf. on e-Health and
Telemedicine pp 171–5
[147] Castrillón R et al 2019 Characterization of the handwriting skills as a biomarker for
Parkinson’s disease Proc. of the 14th Int. Conf. on Automatic Face and Gesture Recognition
pp 478–82
8-56
[148] Vasquez-Correa J C et al 2019 Multimodal assessment of Parkinson’s disease: a deep

learning approach IEEE J. Biomed. Health Inform. 23 1618–30
[149] Shin M and Park S et al 2006 Clinical and empirical applications of the Rey–Osterrieth
complex figure test Nat. Protoc. 1 892–9
[150] Sáenz-Lechón N, Godino-Llorente J, Osma-Ruiz V and Gómez-Vilda P 2006
Methodological issues in the development of automatic systems for voice pathology
detection Biomed. Signal Process. Control 1 120–8
[151] Vásquez-Correa J and Orozco-Arroyave J et al 2017 Multi-view representation learning via
gcca for multimodal analysis of Parkinson’s disease Proc. of ICASSP pp 2966–70
[152] Garca N et al 2018 Multimodal i-vectors to detect and evaluate Parkinson’s disease Proc. of
[153] Ghassemi N H et al 2018 Segmentation of gait sequences in sensor-based movement
analysis: a comparison of methods in Parkinson’s disease Sensors 18 1–15
[154] Vásquez-Correa J C et al 2019 Apkinson: a mobile solution for multimodal assessment of
patients with Parkinson’s disease Proc. of INTERSPEECH pp 964–5
[155] Postuma R B et al 2015 Risk factors for neurodegeneration in idiopathic rapid eye
movement sleep behavior disorder: a multicenter study Ann. Neurol. 77 830–9
[156] Alibiglou L et al 2016 Subliminal gait initiation deficits in rapid eye movement sleep
behavior disorder: A harbinger of freezing of gait? Mov. Disord. 31 1711–9
[157] Pineda-Trujillo N et al 2006 A genetic cluster of early onset Parkinson’s disease in a
colombian population Am. J. Med. Genet. Part B: Neuropsychiatr. Genet. 141B 885–9
[158] Postuma R B, Gagnon J F, Vendette M, Fantini M L, Massicotte-Marquez J and
Montplaisir J 2009 Quantifying the risk of neurodegenerative disease in idiopathic REM
sleep behavior disorder Neurology 72 1296–300
[159] Rusz J et al 2015 Speech disorders reflect differing pathophysiology in Parkinson’s disease,
progressive supranuclear palsy and multiple system atrophy J. Neurol. 262 992–1001
[160] Ebersbach G et al 2013 Clinical syndromes: Parkinsonian gait Mov. Disord. 28 1552–9
[161] Pinto S, Chan A, Guimarães I, Rothe-Neves R and Sadat J 2017 A cross-linguistic
perspective to the study of dysarthria in Parkinson’s disease J. Phon. 64 156–67
[162] Brognara L et al 2019 Assessing gait in Parkinson’s disease using wearable motion sensors:
a systematic review Diseases 7 1–14
[163] Fritsch J et al 2019 Automatic diagnosis of alzheimer’s disease using neural network
language models Proc. of ICASSP (Piscataway, NJ: IEEE), pp 5841–5
[164] Perez-Toro P et al 2019 Natural language analysis to detect Parkinson’s disease Proc. of
Text, Speech, and Dialogue pp 82–90
[165] Garca A M et al 2016 How language flows when movements don’t: an automated analysis
of spontaneous discourse in Parkinson’s disease Brain Lang. 162 19–28
8-57
IOP Publishing

Chapter 9
‘They labelled me ignorant’: the role
of neuroscience to support students with a
profile of dyslexia
Stephen Camilleri, Deborah Chetcuti and Ruth Falzon
Dyslexia is usually associated with individuals who have difficulties with reading and
writing. Although the challenges and positive skills associated with dyslexia have
been described in a number of research studies, the underlying factors that cause
dyslexia remain unclear. In recent years, there have been attempts to link neuro-
science research to educational research in order to try and better understand the
causes of dyslexia improve diagnosis and increase the effectiveness of intervention
programs. This chapter tries to develop a link between the narratives of Maltese
students with a profile of dyslexia and current scientific research in neuroscience.
Based on a qualitative approach, the chapter (1) gives voice to students with a profile
of dyslexia and the space to tell their personal story, and (2) reviews the literature to
demonstrate the importance of neuroscience as an interdisciplinary response to the
educational debates surrounding dyslexia. Evidence from the student narratives and
the neuroscience literature are then used to establish dyslexia as a real learning
difficulty, to emphasise the importance of early diagnosis and intervention, and to
highlight the implications for policy making regarding dyslexia in educational
practices.
9.1 Introduction
Dyslexia is usually associated with children who have difficulties with reading and
spelling (Crisp et al 2012). In dyslexic individuals, there is persistent difficulty
with literacy. Reading accuracy may be weak and reading fluency is below that
expected for a person of their level of education, intelligence, or professional status

(Shaywitz and Shaywitz 2008, p 1333). The British Dyslexia Association (BDA)
defines dyslexia as:
a specific learning difficulty which mainly affects the development of literacy

and language related skills. It is likely to be present at birth and to be life-long
in its effects. It is characterized by difficulties with phonological processing,
rapid naming, working memory, processing speed and the automatic develop-
ment of skills that may not match up to an individual’s other cognitive
abilities. (BDA 2017, para 4).
Various hypotheses have been suggested to explain the nature of dyslexia ranging
from poor phonological representation to deficits in visual processing and attention
span (Dresler et al 2018). However, to date, there is no one unifying theory that can
explain dyslexia and the International Dyslexia Association describes dyslexia as a
multi-dimensional profile.
An understanding of the nature of dyslexia is important to inform educational
practice and enable students with a profile of dyslexia to succeed within an educa-
tional system that places great emphasis on academic success, access, and assessment
of knowledge through reading and writing. As parents of young adults with dyslexia,
as teachers of students, and as educators in an institution of higher education, we
have lived through the achievements and failures of our children within an educa-
tional system that focuses mainly on academic success using traditional access to
material and presentation of knowledge. We have experienced first hand, what
Elwood et al (2017) describe as the significant impact that academic success has on
students’ life chances and opportunities. We have also observed the debilitating stress
and anxiety caused by school events such as tests, and high-stakes end-of-school
examinations that influence personal, social and academic development and self-
worth (e.g. Burden 2008, Schunk and Pajares 2002, Ziegler-Hill et al 2011). The ‘No
To Failure Report’ (2009) suggested that there is a very clear link between academic
failure and dyslexia. This means that one in five students affected by dyslexia are at a
greater risk of failure, anxiety and physical and mental health consequences (e.g.
Crocker and Knight 2005, Kannangara 2015, Vettiyadan et al 2018). Yet, with the
right support and using inclusive strategies of teaching and learning, dyslexic students
can do just as well as others and ‘there is simply no need for these children to be
slipping through the academic net’ (No To Failure Report 2009).
In recent years, in order to try and find the necessary support for students with a
profile of dyslexia an attempt has been made to link basic neuroscience research with
educational research in order to (a) try and develop an understanding of dyslexia
from a neurological perspective, and (b) improve diagnosis and the efficacy of
training and intervention programs (Gabrieli 2016). In this regard:
… neuroimaging is in general potentially helpful to reveal the underlying

cognitive mechanisms associated with LD, thereby improving the precision of
9-2
the differential diagnoses. Furthermore, it can help to detect children at risk

early on, to better understand the prognosis, and to develop more effective
interventions. Finally, neuroimaging can—when addressed adequately—be
used as a powerful illustrative tool to improve the communication of scientific
results to educators, policy-makers and the community in general (Dresler et al
2018, p 9).
This chapter, therefore, tries to establish a link between neuroscience and its
impact on educational practices in relation to dyslexia. Like Gabrieli et al (2015), we
would argue that studies in neuroscience have great potential in improving the lives
of individuals with a profile of dyslexia. ‘[N]euroscience research, research on
educational processes, and research on learning can be mutually informing … and
neuroscience can complement educational research’ (Hruby and Goswami 2011,
p 168). In other work (Camilleri et al 2019), we explored dyslexic students’
perspectives on their experiences of the national examinations in the Maltese
context. In the current chapter, we would like to use these narratives to support
Shaywitz and Shaywitz’s (2008) argument that, in addition to current educational
research on dyslexia, there is also the great potential of neuroscience research that
can lead to a deeper understanding of the identification, diagnosis and management
of dyslexia. Such hard evidence may also prove to be a strong political and
persuasive lobby tool for change.
9.2 Method
The main aim of this chapter was to compare the views of Maltese students with a
profile of dyslexia and current research on neuroscience in order to establish a link
between narratives of dyslexic youth and scientific research. Two main methods
were involved. First, a review of current literature on neuroscience research in
relation to dyslexia and secondly the use of youth narratives. The youths’ narratives
were collected for a research study (Camilleri et al 2019) that explored the views of
Maltese youth regarding national examinations. This study was a qualitative study
presenting open-ended, inductive, insightful data which represented the participants’
voice (Chetcuti et al 2016). Within this study, the youth’s narratives were used to
‘illustrate and suggest [not] explain and evaluate’ (Speedy 2018, p 142). The
narratives were also used to ‘evoke surprise and perceive events and experience
with a different lens which can act as an agent for change’ (Antonelli et al 2014, p 6).
In our previous study (Camilleri et al 2019), we wanted to capture students’ views
and experiences about issues that directly affect them; and to use their experiences to
bring about significant changes in their lives, learning and well-being (Elwood 2012).
The narratives gave the youth the space to narrate their otherwise untold stories.
‘[I]n our attempt to present these narratives we are both interpreting and allowing
the readers to create their own meaning of the youngsters’ experiences through the
voices of the youngsters [themselves]’ (Antonelli et al 2014, p 6).
However, this present research links the narrative account of students’ experience
to current neuroscience research in order to explore the complex dynamics involved
9-3
in the experiences of students with a profile of dyslexia. Maisog et al (2008) noted

that there has been very little attempt to link the narrative review of the literature on
neuroimaging with educational research on dyslexia. We believe that ‘a successful
interdisciplinary conversation could helpfully address many questions about literacy
and its instruction and development’ (Hruby and Goswami 2011, p 169).
9.2.1 The participants

The narratives present the voices of eight youths with a profile of dyslexia who had
just sat for the Maltese Secondary Education Certificate (SEC) which, within the
local Maltese context, is held at the end of compulsory schooling, when students are
15–16 years of age. This is the first set of national examinations which allow students
to proceed to academic postsecondary education, namely the national Matriculation
Certificate (MATSEC). MATSEC is the postsecondary entry certification needed
for university education (University of Malta 2013) and is comparable to the
International Baccalaureate or the Scholastic Assessment Test (SAT) in the
American context (Frey and Detterman 2004).
The youths had been formally assessed by educational psychologists or specialists
in dyslexia, and certified as having a profile of dyslexia. Participant recruitment was
through the Malta Dyslexia Association who acted as a gatekeeper to ensure an opt-
in decision-making process. Purposive sampling through snowballing was also used
(Voicu and Banonea 2011). The youths opted to participate in the study on a
voluntary basis. They all came from different backgrounds with parents who worked
in diverse areas ranging from health care to education to industry. We were aware
that, due to the small number of participants and the method of recruitment, the
findings of the study would remain ‘limited to the confines of the accessed network of
the participants’ (Penrod et al 2003, p 102).
9.2.2 Data collection and analysis

The experiences and stories of the participants of the study were collected using
semi-structured interviews that were carried out individually or in pairs. Following
Cohen et al (2011), the interviews were based on a number of open-ended questions
that were prepared beforehand but, at the same time, the participants were
encouraged to share other experiences that they deemed relevant to the issues being
discussed. The face-to-face interaction helped us to ensure that what the youths were
describing was rich, trustworthy and authentic (Fontana and Frey 2005). The
interviews were audio-recorded and later transcribed verbatim (Camilleri et al 2019).
Analysis of the data was based on a narrative-discursive approach (Taylor and
Littleton 2006). We immersed ourselves in the youth’s stories and, by ‘drowning in
the data’ (Goodson 2013, p 40), we were able to elicit a reflexive multi-voiced
narrative (Denzin and Lincoln 2005) that reflected the experiences of the partic-
ipants. We developed patterns across narratives that made meanings clearer (Taylor
and Littleton 2006) and evolved into a bricolage of stories (Goodson 2013) that
encapsulated the participants’ lived experiences of dyslexia.
9-4
All throughout the collection and analysis of the youths’ narratives we adhered to
the ethical considerations proposed by the American Psychological Association.
Anonymity, trust, fidelity, authenticity and respect for individuals were adhered
to (American Psychological Association 2020). We also sought approval from
the University of Malta’s Research Ethics Committee (UREC) and respected the
principles of informed consent, anonymity and the possibility to opt out of the study
at any point in time during the research process.
9.3 Bridging the gap: linking neuroscience and educational research

The integration of student narratives and current educational research on dyslexia
with findings from neuroscience research offers exciting opportunities for teaching,
learning, assessment practices, and policy making with respect to high-stakes
examinations (Szucs and Goswami 2007, Willingham and Lloyd 2007). Drawing
on an interdisciplinary model and using literature on neuroscience research and
student narratives, this chapter:
(1) demonstrates the importance of neuroscience in the identification of
dyslexia as an authentic learning difficulty;
(2) emphasises the importance of early diagnosis and intervention for dyslexic
students, and;
(3) discusses the implications of neuroscience research on educational practices
and policy making in high-stakes examinations.
9.3.1 The dyslexia debate: is dyslexia a real learning difficulty or is it a myth?

The characteristics of a dyslexic profile have been described in a number of studies
(e.g. Bender 2004, Catts 1989, Lerner and Johns 2011) and usually include
difficulties with reading fluency and accuracy, challenges with auditory, sequential
short-term memory (ASM), sequencing and organisation of ideas, factors that all
impact on the academic performance of students. Other studies (e.g. Alexander‐
Passe 2008, Crisp et al 2012, Shaywitz et al 2006a) also describe a number of
difficulties experienced by dyslexic students such as recognition of letters in the text,
blurring of letters, letters seen as mirror images and a confusing combination of
letters, all of which make it more difficult to understand a text and which can affect
comprehension. From a clinical perspective, the fifth edition (American Psychiatric
Association 2013) of the Diagnostic and Statistical Manual of Mental Disorders
(DSM-5) and the 2010 tenth edition of the International Classification of Diseases,
Clinical Modification (ICD-10) published by the World Health Organisation (WHO
2010) describe dyslexia as:
a specific and persistent learning disability affecting the acquisition and

development of the written language code (reading and spelling) and causing
significant handicap to academic achievement and/or activities of daily life;
it is considered where reading performance is poorer than one standard
deviation below the expected level for a given age and intelligence and where
these deficits cannot be explained by sensorial, neurological, psychiatric,
9-5
motivational, or other causes or by inadequate educational instruction. (Habib

and Giraud 2013, p 64)
Looking at the experiences of the youth participants, it becomes evident that

dyslexia is a realistic challenge that may significantly affect daily life, academic
achievements, academic self-concept, and self-esteem (Glazzard 2010). Bennett
(2015) argued that these challenges make the life of dyslexic youth twice as difficult:
Dyslexia is a learning disability. It is a different way of thinking. It makes me

process words in a different way and I have trouble with reading and writing.
I’m not so good in spelling and I also have trouble with remembering things. I
think that my brain is wired differently and does not work in the same way as
that of my friends. (Youth-participant Mark)
Other definitions of dyslexia try to move beyond the difficulties and challenges
experienced by youth with a profile of dyslexia and focus on a number of positive
skills. These positive skills include ‘big-picture thinking, problem-solving and lateral
thinking abilities, an instinctive understanding of how things work, originality,
creativity and exceptional visual spatial skills’ (British Dyslexia Association 2017,
para 4). In fact, the youths were keen to explain that, rather than looking at dyslexia
as a deficit, they looked at it as something that enabled them to be creative, and
helped them to be different in a positive manner:
I try to look at dyslexia from a positive side. I think that because of my

dyslexia I can think out of the box. I have a different way of looking at things.
I can solve problems in a way other teenagers do not even dream about.
Dyslexia is neither an ability nor a disability. It is something that is. There are
positives and negatives to it. Each of them outweighs the other. (Youth-
participant Rachel)
What is very evident from these youths’ experiences is that, for them, dyslexia is
something very real. It influences and affects their school life, their work life, their
leisure time, their relationships with family and peers, and their day-to-day
experiences.
Nothing is wrong with me. I have dyslexia. It is not a good thing. It is not a
bad thing. It’s ok to have it. My brain is just wired differently and I need to
take a few more steps before I can be the same as my friends. (Youth-
participant Samuel)
However, within the educational academic community and especially amongst

some teachers, educators and policy makers there is still some scepticism about what
dyslexia really is, what causes it, how it can be identified and what can be done about
it (Riddell and Weedon 2006). In the assessment community, for example, there is
still the persistent notion that students with dyslexia face difficulties and challenges
9-6
that are not very different from other students who have weak reading abilities or
lower cognitive skills (Crisp et al 2012). The debate has been compounded by the
conceptualisation of ‘dyslexia as a myth’ in a number of newspaper articles and
television news stories.
In 2005, for example, during a television interview a university professor, Julian
Elliot argued that the term dyslexia, is a meaningless label and that there is really no
difference between dyslexic individuals and poor readers. In his point of view,
dyslexia is just fiction, made up by middle-class parents who do not accept that their
children just cannot read (Moorhead 2005). Such reasoning becomes problematic
and has serious implications for teachers, educators and examiners, since if these
individuals do not believe in dyslexia, then they will not invest in trying to help
students with a profile of dyslexia overcome their challenges.
The foundation of such debates on the existence of dyslexia as a real learning
difficulty arises from the fact that, until recently, most research studies regarding
dyslexia were the result of observations of students with dyslexia and how they
performed in the classroom or on a number of psychological tests. This points to the
need for more scientific evidence to support the neurobiological basis of dyslexia.
The contributions from neuroscience offer the possibility of interdisciplinary
integration of brain, social, cognitive and cultural perspectives in ways that will
be beneficial to all individuals with a profile of dyslexia (Hruby and Goswami 2011).
9.3.1.1 Supporting evidence from neuroscience

Diagnosis of dyslexia has traditionally depended on the use of standardised
psychological tests and tools that assess reading and spelling as well as underlying
literacy skills: phonological skills, the ability to segment, syllabise and blend,
cognitive skills such as speed of processing and auditory sequential short term
memory (e.g. Habib and Giraud 2013, Vellutino et al 2004, Ziegler et al 2010).
However, more recent studies have focused on the use of neuroimaging technologies
such as electroencephalography (EEG), functional magnetic resonance imaging
(fMRI), and magnetoencephalography (MEG) amongst others to identify and
diagnose dyslexia (e.g. Maisog et al 2008, Shaywitz et al 2006b, Habib and
Giraud 2013). Research studies carried out in different countries have now
established what Shaywitz and Shaywitz (2008) describe as ‘a neural signature for
dyslexia’ (p 1333). These neuroimaging studies have led to a better understanding of
the mechanisms involved in dyslexia (Dresler et al 2018). They provide evidence that
learning disabilities, within which dyslexia is the largest population (Ferrer et al
2010) are a real disability that can be explained by atypical neural activity in the
brain, and challenge the description of dyslexia as a ‘hidden impairment’
(Shakespeare 2008, p 13).
These neuroimaging techniques measure changes in metabolic activity and blood
flow in specific brain regions while individuals are carrying out specific cognitive
functions such as reading (Shaywitz et al 2006a). They have provided evidence that
determines the neural basis of dyslexia as a hypoactivation of the left posterior
system in dyslexia, and also identified the cerebellum as playing a potential role in
reading (Maisog et al 2008). What is of particular significance and of relevance for
9-7
students, teachers, educators and examiners is that the studies and ‘data from
laboratories around the world indicate that there are a number of interrelated neural
systems used in reading, at least two in the posterior brain regions as well as distinct
related systems in anterior regions’ (Shaywitz and Shaywitz 2008, p 1333). This
means that dyslexia can be directly linked to functional under-activation in two
posterior systems of the left hemisphere which are referred to as the Visual Word
Form Area (VWFA). In non-dyslexic individuals, activation of these brain areas
leads to automatic decoding of the grapheme–phoneme used in reading, while an
under-activation of these areas is associated with problems in phonological reading
(Linkersdorfer et al 2012). Other evidence from neuroimaging studies using EEG
and MEG also suggest that there are also anomalies in the auditory processing of
linguistic stimuli in students with a profile of dyslexia (Habib and Giraud 2013).
These studies with dyslexic students are particularly important because ‘they
indicate that dysfunction in left hemisphere posterior reading circuits is already
present in dyslexic children and cannot be ascribed simply to a lifetime of poor
reading’ (Shaywitz and Shaywitz 2008, p 1336). This confirms the existence of the
dyslexia profile. Like Shaywitz et al (2006a) we would therefore argue that ‘these
findings have implications for the acceptance of dyslexia as a valid disorder—a
necessary condition for its identification and treatment. Simply put, such studies
provide for the first time, convincing, irrefutable evidence that what has been
considered a hidden disability is real’ (p 281). The capacity of neuroscience to reveal
the inner workings of the brain is a unique contribution to educational research, to
the provision of resources for students and to policy making that can result in
positive outcomes for students with a profile of dyslexia (Gabrieli 2016).
9.3.2 Paying attention to early diagnosis, support and intervention

Reading and writing play an important role in academic success in school, in the
world of employment and in one’s personal life. The ease with which some children
learn to read contrasts sharply with the challenges and struggles faced by dyslexic
students as they try to extract meaning out of print (Shaywitz and Shaywitz 2008).
The lack of fluent reading and writing, the numerous reading and spelling mistakes
and, at times, difficulties with ASM, organisation and motor skills are observed by
teachers and educators in classrooms and are often considered to be the hallmark of
dyslexia (e.g. Baddeley et al 1988, Lyon et al 2003, Williams and Lynch 2010). In
some cases, however, teachers fail to recognize the difficulties being demonstrated by
the students as being characteristic of a learning difficulty. This results in students
feeling stupid when comparing themselves to others (Glazzard 2010); that they have
no control over their own success and that no matter how much effort they make
they will not be able to learn and improve their results (Humphreys and Mullins
2002). This very often leads to a sense of learned helplessness and low self-esteem
(e.g. Burden 2008, Burden and Burdett 2007, Kannangara 2015).
Dyslexic youth very often associate academic success with their sense of self-
worth. They then not only experience low academic self-concept, but generalize this
to every aspect of their lives seeing themselves as failures (e.g. Alexander-Passe 2008,
9-8
Riley and Rustique-Forrester 2002, Vettiyadan et al 2018). These emotional scars

have a significant impact on the wellbeing of teens and their future aspirations. This
sense of frustration, despair and seemingly a call for help is very evident in the
experiences of the youth who participated in the study:
I do not think that my dyslexia makes me less intelligent, less able of complex
thinking. If I had to think of one thing about dyslexia, it’s the frustration that
it makes me feel. The feelings of inadequacy and the constant emotional
struggles that I am never good enough. That I will never succeed in life. It
makes me feel less than my friends. And I always downplay my successes. The
problem that my teachers fail to understand is the feeling of uselessness that
wells up inside. (Youth-participant Matthew)
The students narrated experiences of being misunderstood by their teachers,

feeling inadequate, and an underlying desire to be treated with the respect they feel
they deserve:
When I was nine I was tested. I found out I was dyslexic. But when my mother
told the teacher she did not believe her. Instead I was told that I was lazy. That
I did not do well on purpose. I was put at the back of the class. I was ignored
and picked on. That really hurt me. Once we had a test and I did not do well.
The teacher shouted at me and told me that I did not study. I felt really bad. I
felt as if I was nothing. I felt as if I did not exist. (Youth-participant Susanna)
This heartfelt cry for help suggests that it is very important to diagnose students
as early on as possible and also to make teachers and educators aware of the
characteristics of dyslexia so that they can avoid labelling students as lazy and stupid
(e.g. Burden 2008, Gwernan‐Jones and Burden 2010, Nascimento et al 2018). The
support of parents, teachers and educators is extremely important in encouraging
dyslexic students and helping them to overcome their challenges (Nalavany and
Carawan 2012):
What helped me to overcome the challenges of my dyslexia was my English

teacher… She was very different from my other teachers who always blamed
my low achievement on lack of effort or lack of ability. Instead she encouraged
me. She told me ‘you’re getting there, don’t give up, your spelling is getting
better’. This made me feel good about myself because I knew that at least I was
not disappointing my teacher and that she cared in spite of all my mistakes.
(Youth-participant Matthew)
Another important aspect linked with early diagnosis, is that of intervention. A

number of teachers and educators assume that students with a profile of dyslexia
cannot learn in the same way as other students and therefore that they cannot
improve (e.g. Pumfrey and Reason 2013, Soriano-Ferrer et al 2016, Washburn et al
2014). This can be very disheartening for students with a profile of dyslexia,
9-9
especially when these students know that, with some extra help, attention and with
appropriate pedagogic interventions, their learning can improve drastically (Gibson
and Kendall 2010, Glazzard and Dale 2015):
I know that I can learn. I need a little extra help but eventually I do get there.
If I have to learn a poem I think of it as a story. I make a video of it in my
mind. I use a lot of visual aids. When I want to learn history I think of events
as stories. When I want to learn science I think of diagrams and experiments. I
wish my teachers would teach me in this way. It is not really difficult and it
would really make a difference for me. (Youth-participant Sarah)
From an educational perspective, these youths’ narratives illustrate the impor-

tance of providing improved opportunities for students to learn in the ways that are
suitable for them (Scott et al 2014). These authors argue that if we want to make
learning and education more accessible for all students, including those with a
profile of dyslexia, then there needs to be ‘a heightened awareness of educators of the
diversity among students in their classrooms and the increased expectations of
society that educators will address the learning needs of all students’ (p 68).
9.3.2.1 Neuroscience and its implications for early diagnosis and intervention
in dyslexia
The early diagnosis of dyslexia can be very complex. However, early diagnosis of
dyslexia is especially important because the use of the written language is usually the
medium for learning both in school as well as in daily life (Habib and Giraud 2013).
Using traditional assessment tools, diagnosis for dyslexia cannot be made until a
child has started to learn a language and its literacy. To date, diagnosis is carried out
when children are observed and investigated using standardized testing material
around the ages of six–seven (Snowling 2013). However, most of the time, problems
with reading are only evident after the student has failed in one way or another
(Gabrieli 2016). Neuroscience can help to identify the characteristics of the learning
difficulty and point to the required educational interventions (Gabrieli 2016) before
observed experiences of failure. Neurological imaging is therefore important in the
early diagnosis of dyslexia and in mapping what is going on inside the brain when
children learn how to read (Kraft et al 2016). This does not mean that every child
needs to undergo neuroimage testing, but research using neuroimaging gives us an
indication about how reading development takes place and the kind of educational
interventions that can help dyslexic children improve their reading and writing skills
(Kraft et al 2016, Snowling 2013). The literature presents ample research findings
which point to the use of neuroscience in early detection and diagnosis for dyslexia
(box 9.1).
Three important conclusions that emerge from neuroimaging studies regarding
the early diagnosis and intervention in dyslexia include the ideas that (1) learning to
read depends on the development of phonological awareness—the sensitivity to the
sound structure of spoken words; (2) the neural systems for skilled and fluent reading
are malleable and able to respond to effective reading interventions; and (3) poor
9-10
Box 9.1. Neurological studies and early detection and diagnosis.
• Microstructural level studies attributing variations related to disruptions of

neuronal migration during neocortical development (e.g. Szalkowski et al
2012, Tammimies et al 2013, Gabel et al 2010, 2011).
• Structural anomalies in the cortical layers of the perisylvian brain regions leading
to a functional disruption of the local microcircuitry (Galaburda et al 2006,
Giraud and Ramus 2013).
• Significantly increased cortical thickness (CortT) of the left temporo-parietal
regions, the angular and supramarginal gyri (SMG) (Darki et al 2014).
• fMRI studies also describing hypoactivations during reading and reading-related
tasks in temporo-parietal brain regions, superior temporal, inferior parietal, and
in inferior frontal brain regions (Richlan et al 2009, 2013).
• Structural MRI studies reporting neuroanatomical differences in the grey matter
of the dorsal reading network, including the left superior temporal cortex
(Richlan et al 2013), the left inferior parietal cortex (Darki et al 2014) and the
ventral reading network, including the left fusiform gyrus (Skeide et al 2016,
Altarelli et al 2013).
• Disruptions of interregional connectivity in specific brain regions and (Skeide
et al 2015, Vandermosten et al 2012, 2015, Yeatman et al 2011, Rimrodt et al
2010).
• Anomalies within the inferior fronto-occipital fascicle (IFOF), a long-distance
fiber tract thought to be related to visual word form recognition (Vandermosten
et al 2012, 2015, Jobard et al 2003).
• Disruptions in the connectivity of the dorsal reading network (Saygin et al 2013).
readers do not always use phonologic strategies to analyze unfamiliar words but
they appear to rely more on memory-based strategies (Shaywitz et al 2006b).
The data obtained from various studies using neuroimaging techniques indicate
that reading intervention strategies bring about significant changes in brain
organization, suggesting that learning has taken place. In a meta-analysis of studies
regarding neuroimaging and reading intervention, Barquero et al (2014) revealed
‘differences in brain activity are associated with intervention’ (p 13). This has
important implications for educational practices and policy making in schools, since
it means that developing effective intervention becomes a more realistic goal when
these are based on neuroimaging studies (Barquero et al 2014). Shaywitz and
Shaywitz (2008) advocate that ‘the provision of an evidence-based reading inter-
vention at an early age improves reading and facilitates the development of those
neural systems that underlie reading’ (p 1340).
Such findings offer hope for individuals with a profile of dyslexia since, through
proper interventions, they can find themselves on a level-playing field with their
peers who do not have a learning difficulty. The utilization of advances in neuro-
science can go a long way in informing and, more importantly, persuading teachers
and educators so that they understand the characteristics of dyslexia, and learn how
to intervene to help their students achieve the best that they can. With these
9-11
advances in neuroscience, the experiences of low self-worth and lack of respect

described by the youth participants can become a thing of the past.
Notwithstanding, in spite of empirical evidence relating to neuroanatomical
predictability of dyslexia, such evidence is ‘sparse compared to the intensively
investigated behavioral predictors’ (Kraft et al 2016, p 379). However, the hope is
that in the future neuroimaging can play an important part in early detection and be
used in screening in as much as hearing tests and vision tests are carried out in
schools. This may further support the practice of early detection and intervention as:
[c]onsistent with much of the cited research, factors which facilitate individuals
in enjoying academic success with dyslexia are: early identification of dyslexia,
supportive families, developing a positive ‘dyslexic identity’, developing self-
advocacy skills and being provided with adequate learning resources. (Pitt and
Sonic 2017, p 403)
9.3.3 ‘They labelled me ignorant’—Examinations and dyslexia

At one time or another, students who attend school need to sit for tests and
examinations. When these examinations are held at the end of secondary or post-
secondary school, they become very high-stakes as they can determine the life
chances of individuals (Elwood et al 2017). They provide the key for entry into post-
secondary education or into the world of work (Elwood 2012). Students with a
profile of dyslexia often believe that they will be judged on their academic success
and the realization that their dyslexic profile will prevent them from being successful
can lead to a number of emotional challenges including depression, withdrawal and
feelings of exclusion (e.g. Burden 2008, Falzon and Camilleri 2010, Kannangara
2015):
I used to be very worried because I used to think that how I perform in the
exam is going to affect my whole life. If I don’t manage to pass the exams I
won’t be able to do what I really want to do… Sometimes you have
opportunities but you cannot take them because you do not have the
qualifications. (Youth-participant Samuel)
High-stakes examinations are stressful for all students, but for students who enter
the examinations race at a disadvantage, such as those with a profile of dyslexia, the
pressures may be magnified (Gardner et al 2009). This gives rise to emotional
turmoil (e.g. Alexander‐Passe 2008, Burden 2008, Kannangara 2015) and a sense of
despair resulting in low self-worth (Covington 1992):
How can I ever achieve my full potential? Must all the work, effort and
struggle throughout a course of studies be valued on a single day? Should
examinations create so much anxiety and nervousness? Examiners fail to
understand that these negative experiences have repercussions. Self-worth is
9-12
destroyed along with self-respect and the notion that in life you get what you
deserve. I wonder how many great minds were lost simply because the type of
intelligence and their ideas were not the ones the examiners wanted? (Youth-
participant Matthew)
During examinations, students with a profile of dyslexia have to regularly engage

with tasks that might be unnecessarily extremely difficult for them (Burden 2008)
due to the medium of access and presentation of work demanded (e.g. Douglas et al
2011, Shaw et al 2017). These tasks may include reading, spelling, writing,
organizing or mathematics which are all strongly related to the challenges faced
by dyslexic students in their day-to-day life (Bender 2004, Lerner and Johns 2011).
Dyslexic students may also experience additional anxiety due to test conditions that
do not allow them to perform at their best. Such conditions may include time
constraints, presentation of paper, access to print, size of print, the way in which
questions are asked and the environment in which the examination itself is taking
place (Camilleri et al 2019):
I had a question and I could not understand what the examiner was asking. I
might have known the answer but the wording was really confusing. What did
the examiner want? I got scared, disappointed. The paper was too challenging.
I did not have enough light on the paper. The examiner walked around me.
Tick…tick…tick…of her heels. It was so distracting. I was running a race
against time. I started to write fast… and I began to make more mistakes. I
didn’t have time to revise my work. I couldn’t breathe. (Youth-participant
Sarah)
The youths’ narratives suggest that taking tests and examinations led to great
emotional stress and that, as Sarah argued, this could affect their actual performance
on the examination. This is a double-edged sword because the dyslexic youths, in
addition to being at a disadvantage as compared to their peers at the start of the
examination, also knew that they could never perform at their best, no matter how
hard they tried. This led to a sense of learned helplessness and a low academic self-
concept that was usually generalized to other aspects of the youths’ lives (Humphrey
and Mullins 2002):
I used to feel very badly about the fact that I knew that I deserved a B but I
knew that I was going to get a C. That was something that bothered me. It
used to make me feel down because I knew that the marks did not reflect my
true potential. A pity, but there was nothing I could do…. (Youth participant
Robert)
The stress and anxiety caused by examinations can be relieved when students with
a profile of dyslexia are provided with examination access arrangements (EAA).
Within the Maltese context, the MATSEC examinations board provides EAAs
which may include extra time, a reader, a prompter or a scribe depending on the
9-13
severity of the dyslexia. Further, the University of Malta has lately accepted the use
of technology as an examination access arrangement for university students as from
academic year October 2019 (University of Malta 2018). This will soon also be
considered for national examinations. Notwithstanding, the assigning of EAAs
remains limited, also due to assessment and test-scores interpretations, or rather
misinterpretations (Falzon and Camilleri 2014). The youth-participants lamented
that the examination board and examiners did not acknowledge and understand
their literacy challenges (Camilleri et al 2019). This is mainly due to the belief of
some stakeholders, including parents, educators and examiners that EAA might
inadvertently advantage dyslexic students over students who did not have any
learning difficulties (Grima and Ventura 2006). This position, taken by the
examination board perplexed the youth-participants. Like the participants in a
study by Elliot and Marquart (2004), the youths believed that their requests were not
affecting the examination objectives or putting them at an advantage:
If I need to wear glasses I am allowed to wear them in the exam. This does not
give me an advantage over someone who does not need glasses. Similarly, for
my dyslexia if you give me some help like some more time you are not giving
me something extra. You are just giving me what I need to start off at par with
others who do not have dyslexia. (Youth-participant Fabian)
The youth-participants noted that the allocation of EAA would make a difference
and could have helped them to feel less anxious when sitting for high-stakes
examinations. They insisted that, whilst definitely not asking for or wanting any
preferential treatment, they wanted to be treated fairly and on the same level-playing
field as other students (Cribb and Gerwitz 2003). Their argument was that ‘tradi-
tional forms of assessment are fundamentally discriminatory and … the onus lies
with the institution to find new forms of assessment which will no longer penalize
students with learning difficulties’ (Riddell and Weedon 2006, p 58):
It is hard to say whether my grade would change or not. If I had extra time, my
level of anxiety would probably have been less. Even if I did not use that extra
time, the thought that I had that extra 15 min would help to calm me down.
(Youth-participant Matthew.)
The issues described by the youth-participants point to issues of fairness and

equity. In other words, being given the opportunity to show their diverse thinking
and creativity skills through media they are comfortable with. Unfortunately,
examination boards seem to be more concerned with the technical and logical
aspects of examinations and fail to recognize the emotional impact that examina-
tions may have on students with dyslexia (Camilleri et al 2019, Kirwan and Leather
2011) and to the irrelevance of some of the requirements insisted upon (Falzon 2011,
Falzon and Camilleri 2014). The youth-participants call for an assessment system
that recognizes and celebrates success in a range of areas (Camilleri et al 2019,
Glazzard 2010).
9-14
9.3.3.1 Neuroscience research and implications for policy making regarding

examinations
As indicated in the narratives of the youth-participants, neuroscience research has a
number of practical implications for policy making regarding dyslexia in relation to
high-stakes examinations. The main finding from neuroscience is the establishment of
dyslexia as a non-hidden profile, where the covert is now overt (Shaywitz et al 2006b).
Such findings may also have implications on policy regarding the allocation of EAAs for
students with a profile of dyslexia in primary, secondary and post-secondary education.
Shaywitz and Shaywitz (2008), for example, stressed that fMRI studies ‘now
indicate a neural basis for the accommodation of extra time provided on stand-
ardised tests to dyslexic students’ (p 1343). According to these authors, this neural
basis is the evidence that dyslexic students show a disruption of posterior systems
when they try to engage in skilled reading and they try to compensate for this by
using ancillary pathways anteriorly and in the right hemisphere. This means that
whilst dyslexic students are able to decode words, they may do so slowly and not
automatically to the detriment of reading fluency and access to reading compre-
hension (e.g. Pikulski and Chard 2005, Kim et al 2011, Wise et al 2010). The
practical implication of these findings for examination boards is that ‘this lack of
automatic, fluent reading means that the dyslexic reader may be able to decode
words, but is still not able to read quickly and continues to be at a disadvantage
when compared to non-dyslexic peers when taking high-stakes standardized tests’
(Shaywitz and Shaywitz 2008, p 1343). In this context, therefore:
[s]uch findings should make policymakers more willing to allow children and
adolescents with dyslexia to receive accommodations (such as extra time) on
high-stakes tests. That would allow dyslexic readers with a disruption in the
word-form area (influencing skilled, fluent reading) to be on a level-playing
field with their peers who do not have a reading disability. The utilization of
advances in neuroscience to inform educational policy and practices provides
an exciting example of translational science being used for the public good.
(Shaywitz et al 2006b, p 281)
9.4 Final reflections

In this chapter we present a narrative that tries to link the views and perspectives of
students with a profile of dyslexia and current research on neuroscience. We argue in
favor of the added value of neuroscience to inform educational practices and
intervention in relation to learning disabilities, more specifically to dyslexia (e.g.
Dresler et al 2018, Gabrieli 2016, Shaywitz et al 2006b). We disagree with Bowers
(2016) and Bishop (2014) who stressed that neuroscience gives us plenty of
information about the brain but, as researchers, we should not pretend that this
research has any practical or educational relevance. The presentation of neuro-
logical hard evidence can make the journey to legislation and implementation much
easier (Thomson et al 2007, Lemos and Morehouse 2005). At times hard evidence
9-15
can be more influential than social science. In her doctoral thesis, Falzon (2012)
reflected that ‘[t]he concept of using a quantitative method as a “confirmation” was a
political and not an academic or philosophical decision. I felt that I needed to speak
the language of the people I wanted to convince to take action’ (p 134). Such
research, thus, augurs well for persons with dyslexia. Whilst the commitment to
evidence-based policy ‘provide[s] an astonishing new opportunity for researchers in
material, intellectual and political terms[,] [l]et us hope we are up to it. But let us
remain modest in our claims to improve the conduct of public affairs’ (Solesbury
2001, pp 9–10). Whilst we need to consider all forms of research that can be used to
improve the lives of youth with a profile of dyslexia, we still need to remain cautious
in our interpretation of evidence, and in promises made to students.
In spite of such a possibly gloomy scenario with regard to the power of research
on policy making and implementation, we still believe in its benefits, as shown
through the voices of the youth participants in our study. In the case of dyslexia,
neuroscience has tabled the existence of such a profile and results on the workings of
the brain are verbalised in the youths’ narratives. It is hoped that such research will
help create a cohesive definition of dyslexia across nations, support the concept that
dyslexia is a reading and not a thinking disability, and help pave the way for all
stakeholders to understand that dyslexia should be regarded as a neurodiverse
profile and not a disability or a deficiency (e.g. Baker 2011, Kapp et al 2013).
The beacon for research aims should always be the human experience as there is ‘a
social responsibility in research that transcends the academic discipline of a profession
to which the researcher belongs. The ultimate moral justification for research is that it
makes a contribution to the greater public good, by easing suffering or promoting
truth’ (McLeod 2003, p 175). Let us not lose sight of the true aim of such research, the
awareness that this neurodiverse profile exists and needs to be respected, addressed and
considered in all aspects of the human experience. Let there be no more wastage, brain
drains and loss of human opportunity. Neuroscience can be an asset to the cause.
Let us try and bridge the gap between neuroscience research and educational
research as an ‘interdisciplinary collaboration across the fields of neuroscience,
psychology/cognitive science, and education’ (Dresler et al 2018, p 2) in order to
clarify root causes, improve prognosis, and yield effective remediation. The aim of
this collaboration would be to listen to the stories of dyslexic youth and use them as
a platform from which to embark on neuroscientific studies, that will help improve
the everyday life of dyslexic youth.
References
Alexander‐Passe N 2008 The sources and manifestations of stress amongst school‐aged dyslexics,
compared with sibling controls Dyslexia 14 291–313
Altarelli I, Monzalvo K, Iannuzzi S, Fluss J, Billard C, Ramus F and Dehaene-Lambertz G 2013
A functionally guided approach to the morthometry of occipitotemporal regions in devel-
opmental dyslexia: evidence for differential effects in boys and girls J. Neurosci. 33 11296–301
Antonelli L et al 2014 Drama, performance ethnography, and self-esteem: listening to youngsters
with dyslexia and their parents SAGE Open 4 2158244014534696
9-16
American Psychiatric Association 2013 Diagnostic and Statistical Manual of Mental Disorders
(DSM-5®) (Washington, DC: APA)
American Psychological Association 2020 Publication manual of the American Psychological
Association 7th edn http://doi/10.1037/0000165-000
Baddeley A D, Logie R H and Ellis N C 1988 Characteristics of developmental dyslexia Cognition
29 197–228
Baker D L 2011 The Politics of Neurodiversity: Why Public Policy Matters (Boulder, CO: Lynne
Rienner)
Barquero L A, Davis N and Cutting L E 2014 Neuroimaging of reading intervention: A
systematic review and activation likelihood estimate meta-analysis PLoS One 9 e83668
Bender W N 2004 Learning disabilities: Characteristics, identification, and teaching strategies
(Boston, MA: Allyn & Bacon)
Bennett R E 2015 The changing nature of educational assessment Rev. Res. Educ. 39 370–407
Bishop D V M 2014 What is Educational Neuroscience? Available at: https://figshare.com/articles/
What_is_educational_neuroscience_/1030405
Bowers J S 2016 The practical and principled problems with educational neuroscience Psychol.
Rev. 123 600–12
British Dyslexia Association 2017 Retrieved April 10, 2017, from http://bdadyslexia.org.uk/
educator/what-are-specific-learning-difficulties#Dyslexia
Burden R L 2008 Is dyslexia necessarily associated with negative feelings of self-worth? A review
and implications for future research Dyslexia 14 188–96
Burden R and Burdett J 2007 What’s in a name? Students with dyslexia: their use of metaphor in
making sense of their disability Br. J. Spec. Educ. 34 75–9
Camilleri S, Chetcuti D and Falzon R 2019 ‘They labelled me ignorant’: narratives of Maltese
youth with dyslexia on national examinations SAGE Open 9 1–18
Catts H W 1989 Defining dyslexia as a developmental language disorder Ann. Dyslexia 39 50
Chetcuti D, Falzon R and Camilleri S 2016 D Pebble in My Shoe: Teenagers’ Experiences of
Dyslexia and Examinations (Malta: The authors)
Cohen L, Manion L and Morrison K 2011 Research Methods in Education (New York:
Routledge)
Covington M V 1992 Making the Grade: A Self-worth Perspective on Motivation and School
Reform (New York: Cambridge University Press)
Cribb A and Gerwitz S 2003 Towards a sociology of just practices: an analysis of plural conceptions
of justice Social Justice in Education ed C Vincent (London: Routledge/Falmer), pp 15–29
Crisp V, Johnson M and Novacovic N 2012 The effects of features of examination questions on
the performance of students with dyslexia Br. Educ. Res. J. 38 813–39
Crocker J and Knight K M 2005 Contingencies of self-worth Curr. Dir. Psychol. Sci. 14 200–3
Darki F, Peyrard-Janvid M, Matsson H, Kere J and Klingberg T 2014 DCDC2 polymorphism is
associated with left temporoparietal gray and white matter structures during development
J. Neurosci. 34 14455–62
Denzin N K and Lincoln Y S 2005 Introduction: the discipline and practice of qualitative research
The Sage Handbook of Qualitative Research ed N K Denzin and Y S Lincoln (Los Angeles,
CA: Sage), pp 1–20
Douglas G, McLinden M, McCall S, Pavey S, Ware J and Farrell A M 2011 Access to print
literacy for children and young people with visual impairment: findings from a review of
literature Eur. J. Spec. Needs Educ. 26 25–38
9-17
Dresler T, Bugden S, Gouet C, Lallier M, Oliveira D G, Pinheiro-Chagas P, Pires A C, Wang Y,

Zugarramurdi C and Weissheimer J 2018 A translational framework of educational neuro-
science in learning disorders Front. Integr. Neurosci. 12 1–12
Elliot S N and Marquart A M 2004 Extended time as a testing accommodation: Its effects and
perceived consequences Except. Child. 70 349–67
Elwood J 2012 Qualifications, examinations and assessment: views and perspectives of students in
the 14–19 phase on policy and practice Camb. J. Educ. 42 497–512
Elwood J, Hopfenbeck T and Baird J 2017 Predictability in high-stakes examinations: students’
perspectives on a perennnial assessment dilemma Res. Pap. Educ. 32 1–17
Failure Project & The Dyslexia-SpLD Trust 2009 No to failure report (Bracknell: No to Failure
Project & The Dyslexia-SpLD Trust)
Falzon R 2011 Being able to read or not: Caution from the experts Times of Malta Retrieved from
http://timesofmalta.com/articles/view/20110306/education/being-ableto-read-or-not-caution-
from-the-experts.353295
Falzon R 2012 Early educators’ awareness and knowledge of structured multisensory literary
instruction Doctoral Dissertation Northumbria University
Falzon R and Camilleri S 2010 Dyslexia and the school counsellor: a Maltese case study Couns.
Psychother. Res. 10 307–15
Falzon R and Camilleri J 2014 Request for oral examinations at the SEC and MATSEC levels
National petition presented to the Maltese Parliament https://change.org/p/ministry-for-
education-and-employment-give-the-option-to-students-to-do-their-examinations-in-writing-
orally-or-using-a-word-processor
Ferrer E, Shaywitz B A, Holahan J M, Marchione K and Shaywitz S E 2010 Uncoupling
of reading and IQ over time: Empirical evidence for a definition of dyslexia Psychol. Sci. 21
93–101
Fontana A and Frey J H 2005 The interview: From neutral stance to political involvement The
Sage Handbook of Qualitative Research 3rd edn ed N K Denzin and Y S Lincoln (Thousand
Oaks, CA: Sage), pp 695–727
Frey M C and Detterman D K 2004 Scholastic assessment or g? The relationship between the
scholastic assessment test and general cognitive ability Psychol. Sci. 15 373–8
Gabel L A, Gibson C J, Gruen J R and LoTurco J J 2010 Progress towards a cellular
neurobiology of reading disability Neurobiol. Dis. 38 173–80
Gabel L A, Marin I, LoTurco J J, Che A, Murphy C, Manglani M and Kass S 2011 Mutation of
the dyslexia-associated gene Dcdc2 impairs LTM and visuo-spatial performance in mice
Genes Brain Behav. 10 868–75
Galaburda A M, LoTurco J, Ramus F, Fitch R H and Rosen G D 2006 From genes to behavior
in developmental dyslexia Nat. Neurosci. 9 1213–7
Gabrieli J D 2016 The promise of educational neuroscience: comment on Bowers (2016) Psychol.
Rev. 123 613–9
Gabrieli J D, Ghosh S S and Whitfield-Gabrieli S 2015 Prediction as a humanitarian and
pragmatic contribution from human cognitive neuroscience Neuron 85 11–26
Gardner J, Holmes B and Leitch R 2009 Assessment and Social Justice (Bristol: Futurelab)
Gibson S and Kendall L 2010 Stories from school: Dyslexia and learners’ voices on factors
impacting on achievement Support Learn. 25 187–93
Giraud A-L and Ramus F 2013 Neurogenetics and auditory processing in developmental dyslexia
Curr. Opin. Neurobiol. 23 37–42
9-18
Glazzard J 2010 The impact of dyslexia on pupils’ self‐esteem Support Learn. 25 63–9
Glazzard J and Dale K 2015 ‘It Takes Me Half a Bottle of Whisky to Get through One of Your
Assignments’: exploring one teacher educator’s personal experiences of dyslexia Dyslexia 21
177–92
Goodson I 2013 Developing Narrative Theory: Life Histories and Personal Representation
(Abingdon: Routledge)
Grima G and Ventura F 2006 Auditing a national system of examinations: issues of selectivity in
the Secondary Education Certificate (SEC) examination in Malta Int. Stud. Educ. 16 207–21
Gwernan‐Jones R and Burden R L 2010 Are they just lazy? Student teachers’ attitudes about
dyslexia Dyslexia 16 66–86
Habib M and Giraud K 2013 Dyslexia Handbook of Clinical Neurology, Vol. III (3rd Series),
Pediatric Neurology Part I ed O Dulac, M Lassonde and H B Sarnat (Amsterdam: Elsevier),
pp 229–44
Hruby G G and Goswami U 2011 Neuroscience and reading: A review for reading education
researchers Read. Res. Q. 46 156–72
Humphrey N and Mullins P M 2002 Personal constructs and attribution for academic success and
failure in dyslexia Br. J. Spec. Educ. 29 196–203
Jobard G, Crivello F and Tzourio-Mazoyer N 2003 Evaluation of the dual route theory of
reading: a metanalysis of 35 neuroimaging studies Neuroimage 20 693–712
Kannangara C S 2015 From languishing dyslexia to thriving dyslexia: developing a new
conceptual approach to working with people with dyslexia Front. Psychol. 6 1976
Kapp S K, Gillespie-Lynch K, Sherman L E and Hutman T 2013 Deficit, difference, or both?
Autism and neurodiversity Dev. Psychol. 49 59
Kim Y S, Wagner R K and Foster E 2011 Relations among oral reading fluency, silent reading
fluency, and reading comprehension: A latent variable study of first-grade readers Sci. Stud.
Read. 15 338–62
Kirwan B and Leather C 2011 Students’ voices: A report of the student view of dyslexia study
skills tuition Support Learn. 26 33–41
Kraft I et al 2016 Predicting early signs of dyslexia at a preliterate age by combining behavioral
assessment with structural MRI Neuroimage 143 378–86
Lerner J W and Johns B 2011 Learning Disabilities and Related Mild Disabilities (Minneapolis,
MN: Cengage Learning)
Lemos M C and Morehouse B J 2005 The co-production of science and policy in integrated
climate assessments Glob. Environ. Change 15 57–68
Linkersdorfer J, Lonnemann J, Lindber S, Hasselhorn M and Fiebach C J 2012 Grey matter
alterations co-localize with functional abnormalities in developmental Dyslexia: An ALE
meta-analysis PLoS One 7 1–10
Lyon G R, Shaywitz S E and Shaywitz B A 2003 A definition of dyslexia Ann. Dyslexia 53 1–14
Maisog J M, Einbinder E R, Flowers D L, Turkeltaub P E and Eden G F 2008 A meta-analysis of
functional neuroimaging studies of dyslexia Ann. N. Y. Acad. Sci. 1145 237–59
McLeod J 2003 Doing Counselling Research in Counselling (London: Sage)
Moorhead J 2005 Is dyslexia just a myth? The Guardian (Sept.) https://www.theguardian.com/
education/2005/sep/07/schools.uk
Nalavany B A and Carawan L W 2012 Perceived family support and self‐esteem: The mediational
role of emotional experience in adults with dyslexia Dyslexia 18 58–74
9-19
Nascimento I S D, Rosal A G C and Queiroga B A M D 2018 Elementary school teachers’

knowledge on dyslexia Rev. CEFAC 20 87–94
Penrod J, Preston D B, Cain R E and Starks M T 2003 A discussion of chain referral as a method
of sampling hard-to-reach populations J. Transcult. Nurs. 14 100–7
Pikulski J J and Chard D J 2005 Fluency: Bridge between decoding and reading comprehension
Read. Teach. 58 510–9
Pitt S and Soni A 2017 Students’ experiences of academic success with dyslexia: A call for
alternative intervention Support Learn. 32 387–405
Pumfrey P and Reason R 2013 Specific learning difficulties (dyslexia): Challenges and Responses
(Hoboken, NJ: Routledge)
Richlan F, Kronbichler M and Wimmer H 2009 Functional abnormalities in the dyslexic brain: a
quantitative meta-analysis of neuroimaging studies Hum. Brain Mapp. 30 3299–308
Richlan F, Kronbichler M and Wimmer H 2013 Structural abnormalities in the dyslexic brain: a
meta-analysis of voxel-based morphometry studies Hum. Brain Mapp. 34 3055–65
Riddell S and Weedon E 2006 What counts as reasonable adjustment? Dyslexic students and the
concept of fair assessment Int. Stud. Sociol. Educ. 16 57–73
Riley K and Rustique-Forrester E 2002 Working with Disaffected Students: Why Students Lose
Interest in School and What We Can do About It (London: Paul Chapman)
Rimrodt S L, Peterson D J, Denckla M B, Kaufmann W E and Cutting L E 2010 White matter
microstructural differences linked to left perisylvian language network in children with
dyslexia Cortex 46 739–49
Saygin Z M, Norton E S, Osher D E, Beach S D, Cyr A B, Ozernov-Palchik O and Gabrieli J D
2013 Tracking the roots of reading ability: white matter volume and integrity correlate with
phonological awareness in prereading and earlyreading kindergarten children J. Neurosci. 33
13251–8
Schunk D and Pajares F 2002 The development of academic self-efficacy Development of
Achievement Motivation ed A Wigfield and J S Eccles (Chichester, UK: Wiley), pp 16–29
Scott S, Webber C F, Lupart J L, Aitken N and Scott D E 2014 Fair and equitable assessment
practices for all students Assess. Educ.: Principles Policy Pract. 21 52–70
Shakespeare T 2008 Debating disability J. Med. Ethics 34 11–4
Shaw S C, Malik M and Anderson J L 2017 The exam performance of medical students with
dyslexia: a review of the literature MedEdPublish 6
Shaywitz B A, Lyon G R and Shaywitz S E 2006a The role of functional magnetic resonance
imaging in understanding reading and dyslexia Dev. Neuropsychol. 30 613–32
Shaywitz B A, Mody M and Shaywitz S E 2006b Neural mechanisms in dyslexia Curr. Dir.
Psychol. Sci. 15 278–81
Shaywitz S E and Shaywitz B A 2008 Paying attention to reading: The neurobiology of reading
and dyslexia Dev. Psychopathol. 20 1329–49
Soriano-Ferrer M, Echegaray-Bengoa J and Joshi R M 2016 Knowledge and beliefs about
developmental dyslexia in pre-service and in-service Spanish-speaking teachers Ann. Dyslexia
66 91–110
Skeide M A, Kraft I, Müller B, Schaadt G, Neef N and Brauer J et al 2016 NRSN1 associated
grey matter volume of the visual word form area reveals dyslexia before school Brain 139
2792–803
9-20
Skeide M A, Kirsten H, Kraft I, Schaadt G, Müller B and Neef N et al 2015 Genetic dyslexia risk
variant is related to neural connectivity patterns underlying phonological awareness in
children NeuroImage 118 414–21
Snowling M J 2013 Early identification and interventions for dyslexia: a contemporary view
J. Res. Spec. Educ. Needs 13 7–14
Solesbury W 2001 Evidence based policy: Whence it came and where it’s going Working Paper
Speedy J 2018 Narrative Inquiry and Psychotherapy (New York: Palgrave Macmillan)
Szalkowski C E, Fiondella C G, Galaburda A M, Rosen G D, LoTurco J J and Fitch R H 2012
Neocortical disruption and behavioral impairments in rats following in utero RNAi of
candidate dyslexia risk gene Kiaa0319 Int. J. Dev. Neurosci. 30 293–302
Szucs D and Goswami U 2007 Educational neuroscience: Defining a new discipline for the study
of mental representations Mind Brain Educ. 1 114–27
Tammimies K, Vitezic M, Matsson H, Le Guyader S, Bürglin T R, Ohman T and Tapia-Páez I
2013 Molecular networks of DYX1C1 gene show connection to neuronal migration genes
and cytoskeletal proteins Biol. Psychiatry 73 583–90
Taylor S and Littleton K 2006 Biographies in talk: A narrative-discursive research approach
Qualitative Sociol. Rev. 2 22–38
Thomson G, Howden-Chapman P and Wilson N 2007 The use and misuse of health research by
parliamentary politicians during the development of a national smokefree law Aust. N. Z.
Health Policy 24
University of Malta 2013 MATSEC regulations statute and conduct. Retrieved from http://www.
um.edu.mt/data/assets/pdf_file/0011/302123/regulationsconductfull.pdf
University of Malta 2018 The University of Malta Access Arrangements (October 2018) Retrieved
from https://um.edu.mt/_data/assets/pdf_file/0004/378481/UMGuidelinescomplete2018.pdf
Vandermosten M, Boets B, Poelmans H, Sunaert S, Wouters J and Ghesquière P 2012 A
tractography study in dyslexia: neuroanatomic correlates of orthographic, phonological and
speech processing Brain 135 935–48
Vandermosten M, Vanderauwera J, Theys C, de Vos A, Vanvooren S, Sunaert S and Ghesquière
P 2015 A DTI tractography study in pre-readers at risk for dyslexia Dev. Cognit. Neurosci. 14
8–15
Vellutino F R, Fletcher J M, Snowling M J and Scanlon D M 2004 Specific reading disability
(dyslexia): What have we learned in the past four decades? J. Child Psychol. Psychiatry 45
2–40
Vettiyadan M, Fathima Henna C K and Shibymol C B 2018 Academic self-concept and
emotional maturity among students with learning disability Indian J. Health Wellbeing 9
1685
Voicu M C and Banonea A 2011 Using the snowball method in marketing research on hidden
populations Challenges Knowl. Soc. 1 1341–51
Washburn E K, Binks-Cantrell E S and Joshi R M 2014 What do preservice teachers from the
USA and the UK know about dyslexia? Dyslexia 20 1–18
Stein J 2001 The magnocellular theory of developmental dyslexia Dyslexia 7 12–36
Williams J A and Lynch S A 2010 Dyslexia: What teachers need to know Kappa Delta Pi Record
46 66–70
Willingham D T and Lloyd J W 2007 How educational theories can use neuroscientific data Mind
Brain Educ. 1 140–9
9-21
Wise J C et al 2010 The relationship between different measures of oral reading fluency and
reading comprehension in second-grade students who evidence different oral reading fluency
difficulties Lang. Speech Hear. Serv. Schools 41 340–8
World Health Organisation 2010 International Statistical Classification of Diseases and Related
Health Problems 10th revision http://icdwho.int/browse10/2010/en
Yeatman J D, Dougherty R F, Rykhlevskaia E, Sherbondy A J, Deutsch G K, Wandell B A and
Ben-Shachar M 2011 Anatomical properties of the arcuate fasciculus predict phonological
and reading skills in children J. Cognit. Neurosci. 23 3304–17
Ziegler-Hill V, Besser A and King K 2011 Contingent self-esteem and anticipated reactions to
interpersonal rejection and achievement failure J. Soc. Clin. Psychol. 30 1069–96
Ziegler J C, Pech-Georgel C, Dufau S and Grainger J 2010 Rapid processing of letters, digits and
symbols: what purely visual‐attentional deficit in developmental dyslexia? Dev. Sci. 13 F8–14
9-22
IOP Publishing

Chapter 10
Advances in epilepsy: from gender to genetics
Suchitra Joshi and Denia Ramirez-Montealegre
Epilepsy is a complex neurological disorder associated with recurrent spontaneous

seizures which are associated with clinical features encompassing affected awareness,
abnormal movements of muscle, head or limbs, or whole-body convulsions. The
underlying causes could be genetic or acquired and the seizures can be controlled
with anti-convulsant medications. In instances where seizure control cannot be
achieved with drugs, removal or ablation of seizure focus are the available
therapeutic options. Seizures often occur in a rhythmic way and the rhythmicity
governed by female reproductive hormones is extensively characterized in women
with epilepsy. In this chapter we discuss recent advances in methods of removal of
seizure focus, mechanisms underlying female reproductive hormonal regulation of
seizures, and the role genetic testing has played in identification of genetic causes
underlying seizures and the potential for customizing therapy in such cases.
10.1 Introduction
Epilepsy is a neurological disorder characterized by the presence of recurrent
spontaneous seizures. The unpredictable nature of seizures affects the quality of a
patient’s life, often leads to loss of productivity and contributes to the economic
burden. Seizures increase the risk of accidents and may cause sudden unexpected
death in epilepsy (SUDEP). Brain malformations, brain injury due to trauma or
stroke, fever, inflammation, substance abuse, and certain genetic factors increase the
risk of epilepsy. However, diagnosis of epilepsy can be made only after the onset of
spontaneous seizures, after which seizure control with anticonvulsant medications is
the only available therapeutic option for a majority of patients. Seizures in a fraction
of patients remain refractory to two or more anticonvulsant drugs and these patients
may require surgical removal of the ictogenic region to achieve seizure freedom.
Epilepsy constitutes diverse seizure phenotypes including blank staring spells seen
in absence epilepsy, focal seizures leading to repetitive, uncontrolled jerking of limbs
or facial twitching, and severe tonic–clonic seizures also called grand mal seizures,

which involve convulsions of the entire body [1]. Understanding seizure patterns and
knowing the underlying mechanism(s) could help devise strategies to keep seizures
under control and manage their impact on day-to-day life.
In this chapter we will discuss aspects of minimally invasive surgical procedures to
remove the seizure focus and obtain seizure control. We will also discuss cyclic
changes in seizure frequency, seen particularly in women of reproductive age with
epilepsy, and discuss the role of female reproductive hormones in regulating these
fluctuations. Finally, we will discuss genetic causes underlying seizures and genetic
testing.
10.2 Minimally invasive techniques used to achieve seizure control

Seizure control can be achieved when the seizure focus is surgically removed. Some
of the invasive surgeries include hemispheroctomy—removal of an entire hemi-
sphere, lesionoctomy—lesioning of seizure foci, particularly used in cortical dyspla-
sias and tumors, loboctomy—removal of a lobe harboring the seizure focus, often in
the hippocampus or amygdala, and callosotomy—severing the axonal connections
between bilateral motor cortices are some of the invasive surgeries performed to
achieve seizure freedom or reduce seizure burden and severity [2]. However, these
techniques are associated with the risks linked to invasive surgeries, which could
affect cognitive capabilities, and require intensive rehabilitation. Furthermore,
success is highly dependent on a clear identification of seizure focus. Methods of
minimally invasive surgeries which use radiowaves, laser, or ultrasound to ablate the
seizure focus are evolving to overcome the limitations associated with traditional
epilepsy surgeries [3–5]. These techniques utilize thermal energy to destroy the
seizure focus and can be coupled to live imaging such that the surgeon can
constantly monitor critical brain functions while ablating the focus. MRI-guided
laser interstitial thermal therapy is particularly useful when difficult to access
structures need to be removed, as in the case of hypothalamic hamartomas and
deeper temporal structures in temporal lobe epilepsy [5]. These procedures are
substantially superior to the invasive procedures listed above, as they involve a
shorter hospital stay and offer means of an effective destruction of the seizure focus
along with minimal harm to the surrounding tissue. However, the rate of success
ranges between 50%–60% with these procedures compared to 80%–90% in tradi-
tional surgeries. Since these are evolving techniques the long-term outcome has not
been extensively characterized. Additionally, there is some concern regarding
inadvertent heating of the skull in cases of focused ultrasound and secondary risk
of malignancy, in the case of radiosurgery, particularly in the pediatric patient
population.
In cases of patients with seizure focus in the brain areas involved in critical
functions such as language or motor control, ablation of the seizure focus is not
possible. In such cases, brain stimulation to interfere with neuronal synchronization,
which underlies the seizures, is an option. A significant reduction in seizures has been
obtained in 30%–40% of the patients using vagus nerve stimulation [6–8].
Responsive neurostimulation is another evolving strategy for the treatment of focal
10-2
seizures; this involves continuous monitoring of focal electroencephalogram (EEG)

via transcranial electrodes coupled with cortical stimulation to suppress seizures [9].
Using this technique, a double-blind clinical trial with 191 patients found a sustained
reduction in seizure frequency in 38% of the stimulated patients compared to 17% of
sham [10]. However, for these cases to be effective, it is critical to identify EEG
patterns which precede the seizure onset.
10.3 Rhythms in seizure frequency

Across both adult and pediatric populations, seizures and interictal spikes in
different types of epilepsy often follow a pattern [11–14]. The peak and trough of
seizure frequency could fluctuate in a 24 h-long cycle or follow a cycle spanning days
or weeks. The abnormal brain activity on EEG, including the presence of interictal
spikes, can be used to predict seizures [12, 14]. Importantly, the rhythms in spike
frequency are seen in patients who do not have an obvious seizure pattern [14].
Although these patterns are often patient-specific, once identified for a patient, they
could be used to predict and/or treat seizures.
Seizure frequency data collected through the SeizureTracker app, in which
patients self-report and track seizures, has revealed seizure patterns in 80% of
patients, whereas data collected in the NeuroVista study, which involved continuous
EEG monitoring of 15 patients for months to years found specific seizure patterns in
92% of the patients [12, 13]. Sleep–wake cycle, changes in melatonin levels, and both
environmental and genetic factors are likely to contribute to this rhythmicity. In
addition, hormonal changes also impact seizure frequency. The best described
seizure rhythmicity is linked to cyclic fluctuations in female reproductive hormones,
which makes approximately 30% of women of reproductive age with epilepsy prone
to cyclic seizure exacerbation called catamenial epilepsy [15, 16]. These seizures
represent the most prevalent drug refractory seizures in women with epilepsy.
10.3.1 Hormonal regulation of seizures

The female reproductive hormones estrogen and progesterone are synthesized
peripherally and can easily cross the blood–brain barrier to modulate neuronal
activity. The availability of estrogen and progesterone to the brain is regulated in
part by the concentration of steroid binding proteins in the serum [17]. Neurons and
glia, which express all the enzymes necessary for conversion of cholesterol to
progesterone and other steroid hormones, can synthesize progesterone in a de novo
manner [18]. Thus the concentration of these hormones in the brain is dependent on
the kinetics of uptake and metabolism as well as the efficacy of de novo synthesis.
The steroidogenic enzymes are expressed in the brain throughout the development
and adulthood and across multiple species, indicating that locally synthesized
progesterone and other steroids likely regulate brain function throughout life.
Estrogen is generally believed to be proconvulsant, whereas progesterone exerts
acute anticonvulsant effects. Logothetis and colleagues found that intravenous
administration of premarin (conjugated estrogen) to women with epilepsy increased
EEG activity in a majority of patients and caused seizures in a subset of patients [19].
10-3
On the other hand, progesterone treatment suppressed seizures in some women with
epilepsy [20–22]. However, in a double-blind, phase III clinical trial, progesterone
was no different than placebo in treating seizures in women with a clear pattern of
menstrual cycle-linked seizure exacerbation or those without such a pattern [23, 24].
10.3.2 Catamenial epilepsy

A third of women of reproductive age with epilepsy may experience
intractable seizure precipitation during specific stages of the menstrual cycle. The
levels of reproductive hormones, estrogen and progesterone, are at their lowest at the
start of the cycle; the estrogen levels rise during the follicular phase and this increase
is followed by an increase in progesterone levels during the luteal phase. As both of
these hormones are potent neuromodulators and exert effects on excitatory and
inhibitory neurotransmitter receptors, the changes in their levels are reflected in
seizure frequency. A diagnosis of catamenial epilepsy is performed if the seizure
frequency during a particular phase of the cycle is twice or more than that during
other phases. Catamenial seizures can be divided into three types [16]. The type I
seizure exacerbation occurs during follicular phase and is believed to be due to high
estrogen-to-progesterone ratio. In contrast, seizures in type II occur perimenstrually
and are thought to be due to progesterone withdrawal. Seizures in the type III
catamenial epilepsy could occur during any part of the ovulatory cycle, due to an
inadequate luteal phase and failure of progesterone levels to rise to levels observed
under normal conditions.
10.3.3 Progesterone
Progesterone can exert its effects via multiple mechanisms; the best-studied
mechanism involves the actions of the progesterone metabolite allopregnanolone,
which is a potent allosteric modulator of γ-aminobutyric acid type-A receptors
(GABAARs). In addition, progesterone can also bind to and activate its cognate
receptors, the progesterone receptors (PRs), which belong to the nuclear hormone
receptor family, and mediate the genomic effects of progesterone (figure 10.1). The
role of PRs in regulating reproduction and sexual behavior is well characterized in
the components of hypothalamic–pituitary–axis (HPA). However, the role of these
receptors in regulating brain activity outside the HPA has only recently started to
emerge [25]. In addition to PRs, progesterone can also bind to membrane
progesterone receptors (mPRs) and to progesterone receptor membrane component
1 (PGRMC1), to alter cellular signaling [26, 27].
10.3.3.1 Progesterone receptor-mediated effects

PRs belong to the nuclear hormone family of receptors. Out of the several isoforms
which can be synthesized via use of alternate start sites, omission of exons, and
inclusion of intronic exons, isoforms A and B are the most widely expressed
and studied type of PRs [26, 28, 29]. PR-B is longer than PR-A by 164 amino
acids. The PR peptides are composed of several domains; the ligand binding domain
10-4
is present at the C terminus, separated from the DNA binding domain by a Hinge
region.
Various activation function domains are present at the N terminus of the protein
which are involved in interaction of PRs between themselves as well as with other
transcription factors and co-regulators. The 164 extra amino acids at the N terminus
of PR-B form an activation function three domain; this domain confers on PR-B the
capacity to interact with a unique set of transcription factors and cofactors; these
interactions are proposed to underlie the higher transcriptional activation ability of
this isoform [30, 31]. There are several cell and tissue types which express both
isoforms of PRs; however, despite the overlapping expression and interaction with
some common cofactors, function of the two isoforms appears to be distinct. The
best example of these distinctions is seen in regard to their role in regulating
reproductive functions and mammary gland development; PR-B knockout females
are fertile but have impaired mammary gland development, whereas PR-A knock-
out females are sterile with normally developed mammary glands [32, 33].
PRs are widely expressed in the brain during development and adulthood [34–37].
Estrogen is a potent regulator of PR expression and deletion of estrogen receptor
(ER) isoform α hampers PR expression [38–40]. In contrast, progesterone seems to
suppress PR expression [41]. The estrogen and progesterone-regulation of PR
expression is evident in the fluctuations in PR expression during estrous cycle [41].
Figure 10.1. Peripherally-synthesized estrogen and progesterone can modulate inhibitory and excitatory
neurotransmission. In addition, progesterone and estrogen can be locally synthesized in the brain through
enzymatic conversion of cholesterol. Progesterone is metabolized to allopregnanolone, which binds to
GABAARs and increases chloride influx. Progesterone and estrogen can also bind to PRs and ERs
respectively and trigger gene expression. Recent studies have found that progesterone increases AMPA
receptor expression potentially through a genomic mechanism dependent on PRs.
10-5
10.3.3.2 PR regulation of neuronal activity

Deletion of PRs affects secretion of reproductive hormones such as luteinizing
hormone and follicle stimulating hormone, indicating their role in regulating the
function of hypothalamic neurons [42]. In addition, PR deletion also seems to affect
the activity of kisspeptin neurons and alters the onset of puberty in females [43, 44].
Furthermore, recent studies have begun to highlight PR regulation of neuronal
activity in extrahypothalamic areas. A recent study found an increased expression of
α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) expres-
sion in the hippocampi of female rats treated with progesterone [25]. Concomitant
with the increased AMPAR expression, the AMPAR-mediated synaptic currents
recorded from CA1 pyramidal neurons were also potentiated [25]. Blockade of PRs
with RU-486 blocked these progesterone effects, whereas treatment of animals with
a specific PR agonist Nestorone mimicked these effects [25]. Since AMPARs
contribute to excitatory glutamatergic transmission, these effects are likely to
increase excitability.
The AMPAR expression in the hippocampus also fluctuates during the rodent
estrous cycle. Higher expression of these receptors and augmented AMPAR-
synaptic currents are seen in the hippocampal CA1 neurons of animals in the
estrous stage, which follows the rise in progesterone and estrogen levels during the
proestrus stage [25]. In contrast, in animals in the diestrus stage, AMPAR expression
is lower and the strength of synaptic currents is reduced [25]. Treatment of animals
with anti-progestin, RU-486, during diestrus to estrous stages blocks the estrus-
associated increase in AMPAR expression. The glutamatergic receptors are present
on dendritic spines and the density of these spines on CA1 neurons also fluctuates
during estrous cycle [45]. PR activation seems to regulate the fluctuations in spine
density, at least in part, since RU-486 treatment prevents estrous cycle-linked
changes in spine density [45].
Since activation of AMPARs would contribute to membrane depolarization
which is necessary to reach a threshold to fire action potential, the PR-regulated
changes in AMPAR expression could contribute to neuronal synchronization and
seizures. Indeed, recent studies have found a role of PR activation in regulating
seizures. Using an animal model of perimenstrual seizure exacerbation in which
progesterone levels are elevated with hormone treatment, treatment with finasteride,
which blocks conversion of progesterone to allopregnanolone, causes a dramatic
exacerbation of seizures in female epileptic animals [46, 47]. Treatment of animals
with RU-486 simultaneously with the hormone treatment reduces the extent of
progesterone or neurosteroid withdrawal-induced seizure precipitation [25].
Furthermore, treatment of animals with the PR agonist Nestorone, increases the
frequency of recurrent spontaneous seizures [48]. These findings suggest that PR
activation is detrimental to seizures. Pharmacological agents used to activate or
block PRs may have off-target effects; however, studies with PR knockout animals
also support a proconvulsant role of PRs. Mice lacking PRs are slow to kindle and
progesterone exerts an enhanced anticonvulsant effect in knockout mice [49, 50].
These findings suggest that PR-mediated effects promote kindling epileptogenesis
10-6
and that PR signaling acts against the neuroprotective effects mediated through
allopregnanolone.
10.3.3.3 Allopregnanolone regulation of neuronal activity

Progesterone induces sedative, anesthetic, and anticonvulsant effects through its
metabolite allopregnanolone. Blockade of conversion of progesterone to allopreg-
nanolone prevents the neuroprotective effects of progesterone [51]. The conversion
of progesterone to allopregnanolone occurs within minutes and the inhibitory effects
are seen acutely within minutes of progesterone application, which is distinct from
the slow but persistent effects mediated through PR activation. Allopregnanolone
acutely reduces neuronal firing [52–55], increases the threshold to electrically or
chemically-induced seizures [22, 56–60], retards epileptogenesis [61, 62], and may
also offer protection against the prolonged seizures of status epilepticus [63, 64].
Allopregnanolone exerts its effects through actions on GABAARs [55, 65–69].
These pentameric GABA-gated chloride channels are assembled from various
subunits encoded from 19 genes. Some subunits have multiple isoforms, whereas
other subunits are represented by a single isoform. Most of the receptors contain 2α,
2β and a γ or δ subunit. The subunit composition determines sensitivity to GABA
and other pharmacological agents including benzodiazepines, alcohol, neurosteroids
and also influences channel kinetics and synaptic integration of the receptors [57, 70,
71]. At physiological concentrations in the low nanomolar range, allopregnanolone
acts as a positive allosteric modulator, whereas, at higher concentrations, it can exert
an agonist-like action [72]. Activation of GABAARs mostly causes membrane
hyperpolarization via chloride influx into the neuron; hence progesterone effects
mediated through allopregnanolone are inhibitory in nature. Neurosteroid sensitiv-
ity of GABAARs depends on the subunit composition [56]; hence areas which
express the αxβxδ and α1βxγ2 subunit-containing GABAARs are more responsive
to inhibitory effects of neurosteroids than areas which have a relatively lower
expression of these receptors [73]. Allopregnanolone potentiates the synaptic, as well
as extrasynaptic, GABAAR-mediated inhibition and this can be recorded in the
form of prolongation of the decay and/or enhancement of the amplitude of
inhibitory post-synaptic currents. Extrasynaptic GABAARs mediate a slow persis-
tent form of inhibition called tonic inhibition. Allopregnanolone also potentiates this
current and increases membrane noise [55, 74, 75].
The actions of allopregnanolone in the brain are not limited to potentiation of
GABAAR currents; it also appears to regulate the GABAAR receptor expression.
For example, estrous cycle-linked changes in the expression of δ subunit-containing
GABAARs are blocked by treatment with finasteride [65]. Allopregnanolone also
appears to play a role in progesterone-induced upregulation of α2 subunit expression
[76].
10.3.3.4 Altered expression of GABAARs in epilepsy

The efficacy of allopregnanolone and other neurosteroids in keeping excitability
under control is reduced in epilepsy [54]. GABAARs with diminished neurosteroid
sensitivity are expressed in tissue isolated from epilepsy patients [77] and in the
10-7
hippocampal and cortical tissue isolated from experimental animals with epilepsy
[61, 78–81]. The expression of α1βxγ2 and αxβxδ subunit-containing receptors,
which are more sensitive to neurosteroids than other GABAARs is decreased,
whereas the expression of α4βxγ2 subunit-containing receptors, which are relatively
less sensitive to neurosteroids, is increased [54, 61, 78–84]. Neurosteroids fail to
potentiate GABAAR-mediated whole cell currents of hippocampal dentate granule
cells (DGCs) from epileptic animals [85]. Allopregnanolone and other neurosteroids
potentiate the decay of synaptic currents of DGCs in naïve animals but not in
epileptic animals [79, 80]. Furthermore, neurosteroid potentiation of tonic currents
is also reduced in epilepsy [61, 78, 80].
Some of the changes in GABAARs appear to occur before the onset of recurrent
spontaneous seizures, whereas other coincide with the appearance of spontaneous
seizures. The expression of δ subunit-containing receptors is reduced following
status epilepticus, whereas the expression of α4γ2 subunit-containing receptors is
increased in epileptic animals [61]. The expression of α4 subunits also appears to be
increased in animals which have recently experienced seizures as compared to
animals that have not [82]. The molecular mechanisms underlying the changes in
GABAARs are also different. Studies in cultured hippocampal neurons have
revealed that NMDA receptor activation triggers the down-regulation of δ sub-
unit-containing receptors [86], whereas the down regulation of α1 subunit-contain-
ing receptors is regulated by cAMP response element binding protein
(CREB)- inducible cAMP early repressor (ICER) regulated mechanisms [87]. On
the other hand, brain derived neurotrophic factor (BDNF)–tyrosine kinase B (TrkB)
receptor–early growth response factor (Egr)-regulated mechanisms underlie the
upregulation of α4 subunit-containing receptors [88, 89]. Phosphorylation of
GABAARs also appears to regulate the neurosteroid sensitivity and dephosphor-
ylation of GABAARs is associated with diminished neurosteroid potentiation in
cortical pyramidal neurons [84, 90].
Expression of GABAARs with diminished neurosteroid sensitivity in epilepsy,
combined with the opposing effects of allopregnanolone and PR, could help to
explain type II catamenial seizure exacerbation (figure 10.2). The decline in
progesterone and allopregnanolone levels during perimenstrual period would
compromise inhibitory neurotransmission. At the same time, increased glutamater-
gic transmission through AMPARs regulated through PR-mediated mechanisms
would cause an imbalance between excitation and inhibition, leading to seizure
precipitation.
10.3.4 Estrogen
Generally, estrogen exerts proconvulsant effects. A higher estrogen-to-progesterone
ratio is proposed to underlie periovulatory seizure exacerbation in catamenial
epilepsy. Studies in experimental animals have shown that estrogen treatment
decreases the threshold to evoke seizures with electrical stimulation [91–93], and
decreases the latency to kainic acid-evoked seizures [94]. Estrogens also accelerate
10-8
Figure 10.2. A schematic illustrating fluctuations in progesterone and estrogen levels during menstrual cycle.
The rising progesterone levels during the luteal phase potentiate GABAergic inhibition through neurosteroids.
The increase in progesterone levels would also activate PRs and exert slow genomic effects leading to
enhancement of AMPA receptor-mediated glutamatergic transmission. The decline in progesterone levels
towards the end of luteal phase would affect the strength of GABAergic inhibition, but the glutamatergic
neurotransmission would be maintained. This would open up a period of imbalance between excitation and
inhibition and increase seizure susceptibility during the perimenstural phase. Hence the opposing effects of
progesterone mediated through neurosteroids and through PR activation could explain perimenstrual seizure
exacerbation in catamenial epilepsy.
the rate of kindling [95]. In addition, blockade of estrogen synthesis seems to reduce
the severity of status epilepticus [96].
The cellular effects of estrogens are mediated through activation of estrogen
receptors (ERs), which like PRs, are widely expressed in the brain [97, 98].
Following ligand binding, ERs translocate to the nucleus and trigger transcription.
There are two isoforms of ER, ER-α and ER-β, which can homo- or hetero-dimerize
upon binding with the ligand and translocate to the nucleus. The ER peptide is
comprised of an N terminus A/B domain, which harbors an activation function 1
(AF1) site involved in the interaction with other transcription factors [99, 100]. This
domain is followed by a C domain which is critical for DNA binding, followed with
D/E/F domains. Ligand binding occurs through these C terminal domains and these
domains also harbor an AF2 site, involved in interaction with other transcription
factors [99, 100]. The AF1 domain in ER-α is substantially stronger than that in ER-
β [101]. There is ample expression of ER-α and ER-β in the hippocampus, frontal
cortex, and amygdala, in addition to that in the hypothalamic regions involved in
regulation of sexual functions [102].
Estrogens exert excitatory effects through glutamate receptors [103]. Estrogen
treatment of overiectomized animals triggers formation of dendritic spines on CA1
pyramidal neurons [45, 104]. Estrogens also regulate the fluctuations in the spine
density during the female reproductive cycle [105, 106]. Using electron microscopy
10-9
Woolley and colleagues have found that the increase in spine density following
estrogen treatment leads to formation of synaptic connections between a single
presynaptic neuron and multiple post-synaptic CA1 neurons [107], which predicts
that estrogen treatment is likely to increase the probability of synchronization.
Estrogen treatment of overiectomized rats prolongs the excitatory post-synaptic
potentials (EPSPs) of hippocampal CA1 neurons indicative of increased synaptic
excitability [108]. Acute application of 17β-estradiol potentiates kainic acid-evoked
currents of CA1 neurons [94], enhances the amplitude of kainic acid-induced
currents and causes a rapid and reversible increase in Schaffer collateral-activated
EPSPs. This potentiation was unaffected by blockade of NMDA receptors, but was
blocked by 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) which blocks non-
NMDA receptors. Another study has found that estradiol treatment also increases
the expression of the NR1 subunit of NMDA receptors on CA1 pyramidal neurons
[103]. Estrogen also increases long-term potentiation (LTP), which involves insertion
of AMPARs at the synapses, in hippocampal CA1 neurons [109]. These findings also
support estrogen’s neuronal effects that would increase glutamatergic synaptic
transmission. The enhancement of LTP in estradiol-treated animals was blocked
by antagonism of ERs with tamoxifen [109]. Acute application of estradiol also
increases synaptic efficacy through non-genomic mechanisms [110].
In contrast to the potentiating effects on glutamatergic receptors, estrogen
transiently suppresses GABAergic inhibition of CA1 pyramidal neurons, which
can be observed in the form of reduction in glutamic acid decarboxylase (GAD)
immunoreactivity and lowering of the frequency of miniature inhibitory post-
synaptic currents (mIPSCs) [111]. However, the amplitude of mIPSCs of CA1
neurons is unaltered by estrogen treatment suggesting that the effects are presynaptic
in nature and GABAAR expression is unlikely to be affected. The expression of ER-
α in the dorsal hippocampus also appears to be concentrated in GABAergic
interneurons [112]. However, whether the estrogen-ER-α signaling in interneurons
affects their excitability is currently unknown. Furthermore, since estrogen seems to
positively regulate glutamatergic transmission, this signaling is unlikely to dampen
the activity of interneurons. Since interneurons also provide inhibitory inputs to
other interneurons, the interneurons at the border of stratum lacunosum and
stratum radiatum, which have a stronger expression of ER-α, may regulate the
activity of other interneurons.
10.4 Genetic epilepsies

In addition to acquired epilepsies, genetic causes including mutations and copy
number variations which alter neuronal function, also cause seizures. Genetic testing
has made it feasible to identify mutations underlying epilepsy and associated
neurological deficits seen in patients. The insights obtained through these tests can
be used to devise better treatment strategies, predict the risk of developing epilepsy
in family members, provide prognosis of the disease, and offer information about
neurological deficits which may develop in the future. The first gene identified as
directly causing epilepsy was in 1995 in a Norwegian family [113] and to date, more
10-10
than 200 genes have been identified to cause epilepsy. The number is expected to
increase, so also is our knowledge about heterogeneity and pleiotropy, and how
these control the intricate process that allows for proper neuronal function.
Males and females are susceptible to certain genetic forms of epilepsy in a
disproportionate manner. These seizure disorders are linked to mutations/variations
in the X chromosome, such that some are seen exclusively in females or affect them
more, whereas others affect males [114–117]. Mutations in the protocadherin 19
(PCDH19) gene cause infantile onset epileptic encephalopathy and mental retarda-
tion in women [115]. Deficient allopregnanolone synthesis in these patients could
help explain seizure phenotype, as GABAergic inhibition would be affected. Rett-
like seizure-hypotonia syndrome, which is associated with refractory seizures and
neurodevelopmental regression, is also seen predominantly in women [118, 119]. In
contrast, Christianson syndrome, which is linked to mutations in SLC9A6, which
encodes for a sodium/hydrogen exchanger, primarily affects men, and carrier
females are normal and without neurological deficits [117].
In contrast, mutations in the genes which encode for molecules important for
voltage-gated ion channel function or for neurotransmitter receptors affect males
and females. These mutations cause either a loss of function or a gain of function.
These mutations can be inherited in an autosomal dominant (AD) or autosomal
recessive (AR) manner or arise de-novo (table 10.1). Because these defects primarily
cause disruption of the normal electrochemical gradient in neurons, identification of
the specific mutation has clinical and therapeutic implications, and opens oppor-
tunities for development of novel treatments that could compensate and even correct
the pathological defect.
10.4.1 GABAAR associated epilepsies

The GABAARs mediate inhibitory neurotransmission in the brain and mutations in
the genes encoding α1, α6, β3, γ2, and δ subunits have been linked to epileptic
encephalopathies. The α1 subunit localizes to chromosome 5q34 probably in close
proximity to GABRG2 and GABRA6, genes encoding the γ2 and α6 subunits,
respectively [120]. Mutations in GABRA1 have been associated with two epileptic
syndromes: early infantile epilepsy encephalopathy (EIEE) and idiopathic general-
ized epilepsy (IGE) [121–123]. These mutations are associated with decreased α1
subunit expression, splicing defects, protein misfolding and/or retention of the
receptor in the endothelial reticulum, increased recycling/degradation and an overall
attenuation of GABA currents [124–126]. A point mutation (R46W) in GABRA6
has been reported in a patient with childhood absence epilepsy (CAE) [127]. This
mutation is associated with impaired surface membrane targeting of the receptors in
HEK293 cells and altered gating function [128].
Mutations in GABRB3, which encodes for β3 subunits of GABAARs have been
reported in patients with Prader–Willi syndrome (PWS), Angelman syndrome, and
in patients with EIEE and CAE with or without generalized tonic–clonic seizures
[129, 130]. Most mutations described so far involve deletions and/or unbalanced
translocations in the case of PWS and Angelman syndrome, or missense mutations
10-11
Table 10.1. Summary of genes related to epilepsy channelopathies.
Gene Chromosome Receptor Inheritance Proposed mechanism
GABRA1 5q34 GABA α−1 subunit AD* Loss of function

GABRB3 15q12 GABA β−3 subunit AD Loss of function
GABRG2 5q34 GABA γ−2 subunit AD Loss of function
GABRD 1p36.33 GABA δ subunit AD Loss of function
GRIN2A 16p13.2 NMDA AD Loss of function
Increased channel opening time
GRIN2B 12p13.1 NMDA AD Gain of function
Loss of function
Haploinsufficiency
SCN1A 2q24.3 Voltage-gated sodium channel type 1 AD Reduced sodium currents in GABAergic
α-subunit (Nav1.1) interneurons
SCN2A 2q24.3 Voltage-gated sodium channel type 2, AD Possible gain of function
α-subunit (Nav1.2)
10-12
SCN3A 2q24.3 Voltage-gated sodium channel type 3, AD Possible gain of function with persistent
α-subunit (Nav1.3) currents
In some, loss of function
SCN8A 12q13 Voltage-gated sodium channel type 6, AD Mostly gain of function but small number
α-subunit (Nav1.6) can have loss of function
SCN1B 19q13.11 Voltage-gated sodium channel type I AD or AR Increased peak voltage of action potentials/
β-subunit amplitude in CA3 (hyperexcitability)
KCNA1 12p13.32 Voltage-gated potassium (Kv1.1 α AD Loss of function

subunit)
KCNA2 1p13.3 Voltage-gated potassium (Kv1.2 α AD Loss of function dominant-negative
subunit)
KCNB1 20q13.13 Voltage-gated potassium (Kv2 α subunit) AD Increase inward K+ current and cell
excitability
KCNQ2 20q13.33 Voltage-gated potassium (Kv7 α subunit) AD Loss of function
KCNQ3 8q24.22 Voltage-gated potassium (Kv7 α subunit) AD Loss of function
KCNMA1 10q22.3 Large-conductance voltage and Ca2 AR Loss of function
+
-activated K+ channel (BK channel) Increased excitability
KCND2 Voltage-activated A-type potassium ion AD Unknown
channels Kv4.2
KCNJ10 1q23.2 Potassium inward-rectifying channel AD Unknown
(Kir4.1)
CACNA1A 19p13.13 Voltage-gated calcium channel P/Q-type AD Gain of function
(CaV2.1)
CACNA1H 16P13.3 Voltage-gated calcium channel T-type Increased Gain of function
(Cav3.2) susceptibility Increased calcium influx
CACNB4 2q23.3 Voltage-dependent calcium channel AD Channel inactivation
10-13
β4-subunit
CHRNA2 8p21.2 Nicotinic acetylcholine receptor AD Loss of function
α2−subunit
CHRNA4 20q13.33 Nicotinic acetylcholine receptor AD Decreased function or loss of function
α−4 subunit
CHRNB2 1q21.3 Nicotinic acetylcholine receptor AD Decreased function or loss of function
β2−subunit
and amino acid substitutions in patients with EEIE and CAE [131–134]. Murine
models display behavior similar to that of patients with autistic spectrum disorders
[135]. Furthermore, mutations in GABRB3 either reduce current amplitudes or alter
the kinetic properties of receptors, causing a loss of function phenotype with
decreased inhibition [136]. Interestingly, GABRB3 is in close proximity to
GABRA5, the gene encoding the α5-subunit of the GABAAR [137, 138]; but,
whether this proximity has any functional association is unknown [135].
The γ2 subunit is required for synaptic integration of the receptors and patients
with mutations in GABRG2 usually present with febrile seizures, CAE, or
generalized febrile seizures-plus syndrome (GEFS+); some may also develop
myoclonic or generalized tonic–clonic seizures [139–143]. Most of these cases are
associated with missense mutations with variable penetrance [142–144]. These
mutations appear to reduce the threshold to fever-induced neurological affects,
because a brief increase of the media temperature decreases surface expression and
increases receptor endocytosis under in vitro conditions [145]. In addition, mutations
affecting zinc sensitivity, ER retention of the subunit, and impaired trafficking of the
receptors have been identified [146, 147]. Experimental murine models replicate
some of the human phenotype including abnormal behavior, abnormal EEG and
premature death [148].
The GABRD gene encodes the δ subunit of GABAARs; these receptors are
mutually exclusive with the γ2 subunit and are localized to the peri- and extra-
synaptic membrane. The tonic current is attenuated and synaptic inhibition is
altered in GABRD-null mice [55, 149, 150]. Mutations in GABRD caused by a
number of polymorphisms including exon transversion and amino acid substitu-
tions, affect membrane trafficking and channel kinetics, and are associated with
variable phenotypes including IGE, HME, and GEFS+ [151, 152].
10.4.2 Glutamate receptors associated epilepsies

Glutamate, a major excitatory neurotransmitter in the mammalian brain, acts as a
ligand for two families of receptors: ionotropic and metabotropic. The ionotropic
receptors are comprised of a large family of ligand-gated ion channels that include
N-methyl-D-aspartate (NMDA), AMPA, and kainate receptors. A combination of
subunits forms the pore that opens upon binding of its ligand to allow passage of
Na+, Ca2+, and K+ ions. The NMDARs are composed of NMDAR-1 (or GRIN1)
and NMDAR-2 (or GRIN2) which is further classified into 2A–2C, whereas AMPA
receptors are comprised of GluA1 to GluA4 subunits, and kainate receptors are
composed of GluK1 to GluK5 subunits. Activation of NMDAR requires binding of
glutamate and glycine or serine and some studies have shown that magnesium or
zinc can bind, conferring some voltage-dependent activation [153, 154].
Mutations in GRIN2A have been associated with focal epilepsy with aphasia.
The clinical spectrum varies from early-onset seizures with severe developmental
delay, speech problems and intellectual disability to later onset of seizures that can
remit during adolescence with minimal or even normal development [155–159].
Three main syndromes have been identified: Landau–Kleffner syndrome (LKS)
10-14
[156], continuous spike-wave during slow-wave sleep (CSWS) and autosomal

dominant rolandic epilepsy (ADRE) [155–159]. Electrographic abnormalities
include centro-temporal spike-wave discharges in the temporal lobe or rolandic
region that in the case of CSWS become more noticeable and continuous during
slow-wave sleep. Patients with CSWS have more severe developmental delays and
behavioral problems than children with LKS but tend to respond better to anti-
seizure medications. Patients with ADRE usually have a milder clinical phenotype
with mild developmental delays and better overall neurocognitive outcomes [160].
Some have proposed that rather than different diseases, these represent a disease
spectrum [161].
Mutations in GRIN2B are associated with mental retardation autosomal
dominant syndrome (MRD6). The seizures, when present, are highly variable and
include focal or multifocal seizures, generalized seizures, or more severe abnormal-
ities including hypsarrhythmia. About half of patients progress to refractory epilepsy
[162–164].
Deletions and missense mutations in GRIA2 and GRIA3 genes, which encode for
GluA2 and GluA3 subunits of AMPARs have also been seen in epilepsy patients
[165], and are associated with mental retardation. In addition, mutations in AMPA
receptor auxiliary proteins such as stargazin, which affect receptor trafficking are
also linked with epilepsy [166, 167]. Mutations in kainate receptors are also
associated with developmental delay, mental retardation, and autism spectrum
disorder; however, a link to epilepsy has not been found yet [165].
10.4.3 Voltage-gated sodium channels associated epilepsies

Voltage-gated sodium channels are part of a very large family of channels,
comprised of α and β subunits, that regulate Na+ transport across the membrane.
The channel is composed of four pore-forming sensors (S1–4) along with one or two
linkers, that open the pore by pivoting around a hinge at the base of the pore [168].
The α-subunits encoded by 10 different genes have specific expression patterns and
excitable properties, and are sufficient for proper expression and function of the
channel [169–173]. The NaV1.1, 1.2, 1.3 and 1.6 α-subunits are primarily expressed
in the brain [76]. In contrast, β-subunits (1–4) are less complex and can form
heterodimeric complexes with a single α subunit or homomeric complexes with
2β subunits [174–178]. The β subunits are crucial for channel activation and
inactivation as well as for regulating kinetic properties. Sodium channels are
inhibited by tetrodotoxin, anesthetics, antiarrhythmics and antiepileptics and can
be enhanced by scorpion β-toxins. These drugs bind to a common receptor site in the
pore and impede ion permeation. Furthermore, NaV1.1 and NaV1.2 can be
regulated by dopamine, acetylcholine, serotonin and other neurotransmitters via
G-protein coupled receptors using PKC and PKA pathways [179–181].
Multiple mutations in different subunits have been associated with epilepsy
syndromes and most of these mutations display clinical pleiotropy. In most cases,
these are autosomal dominant diseases, with mutations causing a gain‐of‐function
effect at the molecular and cellular levels. However, in a small number of mutations
10-15
novel mechanisms of indirect hyperexcitability have been proposed, including a loss

of function effect in GABAAR [182].
The SCN1A gene encodes for the Nav1.1 α-subunit, and mutations in this gene
are linked to the largest number of genetic epilepsy syndromes ranging from milder
forms like familial febrile seizures to more complex syndromes like GEFS+, EIEE,
and Dravet syndrome, the largest of the EIEE syndromes [183, 184]. Approximately
95% of patients with Dravet syndrome have de novo heterozygous mutations and
are at a higher risk for SUDEP [185]. Seizures often become refractory to treatment
and may transition to Lennox–Gastaut syndrome [184, 186–196]. EEG usually
shows high voltage and diffuse slow background, myoclonus, migrating focal onset
discharges progressing to multifocal and generalized spike-wave discharges [186,
187, 193–198]. Mutations, clustered at the C-terminus usually confer a gain‐of‐
function with increased and persistent inward sodium current. In some cases, the
indirect effect on GABAergic interneurons facilitates a hyperexcitable state that
results in lowering seizure threshold and epilepsy [183, 199–201]. A wide array of
mutations including missense mutations, frameshift mutations resulting in intragenic
stop codons, splice donor, deletions, insertions, and exon deletions have been
identified in patients [184, 186, 189, 191, 198, 202–214]. A large meta-analysis of
the reported mutations has revealed that severe phenotypes are associated with
changes localized outside of the pore, in the S1–S4 transmembrane regions of the
Nav1.1, and affect hydrophobic properties and disrupt the isoelectric point of the
channel. Similarly, most missense mutations in EIEE tend to occur in the outer pore
regions while mutations of less severe clinical forms like GEFS+ are located inside
the pore [215].
A murine Nav1.1 knock-out model has deficits similar to those described in
human, and the animals die before the first month of life [216]. Interestingly, the
phenotype seems to be rescued in heterozygous animals by introducing a hetero-
zygous point mutation in Scn8A gene [217]. It is posited that the introduction of the
mutation in Scn8A gene, which encodes for the Nav1.6 α-subunit compensates the
disrupted influx of sodium. Additionally, in a loss of function model using knock in
strategy, animals develop seizures within the first month of life but have a slightly
longer lifespan [218].
The SCN2A gene encodes for the Nav1.2 α-subunit of the voltage-gated sodium
channel [219, 220]. The cytoplasmic loop between domains 2 and 3 regulates
targeting of the protein to the axonal initial segment of murine hippocampal
neurons [221]. In rodents and in humans, Nav1.2 and Nav1.3 are physically linked
which suggests that the voltage-gated sodium channel pore may be formed by the
heterodimeric link of at least three different α-subunits [222, 223]. SCN2A mutations
have been linked to benign familial seizures which further classify into neonatal or
infantile (BFNIS), depending on age at seizure onset; and EIEE. In most of the
cases, these seizures resolve completely by 5–6 years of age and are free of
neurocognitive sequelae [224–230]. Nav1.2 is expressed during early development
but is replaced by Nav1.6 (SCN8A) later in life [231]; this developmental switch
could explain the resolution of seizures with age. Prolonged refractory seizures are
observed in patients with EIEE. EEG findings are variable and include abnormal
10-16
background activity, slow waves, focal spikes, focal or multifocal sharp waves,
polyspikes, and/or generalized spikes. Patients with infantile spasms present with
typical hypsarrhythmia [232, 233]. Atrophy of cortical, subcortical, and cerebellar
structures along with intracranial calcification are often observed on MRI of these
patients [231–234].
The SCN3A gene encodes the Nav1.3 α-subunit of the voltage-gated sodium
channel, which is highly expressed in the cerebellum and frontal lobe. The kinetics of
Nav1.3 are similar to Nav1.2, but Nav1.3 has slower recovery from activation and
higher hyperpolarizing potentials [235, 236]. Mutations in this subunit are associated
with familial focal epilepsy with variable foci 4 (FFEVF4) and EIEE. Missense
mutations are generally observed in patients with FFEVF4, who usually present
with focal onset seizures without or with generalized seizures on or before the second
year of life. Seizures are generally well-controlled with medication and brain MRI is
usually normal [236, 237], indicating that the deficit does not pose severe problems.
Amino acid substitutions and a gain-of-function seen in patients of EIEE are usually
associated with refractory seizures in the neonatal or early infantile period.
Furthermore, brain imaging is abnormal and includes thinning of corpus callosum,
delayed myelination and polymicrogyria. Cognitive outcome is also meagre [238].
The mutations in Nav1.6α, encoded by the SCN8A gene, are linked to EIEE,
benign familial infantile epilepsies (BFIS) or familial myoclonic epilepsy (FME).
Seizures along with cognitive impairment with or without cerebellar ataxia are seen
in patients. Nav1.6 α is expressed on distal dendrites of CA1 pyramidal neurons
[239] and plays an important role in the initial phase of membrane depolarization.
Although murine models have motor and coordination problems, spontaneous
seizures are not observed [240–242]. Thus, the mechanism contributing to seizures in
patients is currently unclear.
The SCN1B gene encodes the β1-subunit of voltage-gated sodium channels,
which is responsible for modulating voltage-dependence, channel gating, and
channel expression at the cell surface [243, 244]. Missense mutations, translocations,
transversions, and amino acid substitution have been reported in young patients
with seizures [243, 245–248].
10.4.4 Voltage-gated potassium channels associated epilepsies

A growing number of voltage-gated potassium channel mutations (VGKC) have
been associated with genetic epilepsies. KCNQ genes encode a growing family of
single pore-loop VGKC α-subunits that have been associated with a large number of
cardiac and neurologic conditions. The KCNQ2 and KCNQ3 genes are of
particular importance in early brain development and mutations in these genes
have been associated with the diverse group of EEIE. Both genes encode for VGKC
that regulate outward potassium current (M-current) in neurons. These channels are
activated by changes in the transmembrane potential and have two domains that are
independent from each other: a pore and the voltage-sensor domain. The pore is
formed by four subunits and a single potassium ion is allowed to flow at a time [249–
252]. Most of the mutations identified in patients with epilepsy cause a loss of
10-17
function or alter inward current of potassium. Aside from KCMA1 mutations, all
are inherited in an autosomal dominant fashion and some have variable expression
and/or are expressed only during specific developmental periods.
Mutations in KCNQ2 have been identified in patients with early-onset seizures.
Although seizures usually remit around 4 years of age, the patients can have
significant long-term neurocognitive deficits and intellectual disability. In contrast to
KCNQ2 mutations, those in KCNQ3 have been associated with three different
phenotypes; the milder benign familial neonatal epilepsy (BNFE), which can be
easily controlled with anti-seizure medications and spontaneously remit before the
first year of life [253–256]. The second form includes benign familial infantile
epilepsy (BFIE), is even more clinically diverse and at least three different loci have
been associated with this syndrome: the PRRT2 and SCN2A genes, suggesting
clinical pleiotropy [257–259]. These patients also respond well to medications and
seizures remit around the first year of life. The most severe clinical form occurs with
or without seizures, but all patients will have developmental delays that can progress
to intellectual disability. Seizures may remit or persist and some patients have
cortical blindness [260–263]. There is no specific phenotype/genotype correlation
that could aid with determining risk for intellectual disability.
In addition to KCNQ genes, KCNX genes encode a specific subtype of VGKC
(shaker-related voltage-gated), which play important roles in shaping action
potentials in neurons and other cell types [264, 265]. Two genes within this family
have been identified in a small cohort of patients with epileptic syndromes.
Mutations in KCNA1 mutations have been primarily associated with episodic
ataxia type-1 (EA1), but a small number of patients also have seizures [266–268]. In
contrast, mutations in KCNA2 are associated with febrile, myoclonic, myoclonic–
atonic, focal, generalized and/or hemiclonic seizures. Seizures can remit in half of the
patients in late childhood/early adolescence, but all patients have developmental
delays, residual cognitive problems and intellectual disability. Some patients also
have speech problems and abnormal movements, with myoclonus, ataxia and
tremor being the most commonly reported [269, 270].
Mutations in other potassium channels encoded by KCNB1, KCNMA1
(α-subunit of the complex BK channel), KCND2 and D7, and KCNJ10 genes are
also associated with epilepsies in a minor fraction of patients [151, 271–276].
10.4.5 Voltage-gated calcium channels associated epilepsies

Voltage-gated calcium channels (VGCC) are responsible for transporting Ca2+
across the membrane as well as allowing for specific functions including hormonal
release, muscle contraction, and regulation of gene expression. The VGCC are
broadly classified based on their voltage sensitivities into high and low voltage, and
further subclassified depending on the structural conformation of their α1-subunits,
which conform the pore of the channel (table 10.2). Each channel has ancillary
subunits associated with the main α1-subunit, namely β, δ, γ, that confer specific
kinetic and pharmacologic properties to the channel [277–281]. The pore is formed
by the combination of four homologous α1-membrane domains conforming a
10-18
Table 10.2. Voltage-gated calcium subtypes, subunit composition and expression.
Subtype Subunits Conformation Tissue
L-type Cav1.1–1.4 High voltage 1.1: skeletal muscle

1.2: smooth muscle, pancreas, fibroblasts
and brain (dendrites/dendritic spines)
1.3: smooth muscle, fibroblasts, heart
1.4: eye
P/Q-type Cav2.1 High voltage Brain (cerebellum)

N-type Cav2.2 High voltage CNS/PNS
R-type Cav2.3 High voltage Brain (cerebellum)
T-type Cav3.1–3.3 Low voltage 3.1: heart and brain (VP nucleus)
3.2: smooth muscle, brain (thalamus)
3.3: brain (thalamus)
structure of six membrane spanning helices (S1–S6). Functionally, VGCC are tightly
regulated and any disruption to the function of the channel can result in abnormal
signaling.
The CACNA1A gene codifies for the transmembrane pore-forming subunit of the
P/Q (Cav2.1) VGCC [282]. Epilepsy associated mutations in these channels have
been reported by the EpiK Consortium and Epilepsy Phenome/Genome Project. In
most patients, the clinical phenotype is consistent with EIEE and seizures will begin
shortly after birth, with myoclonic, focal, tonic–clonic, tonic and generalized
seizures with status epilepticus being described. EEG showed generalized slowing,
polyspike-wave discharges and/or multifocal discharges. However, functional anal-
ysis characterizing cellular mechanisms underlying the effect of these mutations is
currently lacking.
The CACNA1H gene encodes for the α1-H subunit (Cav3.2) of the T-type
VGCC. The protein contains the classical four-domain structure, pore loops and
voltage sensors described in VGCC. Kinetics, conductance and voltage-dependence
is characteristic of T-type channels but it doesn’t have an inactivating binding
domain for calcium and it is inactivated via G-protein β2γ2 subunits (GNB2/GNG2)
[283–285]. In the brain, it is highly expressed in the caudate, putamen and amygdala
but it is also heavily expressed in the heart, kidney and liver [283, 286]. Most of the
patients were identified during screening of large sample sets of patients with
epilepsy and seizures phenotype include absence epilepsy, myoclonic/astatic
epilepsy, generalized epilepsy, and febrile seizures [287–289]. In some cases, the
mutations were identified in unaffected controls and this has led to the hypothesis
that mutations in Cav3.2 may be part of a multifactorial disorder and that mutations
may predispose to seizures but not be fully responsible for causing epilepsy
[287, 288].
10-19
The CACNB4 gene encodes for the β4 isoform of the VGCC. Similar to other
voltage-gated channels, β-subunits regulate kinetics and exert their control by
binding to the α-subunits through an α-interaction domain (AID) and are further
stabilized by two β-binding domains (BID) that are important for the structure but
do not participate directly in binding α-subunits: a guanylate kinase domain and a
Src homology-3 (SH3) domain [290, 291]. Mutations in CACNB4 have been linked
to IGE, JME and absence epilepsy (classical and juvenile). The mutated β4 subunits
are unable to bind the α-subunit, which leads to defective inactivation of the
channel. Mutations in CACNB4 have also been associated with episodic ataxia
[292].
10.4.6 Nicotinic acetyl-choline receptor associated epilepsies

Mammalian nicotinic acetylcholine receptors (nAChRs) belong to the superfamily
of ligand-gated ion channels that have a Cys-loop [293–296]. It is a large and
complex family of channels conformed by multiple α and β-subunits. All nAChRs
are composed of five subunits arranged around a water-filled pore [297–305]. With
multiple options for assembly, nAChRs display significant structural and functional
diversity and depending on the subunit composition, the kinetics, ion conductance,
cation selectivity, sites for specific inhibitors and overall properties will differ. Even
when in theory, subunits can combine in almost any way, there are combinations
that are favored and some are only expressed in the brain and/or in specific organs.
In the brain, the most common nAChR combinations are the α4β2 heteromeric
receptor—which has high nicotinic affinity— and the α7 homomeric receptor which
primarily binds α bungarotoxin in the brain. In the α4β2 channel, a single molecule
of acetylcholine binds specifically between the α4 and β2 subunits.
The nAChR has three different conformational states: the closed state at rest, the
open state, and the desensitized state. In the closed state, hydrophobic residues
prevent cations from being transported [296, 306]. Upon removal of the hydro-
phobic residues radially and away from the pore, the channel opens and allows for
the transport of Na+, K+ or Ca2+ [296, 305, 307, 308]. Which cation is being
transported depends solely on the affinity dictated by the subunits forming the pore,
but Na+ and K+ are usually favored. Upon binding of the agonist, the channel opens
for milliseconds prior to closing and reaching a desensitized state, one in which even
in the presence of agonists, the pore will not be able to re-open [309–312].
Mutations in some of the subunits of nAChR have been associated with nocturnal
frontal lobe epilepsy. Although different genes in different loci have been identified,
they all result in a similar phenomenon: disruption or complete loss of function of
the receptor, representing a classical example of genetic heterogeneity. The
CHRNA2 and CHRNA4 genes encode for the α2 and α4 subunits of the
nAChR, whereas CHRNB2 gene encodes for the β2-subunit. Mutations in these
subunits cause nocturnal frontal lobe epilepsy [113, 313–322], in which seizures are
characterized by clusters of sleep-related hypermotor seizures, which can be
controlled by oxcarbazepine [313, 323].
10-20
10.5 Conclusions
Epilepsy is comprised of a spectrum of seizure disorders with genetic or non-genetic
underlying causes. The seizures occur in rhythms in a majority of patients and
particularly in women of reproductive age. Female reproductive hormones affect
seizure susceptibility and contribute to clustering of seizures during particular phases
of the menstrual cycle. Modulation of neuronal activity by progesterone and
estrogen through complex cellular mechanisms and neuroactive metabolites under-
lies this hormonal regulation of seizures. Although a diagnosis of epilepsy is made
only after the onset of neurological symptoms understanding the underlying causes
could help devise strategies of seizure control and genetic testing can play a critical
role in this regard [324]. Finally, evolution of minimally invasive techniques of
ablation of seizure focus may now aid in attaining seizure freedom without the risks
of invasive brain surgeries.
References
[1] Fisher R S, Cross J H and French J A et al 2017 Operational classification of seizure types
by the International League Against Epilepsy: Position paper of the ILAE Commission for
Classification and Terminology Epilepsia 58 522–30
[2] Rugg-Gunn F, Miserocchi A and McEvoy A 2019 Epilepsy surgery Pract. Neurol. 2019
002192
[3] LaRiviere M J and Gross R E 2016 Stereotactic laser ablation for medically
intractable epilepsy: The next generation of minimally invasive epilepsy surgery Front.
Surg. 3 64
[4] Narayanasamy G, Morrill S and Cousins D et al 2019 Radiosurgery for mesial temporal
lobe epilepsy following ROSE trial guidelines—A planning comparison between Gamma
Knife, Eclipse, and Brainlab J. Appl. Clin. Med. Phys. 20 134–41
[5] Monteith S, Sheehan J and Medel R et al 2013 Potential intracranial applications of
magnetic resonance-guided focused ultrasound surgery J. Neurosurg. 118 215–21
[6] DeGiorgio C M, Schachter S C and Handforth A et al 2000 Prospective long-term study of
vagus nerve stimulation for the treatment of refractory seizures Epilepsia 41 1195–200
[7] 1995 A randomized controlled trial of chronic vagus nerve stimulation for treatment of
medically intractable seizures Neurology 45 224
[8] Handforth A, DeGiorgio C M and Schachter S C et al 1998 Vagus nerve stimulation
therapy for partial-onset seizures Neurology 51 48
[9] Karsy M, Guan J, Ducis K and Bollo R J 2016 Emerging surgical therapies in the treatment
of pediatric epilepsy Transl. Pediatr. 5 67–78
[10] Morrell M J 2011 Responsive cortical stimulation for the treatment of medically
intractable partial epilepsy Neurology 77 1295
[11] Quigg M 2000 Circadian rhythms: interactions with seizures and epilepsy Epilepsy Res. 42
43–55
[12] Karoly P J, Freestone D R and Boston R et al 2016 Interictal spikes and epileptic seizures:
their relationship and underlying rhythmicity Brain 139 1066–78
[13] Sánchez Fernández I and Loddenkemper T 2018 Chronotherapeutic implications of cyclic
seizure patterns Nat. Rev. Neurol. 14 696–7
10-21
[14] Baud M O, Kleen J K and Mirro E A et al 2018 Multi-day rhythms modulate seizure risk in
epilepsy Nat. Commun. 9 88
[15] Herzog A G 2008 Catamenial epilepsy: Definition, prevalence pathophysiology and
treatment Seizure 17 151–9
[16] Herzog A G, Klein P and Ransil B J 1997 Three patterns of catamenial epilepsy Epilepsia
38 1082–8
[17] Hammond G L 2016 Plasma steroid-binding proteins: primary gatekeepers of steroid
hormone action J. Endocrinol. 230 R13–25
[18] Stoffel-Wagner B 2003 Neurosteroid biosynthesis in the human brain and its clinical
implications Ann. N. Y. Acad. Sci. 1007 64–78
[19] Logothetis J, Harner R, Morrell F and Torres F 1959 The role of estrogens in catamenial
exacerbation of epilepsy Neurology 9 352
[20] Herzog A G 1986 Intermittent progesterone therapy and frequency of complex partial
seizures in women with menstrual disorders Neurology 36 1607–10
[21] Herzog A G 1999 Progesterone therapy in women with epilepsy: A 3-year follow-up
Neurology 52 1917–8
[22] Wu Y V and Burnham W M 2018 Progesterone, 5α-dihydropogesterone and allopreg-
nanolone’s effects on seizures: A review of animal and clinical studies Seizure 63 26–36
[23] Dana-Haeri J and Richens A 1983 Effect of Norethisterone on seizures associated with
menstruation Epilepsia 24 377–81
[24] Herzog A G, Fowler K M and Smithson S D et al 2012 Progesterone vs placebo therapy for
women with epilepsy: A randomized clinical trial Neurology 78 1959–66
[25] Joshi S, Sun H, Rajasekaran K, Williamson J, Perez-Reyes E and Kapur J 2018 A novel
therapeutic approach for treatment of catamenial epilepsy Neurobiol. Dis. 111 127–37
[26] Brinton R D, Thompson R F and Foy M R et al 2008 Progesterone receptors: Form and
function in brain Front. Neuroendocrinol. 29 313–39
[27] Mani S K and Blaustein J D 2012 Neural progestin receptors and female sexual behavior
Neuroendocrinology 96 152–61
[28] Li X and O’Malley B W 2003 Unfolding the action of progesterone receptors J. Biol. Chem.
278 39261–4
[29] Conneely O M, Mulac-Jericevic B and Lydon J P 2003 Progesterone-dependent regulation
of female reproductive activity by two distinct progesterone receptor isoforms Steroids 68
771–8
[30] Richer J K, Jacobsen B M, Manning N G, Abel M G, Wolf D M and Horwitz K B 2002
Differential gene regulation by the two progesterone receptor isoforms in human breast
cancer cells J. Biol. Chem. 277 5209–18
[31] Tetel M J, Leonhardt S A, Edwards D P, McDonnell D P and Giangrande P H 1999
Hormone-dependent interaction between the amino- and carboxyl-terminal domains of
progesterone receptor in vitro and in vivo Mol. Endocrinol. 13 910–24
[32] Mulac-Jericevic B, Lydon J P, DeMayo F J and Conneely O M 2003 Defective mammary
gland morphogenesis in mice lacking the progesterone receptor B isoform Proc. Natl. Acad.
Sci. 100 9744–9
[33] Mulac-Jericevic B, Mullinax R A, DeMayo F J, Lydon J P and Conneely O M 2000
Subgroup of reproductive functions of progesterone mediated by progesterone receptor-B
isoform Science 289 1751–4
10-22
[34] Hagihara K, Hirata S, Osada T, Hirai M and Kato J 1992 Distribution of cells containing
progesterone receptor mRNA in the female rat di- and telencephalon: an in situ hybrid-
ization study Mol. Brain Res. 14 239–49
[35] Guerra-Araiza C, Coyoy-Salgado A and Camacho-Arroyo I 2002 Sex differences in the
regulation of progesterone receptor isoforms expression in the rat brain Mol. Brain. Res. 59
105–9
[36] Camacho-Arroyo I, Guerra-Araiza C and Cerbon M A 1998 Progesterone receptor
isoforms are differentially regulated by sex steroids in the rat forebrain Neuroreport 9
3993–6
[37] Quadros P S, Pfau J L and Wagner C K 2007 Distribution of progesterone receptor
immunoreactivity in the fetal and neonatal rat forebrain J. Comput. Neurol. 504 42–56
[38] Maclusky N J and McEwen B S 1978 Oestrogen modulates progestin receptor concen-
trations in some rat brain regions but not in others Nature 274 276–8
[39] Hewitt S C and Korach K S 2000 Progesterone action and responses in the αERKO mouse
Steroids 65 551–7
[40] Shughrue P J, Lubahn D B, Negro-Vilar A, Korach K S and Merchenthaler I 1997
Responses in the brain of ER-α-disrupted mice Proc. Natl. Acad. Sci. U.S.A. 94 11008–12
[41] Guerra-Araiza C, Villamar-Cruz O, Gonźlez-Arenas A, Chavira R and Camacho-Arroyo I
2003 Changes in progesterone receptor isoforms content in the rat brain during the oestrous
cycle and after oestradiol and progesterone treatments J. Neuroendocrinol. 15 984–90
[42] Lydon J P, DeMayo F J and Funk C R et al 1995 Mice lacking progesterone receptor
exhibit pleiotropic reproductive abnormalities Genes Dev. 9 2266–78
[43] Stephens S B Z, Tolson K P and Rouse J et al 2015 Absent progesterone signaling in
kisspeptin neurons disrupts the LH surge and impairs fertility in female mice Endocrinology
156 3091–7
[44] Kubota K, Cui W and Dhakal P et al 2016 Rethinking progesterone regulation of female
reproductive cyclicity Proc. Natl. Acad. Sci. U.S.A. 113 4212–7
[45] Woolley C S and McEwen B S 1993 Roles of estradiol and progesterone in regulation of
hippocampal dendritic spine density during the estrous cycle in the rat J. Comput. Neurol.
336 293–306
[46] Lawrence C, Martin B S, Sun C, Williamson J and Kapur J 2010 Endogenous neurosteroid
synthesis modulates seizure frequency Ann. Neurol. 67 689–93
[47] Reddy D S, Kim H Y and Rogawski M A 2001 Neurosteroid withdrawal model of
perimenstrual catamenial epilepsy Epilepsia 42 328–36
[48] Shiono S, Williamson J, Kapur J and Joshi S 2019 Progesterone receptor activation
regulates seizure susceptibility Ann. Clin. Transl. Neurol. 6 1302–10
[49] Reddy D S and Mohan A 2011 Development and persistence of limbic epileptogenesis are
impaired in mice lacking progesterone receptors J. Neurosci. 31 650–8
[50] Reddy D S, Gangisetty O and Briyal S 2010 Disease-modifying activity of progesterone in
the hippocampus kindling model of epileptogenesis Neuropharmacology 59 573–81
[51] Kokate T G, Banks M K, Magee T, Yamaguchi S and Rogawski M A F 1999 A 5α-
reductase inhibitor, blocks the anticonvulsant activity of progesterone in mice J. Pharmacol.
Exp. Ther. 288 679–84
[52] Grassi S, Frondaroli A, Dieni C, Dutia M B and Pettorossi V E 2007 Neurosteroid
modulation of neuronal excitability and synaptic transmission in the rat medial vestibular
nuclei Eur. J. Neurosci. 26 23–32
10-23
[53] Jin X, Zhong W and Jiang C 2013 Time-dependent modulation of GABAA-ergic synaptic
transmission by allopregnanolone in locus coeruleus neurons of Mecp2-null mice Am. J.
Physiol. Cell Physiol. 305 C1151–60
[54] Peng Z, Huang C S, Stell B M, Mody I and Houser C R 2004 Altered expression of the δ
subunit of the GABAA receptor in a mouse model of temporal lobe epilepsy J. Neurosci. 24
8629–39
[55] Stell B M, Brickley S G, Tang C Y, Farrant M and Mody I 2003 Neuroactive steroids
reduce neuronal excitability by selectively enhancing tonic inhibition mediated by δ subunit-
containing GABAA receptors Proc. Natl. Acad. Sci. U.S.A. 100 14439–44
[56] Belelli D and Lambert J J 2005 Neurosteroids: endogenous regulators of the GABAA
receptor Nat. Rev. Neurosci. 6 565–75
[57] Joshi S, Rajasekaran K and Kapur J 2013 GABAergic transmission in temporal lobe
epilepsy: The role of neurosteroids Exp. Neurol. 244 36–42
[58] Reddy D S and Rogawski M A 2012 Neurosteroids - Endogenous regulators of seizure
susceptibility and role in the treatment of epilepsy Jasper’s Basic Mechanisms of Epilepsies
4th edn ed J L Noebels, M A Rogawski, R W Olsen and A V Delgado-Escueta (Bethesda,
MD: Oxford University Press), pp 984–1002
[59] Martin-Garcia E and Pallares M 2005 The intrahippocampal administration of the
neurosteroid allopregnanolone blocks the audiogenic seizures induced by nicotine Brain
Res. 1062 144–50
[60] Gasior M, Carter R B, Goldberg S R and Witkin J M 1997 Anticonvulsant and behavioral
effects of neuroactive steroids alone and in conjunction with diazepam J. Pharmacol. Exp.
Therap. 282 543
[61] Joshi S, Rajasekaran K, Williamson J and Kapur J 2017 Neurosteroid-sensitive δ-GABAA
receptors: A role in epileptogenesis? Epilepsia 58 494–504
[62] Biagini G, Longo D and Baldelli E et al 2009 Neurosteroids and epileptogenesis in the
pilocarpine model: evidence for a relationship between P450scc induction and length of the
latent period Epilepsia 50 53–8
[63] Kokate T G, Cohen A L, Karp E and Rogawski M A 1996 Neuroactive steroids protect
against pilocarpine- and kainic acid-induced limbic seizures and status epilepticus in mice
Neuropharmacology 35 1049–56
[64] Zolkowska D, Wu C-Y and Rogawski M A 2018 Intramuscular allopregnanolone and
ganaxolone in a mouse model of treatment-resistant status epilepticus Epilepsia 59 220–7
[65] Maguire J and Mody I 2007 Neurosteroid synthesis-mediated regulation of GABAA
receptors: Relevance to the ovarian cycle and stress J. Neurosci. 27 2155–62
[66] Smith S S, Shen H, Gong Q H and Zhou X 2007 Neurosteroid regulation of GABAA
receptors: Focus on the α4 and δ subunits Pharmacol. Therap. 116 58–76
[67] Gangisetty O and Reddy D S 2010 Neurosteroid withdrawal regulates GABAA receptor
α4-subunit expression and seizure susceptibility by activation of progesterone receptor-
independent early growth response factor-3 pathway Neuroscience 170 865–80
[68] Maguire J and Mody I 2007 Neurosteroid synthesis-mediated regulation of GABA(A)
receptors: relevance to the ovarian cycle and stress J. Neurosci. 2007 27
[69] Wu X, Gangisetty O, Carver C M and Reddy D S 2013 Estrous cycle regulation of
extrasynaptic δ-containing GABAA receptor-mediated tonic inhibition and limbic epilepto-
genesis J. Pharmacol. Exp. Therap. 346 146–60
10-24
[70] Olsen R W and Sieghart W 2009 GABAA receptors: Subtypes provide diversity of function
and pharmacology Neuropharmacology 56 141–8
[71] Sieghart W 2006 Structure, pharmacology, and function of GABAA receptor subtypes Adv.
Pharmacol. 54 231–63
[72] Bianchi M T and MacDonald R L 2003 Neurosteroids shift partial agonist activation of
GABAA receptor channels from low- to high-efficacy gating patterns J. Neurosci. 23
10934–43
[73] Belelli D and Herd M B 2003 The contraceptive agent provera enhances GABAA receptor-
mediated inhibitory neurotransmission in the rat hippocampus: Evidence for endogenous
neurosteroids? J. Neurosci. 23 10013
[74] Mtchedlishvili Z and Kapur J 2006 High-affinity, slowly desensitizing GABAA receptors
mediate tonic inhibition in hippocampal dentate granule cells Mol. Pharmacol. 69 564–75
[75] Brickley S and Mody I 2012 Extrasynaptic GABAA receptors: Their function in the CNS
and implications for disease Neuron 73 23–34
[76] Reddy D S, Gangisetty O and Wu X 2017 PR-independent neurosteroid regulation of
α2-GABAA receptors in the hippocampus subfields Brain Res. 1659 142–7
[77] Jansen L A, Peugh L D and Ojemann J G 2008 GABAA receptor properties in catastrophic
infantile epilepsy Epilepsy Res. 81 188–97
[78] Rajasekaran K, Joshi S, Sun C, Mtchedlishvilli Z and Kapur J 2010 Receptors with low
affinity for neurosteroids and GABA contribute to tonic inhibition of granule cells in
epileptic animals Neurobiol. Dis. 40 490–501
[79] Sun C, Mtchedlishvili Z, Erisir A and Kapur J 2007 Diminished neurosteroid sensitivity of
synaptic inhibition and altered location of the alpha4 subunit of GABAA receptors in an
animal model of epilepsy J. Neurosci. 27 12641–50
[80] Zhang N, Wei W, Mody I and Houser C R 2007 Altered localization of GABAA receptor
subunits on dentate granule cell dendrites influences tonic and phasic inhibition in a mouse
model of epilepsy J. Neurosci. 27 7520–31
[81] Schwarzer C, Tsunashima K, Wanzenbock C, Fuchs K, Sieghart W and Sperk G 1997
GABAA receptor subunits in the rat hippocampus II: altered distribution in kainic acid-
induced temporal lobe epilepsy Neuroscience 80 1001–17
[82] Grabenstatter H L, Cogswell M and Cruz Del Angel Y et al 2014 Effect of spontaneous
seizures on GABAA receptor α4 subunit expression in an animal model of temporal lobe
epilepsy Epilepsia 55 1826–33
[83] González M I, Grabenstatter H L and Cea-Del Rio C A et al 2015 Seizure-related
regulation of GABAA receptors in spontaneously epileptic rats Neurobiol. Dis. 77 246–56
[84] Kia A, Ribeiro F, Nelson R, Gavrilovici C, Ferguson S S G and Poulter M O 2011
Kindling alters neurosteroid-induced modulation of phasic and tonic GABAA receptor-
mediated currents: role of phosphorylation J. Neurochem. 116 1043–56
[85] Mtchedlishvili Z, Bertram E H and Kapur J 2001 Diminished allopregnanolone enhance-
ment of GABAA receptor currents in a rat model of chronic temporal lobe epilepsy
J. Physiol. 537 453–65
[86] Joshi S and Kapur J 2013 NMDA receptor activation down–regulates expression of δ
subunit-containing GABAA receptors in cultured hippocampal neurons Mol. Pharmacol.
84 1–11
[87] Lund I V, Hu Y and Raol Y H et al 2008 BDNF selectively regulates GABAA receptor
transcription by activation of the JAK/STAT pathway Sci. Signal. 1 ra9
10-25
[88] Roberts D S, Hu Y, Lund I V, Brooks-Kayal A R and Russek S J 2006 Brain-derived

neurotrophic factor (BDNF)-induced synthesis of early growth response factor 3 (Egr3)
controls the levels of type A GABA receptor α4 subunits in hippocampal neurons J. Biol.
Chem. 281 29431–5
[89] Roberts D S, Raol Y H and Bandyopadhyay S et al 2005 Egr3 stimulation of GABRA4
promoter activity as a mechanism for seizure-induced up-regulation of GABAA receptor α4
subunit expression Proc. Natl. Acad. Sci. U.S.A. 102 11894–9
[90] Gavrilovici C, D’Alfonso S, Dann M and Poulter M O 2006 Kindling-induced alterations
in GABAA receptor-mediated inhibition and neurosteroid activity in the rat piriform cortex
Eur. J. Neurosci. 24 1373–84
[91] Edwards H E, Burnham W M, Mendonca A, Bowlby D A and MacLusky N J 1999 Steroid
hormones affect limbic afterdischarge thresholds and kindling rates in adult female rats
Brain Res. 838 136–50
[92] Buterbaugh G G and Hudson G M 1991 Estradiol replacement to female rats facilitates
dorsal hippocampal but not ventral hippocampal kindled seizure acquisition Exp. Neurol.
111 55–64
[93] Buterbaugh G G 1989 Estradiol replacement facilitates the acquisition of seizures kindled
from the anterior neocortex in female rats Epilepsy Res. 4 207–15
[94] Woolley C S 2000 Estradiol facilitates kainic acid—induced, but not flurothyl-induced,
behavioral seizure activity in adult female rats Epilepsia 41 510–5
[95] Ahmad A and Vohora D 2014 Proconvulsant effects of estriol, the third estrogen, in the
mouse PTZ-kindling model Neurol. Sci. 35 1561–6
[96] Sato S M and Woolley C S 2016 Acute inhibition of neurosteroid estrogen synthesis
suppresses status epilepticus in an animal model eLife 5 e12917
[97] Weiser M J, Foradori C D and Handa R J 2008 Estrogen receptor β in the brain: from form
to function Brain Res. Rev. 57 309–20
[98] Sugiyama N, Andersson S and Lathe R et al 2009 Spatiotemporal dynamics of the
expression of estrogen receptors in the postnatal mouse brain Mol. Psychiatry 14 223–32
[99] Nilsson S, Mäkelä S and Treuter E et al 2001 Mechanisms of estrogen action Physiol. Rev.
81 1535–65
[100] Berry M, Metzger D and Chambon P 1990 Role of the two activating domains of the
oestrogen receptor in the cell-type and promoter-context dependent agonistic activity of the
anti-oestrogen 4-hydroxytamoxifen EMBO J. 9 2811–8
[101] Cowley S M and Parker M G 1999 A comparison of transcriptional activation by ER alpha
and ER beta J. Steroid Biochem. Mol. Biol. 69 165–75
[102] Mitra S W, Hoskin E and Yudkovitz J et al 2003 Immunolocalization of estrogen receptor β
in the mouse brain: Comparison with estrogen receptor α Endocrinology 144 2055–67
[103] Gazzaley A H, Weiland N G, McEwen B S and Morrison J H 1996 Differential regulation
of NMDAR1 mRNA and protein by estradiol in the rat hippocampus J. Neurosci. 16 6830
[104] Woolley C S and McEwen B S 1994 Estradiol regulates hippocampal dendritic spine density
via an NMDAR-dependent mechanism J. Neurosci. 14 7680
[105] Woolley C S and McEwen B S 1992 Estradiol mediates fluctuation in hippocampal synapse
density during the estrous cycle in the adult rat J. Neurosci. 12 2549–54
[106] Woolley C S, Gould E, Frankfurt M and McEwen B S 1990 Naturally occurring fluctuation
in dendritic spine density on adult hippocampal pyramidal neurons J. Neurosci. 10 4035–9
10-26
[107] Yankova M, Hart S A and Woolley C S 2001 Estrogen increases synaptic connectivity
between single presynaptic inputs and multiple postsynaptic CA1 pyramidal cells: A serial
electron-microscopic study Proc. Natl. Acad. Sci. 98 3525
[108] Wong M and Moss R L 1992 Long-term and short-term electrophysiological effects of
estrogen on the synaptic properties of hippocampal CA1 neurons J. Neurosci. 12 3217
[109] Smith C C and McMahon L L 2005 Estrogen-induced increase in the magnitude of long-
term potentiation occurs only when the ratio of NMDA transmission to AMPA trans-
mission is increased J. Neurosci. 25 7780
[110] Foy M R, Xu J, Xie X, Brinton R D, Thompson R F and Berger T W 1999 17β-Estradiol
enhances NMDAR-mediated EPSPs and long-term potentiation J. Neurophysiol. 81 925–9
[111] Rudick C N and Woolley C S 2001 Estrogen regulates functional inhibition of hippocampal
CA1 pyramidal cells in the adult female rat J. Neurosci. 21 6532
[112] Weiland N G, Orikasa C, Hayashi S and McEwen B S 1997 Distribution and hormone
regulation of estrogen receptor immunoreactive cells in the hippocampus of male and
female rats J. Comput. Neurol. 388 603–12
[113] Steinlein O K, Mulley J C and Propping P et al 1995 A missense mutation in the neuronal
nicotinic acetylcholine receptor alpha 4 subunit is associated with autosomal dominant
nocturnal frontal lobe epilepsy Nat. Genet. 11 201–3
[114] de Lange I M, Helbig K L and Weckhuysen S et al 2016 De novo mutations of KIAA2022
in females cause intellectual disability and intractable epilepsy J. Med. Genet. 53 850–8
[115] Tan C, Shard C and Ranieri E et al 2015 Mutations of protocadherin 19 in female epilepsy
(PCDH19-FE) lead to allopregnanolone deficiency Hum. Mol. Genet. 24 5250–9
[116] Stevenson R E, Holden K R, Rogers R C and Schwartz C E 2012 Seizures and X-linked
intellectual disability Eur. J. Med. Genet. 55 307–12
[117] Christianson A L, Stevenson R E and van der Meyden C H et al 1999 X linked severe
mental retardation, craniofacial dysmorphology, epilepsy, ophthalmoplegia, and cerebellar
atrophy in a large South African kindred is localised to Xq24-q27 J. Med. Genet. 36 759
[118] Weaving L S, Christodoulou J and Williamson S L et al 2004 Mutations of CDKL5 cause a
severe neurodevelopmental disorder with infantile spasms and mental retardation Am. J.
Hum. Genet. 75 1079–93
[119] Evans J C, Archer H L and Colley J P et al 2005 Early onset seizures and Rett-like features
associated with mutations in CDKL5 Eur. J. Hum. Genet. 13 1113–20
[120] Hicks A A, Bailey M E and Riley B P et al 1994 Further evidence for clustering of human
GABAA receptor subunit genes: localization of the alpha 6-subunit gene (GABRA6) to
distal chromosome 5q by linkage analysis Genomics 20 285–8
[121] Carvill G L, Weckhuysen S and McMahon J M et al 2014 GABRA1 and STXBP1: novel
genetic causes of Dravet syndrome Neurology 82 1245–53
[122] Cossette P, Liu L and Brisebois K et al 2002 Mutation of GABRA1 in an autosomal
dominant form of juvenile myoclonic epilepsy Nat. Genet. 31 184–9
[123] Maljevic S, Krampfl K and Cobilanschi J et al 2006 A mutation in the GABAA receptor
α1-subunit is associated with absence epilepsy Ann. Neurol. 59 983–7
[124] Bradley C A, Taghibiglou C, Collingridge G L and Wang Y T 2008 Mechanisms involved
in the reduction of GABAA receptor alpha1-subunit expression caused by the epilepsy
mutation A322D in the trafficking-competent receptor J. Biol. Chem. 283 22043–50
[125] Ding L, Feng H J, Macdonald R L, Botzolakis E J, Hu N and Gallagher M J 2010
GABAA receptor α1 subunit mutation A322D associated with autosomal dominant
10-27
juvenile myoclonic epilepsy reduces the expression and alters the composition of wild type
GABAA receptors J. Biol. Chem. 285 26390–405
[126] Lachance-Touchette P, Brown P and Meloche C et al 2011 Novel α1 and γ2 GABAA
receptor subunit mutations in families with idiopathic generalized epilepsy Eur. J. Neurosci.
34 237–49
[127] Dibbens L M, Harkin L A and Richards M et al 2009 The role of neuronal GABAA
receptor subunit mutations in idiopathic generalized epilepsies Neurosci. Lett. 453 162–5
[128] Hernandez C C, Gurba K N, Hu N and Macdonald R L 2011 The GABRA6 mutation,
R46W, associated with childhood absence epilepsy, alters 6β22 and 6β2 GABAA receptor
channel gating and expression J. Physiol. 589 5857–78
[129] Wagstaff J, Chaillet J R and Lalande M 1991 The GABAA receptor β3 subunit gene:
characterization of a human cDNA from chromosome 15q11q13 and mapping to a region
of conserved synteny on mouse chromosome 7 Genomics 11 1071–8
[130] Wagstaff J, Knoll J H and Fleming J et al 1991 Localization of the gene encoding the
GABAA receptor β3 subunit to the Angelman/Prader-Willi region of human chromosome
15 Am. J. Hum. Genet. 49 330–7
[131] Epi K C, Epilepsy Phenome/Genome P and Allen A S et al 2013 De novo mutations in
epileptic encephalopathies Nature 501 217–21
[132] Holopainen I E, Metsahonkala E L and Kokkonen H et al 2001 Decreased binding of [11C]
flumazenil in Angelman syndrome patients with GABAA receptor β3 subunit deletions
Ann. Neurol. 49 110–3
[133] Tanaka M, Olsen R W and Medina M T et al 2008 Hyperglycosylation and reduced GABA
currents of mutated GABRB3 polypeptide in remitting childhood absence epilepsy Am. J.
Hum. Genet. 82 1249–61
[134] Urak L, Feucht M, Fathi N, Hornik K and Fuchs K 2006 A GABRB3 promoter haplotype
associated with childhood absence epilepsy impairs transcriptional activity Hum. Mol.
Genet. 15 2533–41
[135] DeLorey T M, Sahbaie P, Hashemi E, Homanics G E and Clark J D 2008 Gabrb3 gene
deficient mice exhibit impaired social and exploratory behaviors, deficits in non-selective
attention and hypoplasia of cerebellar vermal lobules: a potential model of autism spectrum
disorder Behav. Brain Res. 187 207–20
[136] Janve V S, Hernandez C C, Verdier K M, Hu N and Macdonald R L 2016 Epileptic
encephalopathy de novo GABRB mutations impair gamma-aminobutyric acid type A
receptor function Ann. Neurol. 79 806–25
[137] Russek S J and Farb D H 1994 Mapping of the beta 2 subunit gene (GABRB2) to
microdissected human chromosome 5q34-q35 defines a gene cluster for the most abundant
GABAA receptor isoform Genomics 23 528–33
[138] Sinnett D, Wagstaff J, Glatt K, Woolf E, Kirkness E J and Lalande M 1993 High-
resolution mapping of the GABAA receptor subunit β3 and α5 gene cluster on chromosome
15q11-q13, and localization of breakpoints in two Angelman syndrome patients Am. J.
Hum. Genet. 52 1216–29
[139] Baulac S, Huberfeld G and Gourfinkel-An I et al 2001 First genetic evidence of GABAA
receptor dysfunction in epilepsy: a mutation in the γ2-subunit gene Nat. Genet. 28 46–8
[140] Carvill G L, Heavin S B and Yendle S C et al 2013 Targeted resequencing in epileptic
encephalopathies identifies de novo mutations in CHD2 and SYNGAP1 Nat. Genet. 45
825–30
10-28
[141] Kananura C, Haug K and Sander T et al 2002 A splice-site mutation in GABRG2

associated with childhood absence epilepsy and febrile convulsions Arch. Neurol. 59
1137–41
[142] Marini C, Harkin L A, Wallace R H, Mulley J C, Scheffer I E and Berkovic S F 2003
Childhood absence epilepsy and febrile seizures: a family with a GABAA receptor mutation
Brain 126 230–40
[143] Wallace R H, Marini C and Petrou S et al 2001 Mutant GABAA receptor γ2-subunit in
childhood absence epilepsy and febrile seizures Nat. Genet. 28 49–52
[144] Audenaert D, Schwartz E and Claeys K G et al 2006 A novel GABRG2 mutation
associated with febrile seizures Neurology 67 687–90
[145] Kang J Q, Shen W and Macdonald R L 2006 Why does fever trigger febrile seizures?
GABAA receptor γ2 subunit mutations associated with idiopathic generalized epilepsies
have temperature-dependent trafficking deficiencies J. Neurosci. 26 2590–7
[146] Chaumont S, Andre C, Perrais D, Boue-Grabot E, Taly A and Garret M 2013 Agonist-
dependent endocytosis of GABAA receptors revealed by a gamma2(R43Q) epilepsy
mutation J. Biol. Chem. 288 28254–65
[147] Shen D, Hernandez C C and Shen W et al 2017 De novo GABRG2 mutations associated
with epileptic encephalopathies Brain 140 49–67
[148] Tan H O, Reid C A and Single F N et al 2007 Reduced cortical inhibition in a mouse model
of familial childhood absence epilepsy Proc. Natl. Acad. Sci. U.S.A. 104 17536–41
[149] Maguire J L, Stell B M, Rafizadeh M and Mody I 2005 Ovarian cycle-linked changes in
GABAA receptors mediating tonic inhibition alter seizure susceptibility and anxiety Nat.
[150] Spigelman I, Li Z, Banerjee P K, Mihalek R M, Homanics G E and Olsen R W 2002
Behavior and physiology of mice lacking the GABAA-receptor δ subunit Epilepsia 43 3–8
[151] Dibbens L M, Feng H J and Richards M C et al 2004 GABRD encoding a protein for
extra- or peri-synaptic GABAA receptors is a susceptibility locus for generalized epilepsies
Hum. Mol. Genet. 13 1315–9
[152] Feng H J, Kang J Q, Song L, Dibbens L, Mulley J and Macdonald R L 2006 Delta subunit
susceptibility variants E177A and R220H associated with complex epilepsy alter channel
gating and surface expression of alpha4beta2delta GABAA receptors J. Neurosci. 26 1499–
506
[153] Dingledine R 2012 Glutamatergic mechanisms related to epilepsy: Ionotropic receptors
Jasper’s Basic Mechanisms of the Epilepsies ed J L Noebels, M Avoli, M A Rogawski, R W
Olsen and A V Delgado-Escueta (Bethesda, MD: Oxford University Press)
[154] Dingledine R, Borges K, Bowie D and Traynelis S F 1999 The glutamate receptor ion
channels Pharmacol. Rev. 51 7–61
[155] Carvill G L, Regan B M and Yendle S C et al 2013 GRIN2A mutations cause epilepsy-
aphasia spectrum disorders Nat. Genet. 45 1073–6
[156] Landau W M and Kleffner F R 1957 Syndrome of acquired aphasia with convulsive
disorder in children Neurology 7 523–30
[157] Lemke J R, Lal D and Reinthaler E M et al 2013 Mutations in GRIN2A cause idiopathic
focal epilepsy with rolandic spikes Nat. Genet. 45 1067–72
[158] Lesca G, Rudolf G and Bruneau N et al 2013 GRIN2A mutations in acquired epileptic
aphasia and related childhood focal epilepsies and encephalopathies with speech and
language dysfunction Nat. Genet. 45 1061–6
10-29
[159] Scheffer I E, Jones L, Pozzebon M, Howell R A, Saling M M and Berkovic S F 1995

Autosomal dominant rolandic epilepsy and speech dyspraxia: a new syndrome with
anticipation Ann. Neurol. 38 633–42
[160] Kugler S L, Bali B and Lieberman P et al 2008 An autosomal dominant genetically
heterogeneous variant of rolandic epilepsy and speech disorder Epilepsia 49 1086–90
[161] Stefanatos G 2011 Changing perspectives on Landau–Kleffner syndrome Clin.
Neuropsychol. 25 963–88
[162] Platzer K, Yuan H and Schutz H et al 2017 GRIN2B encephalopathy: novel findings on
phenotype, variant clustering, functional consequences and treatment aspects J. Med.
Genet. 54 460–70
[163] Lemke J R, Hendrickx R and Geider K et al 2014 GRIN2B mutations in West syndrome
and intellectual disability with focal epilepsy Ann. Neurol. 75 147–54
[164] Endele S, Rosenberger G and Geider K et al 2010 Mutations in GRIN2A and GRIN2B
encoding regulatory subunits of NMDA receptors cause variable neurodevelopmental
phenotypes Nat. Genet. 42 1021–6
[165] Yuan H, Low C-M, Moody O A, Jenkins A and Traynelis S F 2015 Ionotropic GABA and
glutamate receptor mutations and human neurologic diseases Mol. Pharmacol. 88 203–17
[166] Zara F, Blanchi A and Avanzini G et al 1995 Mapping of genes predisposing to idiopathic
generalized epilepsy Hum. Mol. Genet. 4 1201–7
[167] Letts V A 2005 Stargazer—a mouse to seize! Epilepsy Curr. 5 161–5
[168] Payandeh J, Scheuer T, Zheng N and Catterall W A 2011 The crystal structure of a voltage-
gated sodium channel Nature 475 353–8
[169] Goldin A L 2001 Resurgence of sodium channel research Annu. Rev. Physiol. 63 871–94
[170] Isom L L 2002 The role of sodium channels in cell adhesion Front. Biosci. 7 12–23
[171] Noda M, Ikeda T and Kayano T et al 1986 Existence of distinct sodium channel messenger
RNAs in rat brain Nature 320 188–92
[172] Noda M, Ikeda T and Suzuki H et al 1986 Expression of functional sodium channels from
cloned cDNA Nature 322 826–8
[173] Noda M, Shimizu S and Tanabe T et al 1984 Primary structure of Electrophorus electricus
sodium channel deduced from cDNA sequence Nature 312 121–7
[174] Hartshorne R P and Catterall W A 1981 Purification of the saxitoxin receptor of the sodium
channel from rat brain Proc. Natl. Acad. Sci. U.S.A. 78 4620–4
[175] Hartshorne R P and Catterall W A 1984 The sodium channel from rat brain. Purification
and subunit composition J. Biol. Chem. 259 1667–75
[176] Hartshorne R P, Messner D J, Coppersmith J C and Catterall W A 1982 The saxitoxin
receptor of the sodium channel from rat brain. Evidence for two nonidentical beta subunits
J. Biol. Chem. 257 13888–91
[177] Isom L L, De Jongh K S and Patton D E et al 1992 Primary structure and functional
expression of the β1 subunit of the rat brain sodium channel Science 256 839–42
[178] Isom L L, Ragsdale D S and De Jongh K S et al 1995 Structure and function of the beta 2
subunit of brain sodium channels, a transmembrane glycoprotein with a CAM motif Cell 83
433–42
[179] Cantrell A R and Catterall W A 2001 Neuromodulation of Na+ channels: an unexpected
form of cellular plasticity Nat. Rev. Neurosci. 2 397–407
10-30
[180] Carr D B, Day M and Cantrell A R et al 2003 Transmitter modulation of slow, activity-
dependent alterations in sodium channel availability endows neurons with a novel form of
cellular plasticity Neuron 39 793–806
[181] Chen Y, Yu F H, Surmeier D J, Scheuer T and Catterall W A 2006 Neuromodulation of
Na+ channel slow inactivation via cAMP-dependent protein kinase and protein kinase C
Neuron 49 409–20
[182] Catterall W A 2010 Ion channel voltage sensors: structure, function, and pathophysiology
Neuron 67 915–28
[183] Lossin C 2009 A catalog of SCN1A variants Brain Dev. 31 114–30
[184] Mulley J C, Scheffer I E, Petrou S, Dibbens L M, Berkovic S F and Harkin L A 2005
SCN1A mutations and epilepsy Hum. Mutat. 25 535–42
[185] Vadlamudi L, Dibbens L M and Lawrence K M et al 2010 Timing of de novo mutagenesis--
a twin study of sodium-channel mutations N. Engl. J. Med. 363 1335–40
[186] Carranza Rojo D, Hamiwka L and McMahon J M et al 2011 De novo SCN1A mutations
in migrating partial seizures of infancy Neurology 77 380–3
[187] Doose H, Lunau H, Castiglione E and Waltz S 1998 Severe idiopathic generalized epilepsy
of infancy with generalized tonic-clonic seizures Neuropediatrics 29 229–38
[188] Jansen F E, Sadleir L G and Harkin L A et al 2006 Severe myoclonic epilepsy of infancy
(Dravet syndrome): recognition and diagnosis in adults Neurology 67 2224–6
[189] Mulley J C, Nelson P and Guerrero S et al 2006 A new molecular mechanism for severe
myoclonic epilepsy of infancy: exonic deletions in SCN1A Neurology 67 1094–5
[190] Riva D, Vago C, Pantaleoni C, Bulgheroni S, Mantegazza M and Franceschetti S 2009
Progressive neurocognitive decline in two children with Dravet syndrome, de novo SCN1A
truncations and different epileptic phenotypes Am. J. Med. Genet. A 149A 2339–45
[191] Sugawara T, Mazaki-Miyazaki E and Fukushima K et al 2002 Frequent mutations of
SCN1A in severe myoclonic epilepsy in infancy Neurology 58 1122–4
[192] Suls A, Velizarova R and Yordanova I et al 2010 Four generations of epilepsy caused by an
inherited microdeletion of the SCN1A gene Neurology 75 72–6
[193] Dravet C and Oguni H 2013 Dravet syndrome (severe myoclonic epilepsy in infancy)
Handb. Clin. Neurol. 111 627–33
[194] Dravet C 2011 Dravet syndrome history Dev. Med. Child Neurol. 53 1–6
[195] Dravet C 2011 The core Dravet syndrome phenotype Epilepsi 52 3–9
[196] Dravet C, Bureau M, Dalla Bernardina B and Guerrini R 2011 Severe myoclonic epilepsy
in infancy (Dravet syndrome) 30 years later Epilepsia 52 1–2
[197] Freilich E R, Jones J M and Gaillard W D et al 2011 Novel SCN1A mutation in a proband
with malignant migrating partial seizures of infancy Arch. Neurol. 68 665–71
[198] Fujiwara T, Sugawara T and Mazaki-Miyazaki E et al 2003 Mutations of sodium channel α
subunit type 1 (SCN1A) in intractable childhood epilepsies with frequent generalized tonic-
clonic seizures Brain 126 531–46
[199] Heinzen E L, Yoon W and Tate S K et al 2007 Nova2 interacts with a cis-acting
polymorphism to influence the proportions of drug-responsive splice variants of SCN1A
Am. J. Hum. Genet. 80 876–83
[200] Lossin C, Wang D W, Rhodes T H, Vanoye C G and George A L Jr 2002 Molecular basis
of an inherited epilepsy Neuron 34 877–84
10-31
[201] Tate S K, Depondt C and Sisodiya S M et al 2005 Genetic predictors of the maximum doses
patients receive during clinical use of the anti-epileptic drugs carbamazepine and phenytoin
Proc. Natl. Acad. Sci. U.S.A. 102 5507–12
[202] Baulac S, Gourfinkel-An I and Picard F et al 1999 A second locus for familial generalized
epilepsy with febrile seizures plus maps to chromosome 2q21-q33 Am. J. Hum. Genet. 65
1078–85
[203] Claes L, Ceulemans B and Audenaert D et al 2003 De novo SCN1A mutations are a major
cause of severe myoclonic epilepsy of infancy Hum. Mutat. 21 615–21
[204] Claes L, Del-Favero J, Ceulemans B, Lagae L, Van Broeckhoven C and De Jonghe P 2001
De novo mutations in the sodium-channel gene SCN1A cause severe myoclonic epilepsy of
infancy Am. J. Hum. Genet. 68 1327–32
[205] Depienne C, Trouillard O and Saint-Martin C et al 2009 Spectrum of SCN1A gene
mutations associated with Dravet syndrome: analysis of 333 patients J. Med. Genet. 46
183–91
[206] Escayg A, Heils A, MacDonald B T, Haug K, Sander T and Meisler M H 2001 A novel
SCN1A mutation associated with generalized epilepsy with febrile seizures plus--and
prevalence of variants in patients with epilepsy Am. J. Hum. Genet. 68 866–73
[207] Escayg A, MacDonald B T and Meisler M H et al 2000 Mutations of SCN1A, encoding a
neuronal sodium channel, in two families with GEFS+2 Nat. Genet. 24 343–5
[208] Mantegazza M, Gambardella A and Rusconi R et al 2005 Identification of an Nav1.1
sodium channel (SCN1A) loss-of-function mutation associated with familial simple febrile
seizures Proc. Natl. Acad. Sci. U.S.A. 102 18177–82
[209] Moulard B, Guipponi M, Chaigne D, Mouthon D, Buresi C and Malafosse A 1999
Identification of a new locus for generalized epilepsy with febrile seizures plus (GEFS+) on
chromosome 2q24-q33 Am. J. Hum. Genet. 65 1396–400
[210] Ohmori I, Ouchida M, Ohtsuka Y, Oka E and Shimizu K 2002 Significant correlation of
the SCN1A mutations and severe myoclonic epilepsy in infancy Biochem. Biophys. Res.
Commun. 295 17–23
[211] Orrico A, Galli L and Grosso S et al 2009 Mutational analysis of the SCN1A, SCN1B and
GABRG2 genes in 150 Italian patients with idiopathic childhood epilepsies Clin. Genet. 75
579–81
[212] Wallace R H, Scheffer I E and Barnett S et al 2001 Neuronal sodium-channel alpha1-
subunit mutations in generalized epilepsy with febrile seizures plus Am. J. Hum. Genet. 68
859–65
[213] Zucca C, Redaelli F and Epifanio R et al 2008 Cryptogenic epileptic syndromes related to
SCN1A: twelve novel mutations identified Arch. Neurol. 65 489–94
[214] Singh N A, Pappas C and Dahle E J et al 2009 A role of SCN9A in human epilepsies, as a
cause of febrile seizures and as a potential modifier of Dravet syndrome PLoS Genet. 5
e1000649
[215] Kanai K, Hirose S and Oguni H et al 2004 Effect of localization of missense mutations in
SCN1A on epilepsy phenotype severity Neurology 63 329–34
[216] Yu F H, Mantegazza M and Westenbroek R E et al 2006 Reduced sodium current in
GABAergic interneurons in a mouse model of severe myoclonic epilepsy in infancy Nat.
10-32
[217] Martin M S, Tang B, Papale L A, Yu F H, Catterall W A and Escayg A 2007 The voltage-
gated sodium channel Scn8a is a genetic modifier of severe myoclonic epilepsy of infancy
Hum. Mol. Genet. 16 2892–9
[218] Ogiwara I, Miyamoto H and Morita N et al 2007 Nav1.1 localizes to axons of
parvalbumin-positive inhibitory interneurons: a circuit basis for epileptic seizures in mice
carrying an Scn1a gene mutation J. Neurosci. 27 5903–14
[219] Ahmed C M, Ware D H and Lee S C et al 1992 Primary structure, chromosomal
localization, and functional expression of a voltage-gated sodium channel from human
brain Proc. Natl. Acad. Sci. U.S.A. 89 8220–4
[220] Cooperman S S, Grubman S A, Barchi R L, Goodman R H and Mandel G 1987
Modulation of sodium-channel mRNA levels in rat skeletal muscle Proc. Natl. Acad. Sci.
U.S.A. 84 8721–5
[221] Garrido J J, Giraud P and Carlier E et al 2003 A targeting motif involved in sodium
channel clustering at the axonal initial segment Science 300 2091–4
[222] Kasai N, Fukushima K and Ueki Y et al 2001 Genomic structures of SCN2A and SCN3A -
candidate genes for deafness at the DFNA16 locus Gene 264 113–22
[223] Malo D, Schurr E, Dorfman J, Canfield V, Levenson R and Gros P 1991 Three brain
sodium channel alpha-subunit genes are clustered on the proximal segment of mouse
chromosome 2 Genomics 10 666–72
[224] Berkovic S F, Heron S E and Giordano L et al 2004 Benign familial neonatal-infantile
seizures: characterization of a new sodium channelopathy Ann. Neurol. 55 550–7
[225] Heron S E, Crossland K M and Andermann E et al 2002 Sodium-channel defects in benign
familial neonatal-infantile seizures Lancet 360 851–2
[226] Kaplan R E and Lacey D J 1983 Benign familial neonatal-infantile seizures Am. J. Med.
Genet. 16 595–9
[227] Malacarne M, Gennaro E and Madia F et al 2001 Benign familial infantile convulsions:
mapping of a novel locus on chromosome 2q24 and evidence for genetic heterogeneity Am.
J. Hum. Genet. 68 1521–6
[228] Shevell M I, Sinclair D B and Metrakos K 1986 Benign familial neonatal seizures: clinical
and electroencephalographic characteristics Pediatr. Neurol. 2 272–5
[229] Sugawara T, Mazaki-Miyazaki E and Ito M et al 2001 Nav1.1 mutations cause febrile
seizures associated with afebrile partial seizures Neurology 57 703–5
[230] Sugawara T, Tsurubuchi Y and Agarwala K L et al 2001 A missense mutation of the Na+
channel alpha II subunit gene Na(v)1.2 in a patient with febrile and afebrile seizures causes
channel dysfunction Proc. Natl. Acad. Sci. U.S.A. 98 6384–9
[231] Liao Y, Anttonen A K and Liukkonen E et al 2010 SCN2A mutation associated with
neonatal epilepsy, late-onset episodic ataxia, myoclonus, and pain Neurology 75 1454–8
[232] Hackenberg A, Baumer A and Sticht H et al 2014 Infantile epileptic encephalopathy,
transient choreoathetotic movements, and hypersomnia due to a De Novo missense
mutation in the SCN2A gene Neuropediatrics 45 261–4
[233] Ogiwara I, Ito K and Sawaishi Y et al 2009 De novo mutations of voltage-gated sodium
channel alphaII gene SCN2A in intractable epilepsies Neurology 73 1046–53
[234] Touma M, Joshi M and Connolly M C et al 2013 Whole genome sequencing identifies
SCN2A mutation in monozygotic twins with Ohtahara syndrome and unique neuro-
pathologic findings Epilepsia 54 e81–5
10-33
[235] Chen Y H, Dale T J, Romanos M A, Whitaker W R, Xie X M and Clare J J 2000 Cloning,
distribution and functional analysis of the type III sodium channel from human brain Eur.
J. Neurosci. 12 4281–9
[236] Vanoye C G, Gurnett C A, Holland K D, George A L Jr and Kearney J A 2014 Novel
SCN3A variants associated with focal epilepsy in children Neurobiol. Dis. 62 313–22
[237] Lamar T, Vanoye C G and Calhoun J et al 2017 SCN3A deficiency associated with
increased seizure susceptibility Neurobiol. Dis. 102 38–48
[238] Zaman T, Helbig I and Bozovic I B et al 2018 Mutations in SCN3A cause early infantile
epileptic encephalopathy Ann. Neurol. 83 703–17
[239] Lorincz A and Nusser Z 2010 Molecular identity of dendritic voltage-gated sodium
channels Science 328 906–9
[240] Burgess D L, Kohrman D C and Galt J et al 1995 Mutation of a new sodium channel gene,
Scn8a, in the mouse mutant ‘motor endplate disease’ Nat. Genet. 10 461–5
[241] De Repentigny Y, Cote P D and Pool M et al 2001 Pathological and genetic analysis of the
degenerating muscle (dmu) mouse: a new allele of Scn8a Hum. Mol. Genet. 10 1819–27
[242] Kohrman D C, Harris J B and Meisler M H 1996 Mutation detection in the med and medJ
alleles of the sodium channel Scn8a. Unusual splicing due to a minor class AT-AC intron
J. Biol. Chem. 271 17576–81
[243] Patino G A, Claes L R and Lopez-Santiago L F et al 2009 A functional null mutation of
SCN1B in a patient with Dravet syndrome J. Neurosci. 29 10764–78
[244] Tammaro P, Conti F and Moran O 2002 Modulation of sodium current in mammalian cells
by an epilepsy-correlated β1-subunit mutation Biochem. Biophys. Res. Commun. 291
1095–101
[245] Audenaert D, Claes L, Ceulemans B, Lofgren A, Van Broeckhoven C and De Jonghe P
2003 A deletion in SCN1B is associated with febrile seizures and early-onset absence
epilepsy Neurology 61 854–6
[246] Ogiwara I, Nakayama T and Yamagata T et al 2012 A homozygous mutation of voltage-
gated sodium channel beta(I) gene SCN1B in a patient with Dravet syndrome Epilepsia 53
e200–3
[247] Wallace R H, Scheffer I E and Parasivam G et al 2002 Generalized epilepsy with febrile
seizures plus: mutation of the sodium channel subunit SCN1B Neurology 58 1426–9
[248] Wallace R H, Wang D W and Singh R et al 1998 Febrile seizures and generalized epilepsy
associated with a mutation in the Na+-channel β1 subunit gene SCN1B Nat. Genet. 19
366–70
[249] Gutman G A, Chandy K G and Grissmer S et al 2005 International Union of
Pharmacology. LIII. Nomenclature and molecular relationships of voltage-gated potassium
channels Pharmacol. Rev. 57 473–508
[250] Murrell-Lagnado R D and Aldrich R W 1993 Energetics of Shaker K channels block by
inactivation peptides J. Gen. Physiol. 102 977–1003
[251] Pongs O, Leicher T and Berger M et al 1999 Functional and molecular aspects of voltage-
gated K+ channel β subunits Ann. N. Y. Acad. Sci. 30 344–55
[252] Yellen G 2002 The voltage-gated potassium channels and their relatives Nature 419 35–42
[253] Fister P, Soltirovska-Salamon A, Debeljak M and Paro-Panjan D 2013 Benign familial
neonatal convulsions caused by mutation in KCNQ3, exon 6: a European case Eur. J.
Paediatr. Neurol. 17 308–10
10-34
[254] Hirose S, Zenri F and Akiyoshi H et al 2000 A novel mutation of KCNQ3 (c.925T-->C) in
a Japanese family with benign familial neonatal convulsions Ann. Neurol. 47 822–6
[255] Li H, Li N and Shen L et al 2008 A novel mutation of KCNQ3 gene in a Chinese family
with benign familial neonatal convulsions Epilepsy Res. 79 1–5
[256] Ryan S G, Wiznitzer M, Hollman C, Torres M C, Szekeresova M and Schneider S 1991
Benign familial neonatal convulsions: evidence for clinical and genetic heterogeneity Ann.
Neurol. 29 469–73
[257] Heron S E, Grinton B E and Kivity S et al 2012 PRRT2 mutations cause benign familial
infantile epilepsy and infantile convulsions with choreoathetosis syndrome Am. J. Hum.
Genet. 90 152–60
[258] Zara F, Specchio N and Striano P et al 2013 Genetic testing in benign familial epilepsies of
the first year of life: clinical and diagnostic significance Epilepsia 54 425–36
[259] Striano P, Bordo L and Lispi M L et al 2006 A novel SCN2A mutation in family with
benign familial infantile seizures Epilepsia 47 218–20
[260] Allen N M, Mannion M and Conroy J et al 2014 The variable phenotypes of KCNQ-
related epilepsy Epilepsia 55 e99–e105
[261] Bosch D G, Boonstra F N and de Leeuw N et al 2016 Novel genetic causes for cerebral
visual impairment Eur. J. Hum. Genet. 24 660–5
[262] Deciphering Developmental Disorders Study 2017 Prevalence and architecture of de novo
mutations in developmental disorders Nature 542 433–8
[263] Rauch A, Wieczorek D and Graf E et al 2012 Range of genetic mutations associated with
severe non-syndromic sporadic intellectual disability: an exome sequencing study Lancet
380 1674–82
[264] Coetzee W A, Amarillo Y and Chiu J et al 1999 Molecular diversity of K+ channels Ann.
N. Y. Acad. Sci. 868 233–85
[265] Gutman G A, Chandy K G and Adelman J P et al 2003 International Union of
Pharmacology. XLI. Compendium of voltage-gated ion channels: potassium channels
Pharmacol. Rev. 55 583–6
[266] Imbrici P, Cusimano A, D’Adamo M C, De Curtis A and Pessia M 2003 Functional
characterization of an episodic ataxia type-1 mutation occurring in the S1 segment of
hKv1.1 channels Pflugers Arch. 446 373–9
[267] Rogers A, Golumbek P and Cellini E et al 2018 De novo KCNA1 variants in the PVP motif
cause infantile epileptic encephalopathy and cognitive impairment similar to recurrent
KCNA2 variants Am. J. Med. Genet. A 176 1748–52
[268] Zuberi S M, Eunson L H and Spauschus A et al 1999 A novel mutation in the human
voltage-gated potassium channel gene (Kv1.1) associates with episodic ataxia type 1 and
sometimes with partial epilepsy Brain 122 817–25
[269] Pena S D and Coimbra R L 2015 Ataxia and myoclonic epilepsy due to a heterozygous new
mutation in KCNA2: proposal for a new channelopathy Clin. Genet. 87 e1–3
[270] Syrbe S, Hedrich U B S and Riesch E et al 2015 De novo loss- or gain-of-function
mutations in KCNA2 cause epileptic encephalopathy Nat. Genet. 47 393–9
[271] Torkamani A, Bersell K and Jorge B S et al 2014 De novo KCNB1 mutations in epileptic
encephalopathy Ann. Neurol. 76 529–40
[272] Kousi M, Anttila V and Schulz A et al 2012 Novel mutations consolidate KCTD7 as a
progressive myoclonus epilepsy gene J. Med. Genet. 49 391–9
10-35
[273] Staropoli J F, Karaa A and Lim E T et al 2012 A homozygous mutation in KCTD7 links
neuronal ceroid lipofuscinosis to the ubiquitin-proteasome system Am. J. Hum. Genet. 91
202–8
[274] Van Bogaert P, Azizieh R and Desir J et al 2007 Mutation of a potassium channel-related
gene in progressive myoclonic epilepsy Ann. Neurol. 61 579–86
[275] Singh B, Ogiwara I and Kaneda M et al 2006 A Kv4.2 truncation mutation in a patient
with temporal lobe epilepsy Neurobiol. Dis. 24 245–53
[276] Marini C, Scheffer I E and Crossland K M et al 2004 Genetic architecture of idiopathic
generalized epilepsy: clinical genetic analysis of 55 multiplex families Epilepsia 45 467–78
[277] Dunlap K, Luebke J I and Turner T J 1995 Exocytotic Ca2+ channels in mammalian
central neurons Trends Neurosci. 18 89–98
[278] Lehmann-Horn F and Jurkat-Rott K 1999 Voltage-gated ion channels and hereditary
disease Physiol. Rev. 79 1317–72
[279] Simms B A and Zamponi G W 2014 Neuronal voltage-gated calcium channels: structure,
function, and dysfunction Neuron 82 24–45
[280] Zamponi G W 2017 A crash course in calcium channels ACS Chem. Neurosci. 8 2583–5
[281] Zamponi G W, Striessnig J, Koschak A and Dolphin A C 2015 The physiology, pathology,
and pharmacology of voltage-gated calcium channels and their future therapeutic potential
Pharmacol. Rev. 67 821–70
[282] Kordasiewicz H B, Thompson R M, Clark H B and Gomez C M 2006 C-termini of P/Q-
type Ca2+ channel alpha1A subunits translocate to nuclei and promote polyglutamine-
mediated toxicity Hum. Mol. Genet. 15 1587–99
[283] Cribbs L L, Lee J H and Yang J et al 1998 Cloning and characterization of alpha1H from
human heart, a member of the T-type Ca2+ channel gene family Circ. Res. 83 103–9
[284] Khosravani H, Altier C and Simms B et al 2004 Gating effects of mutations in the Cav3.2
T-type calcium channel associated with childhood absence epilepsy J. Biol. Chem. 279
9681–4
[285] Wolfe J T, Wang H, Howard J, Garrison J C and Barrett P Q 2003 T-type calcium channel
regulation by specific G-protein betagamma subunits Nature 424 209–13
[286] Williams M E, Washburn M S and Hans M et al 1999 Structure and functional
characterization of a novel human low-voltage activated calcium channel J. Neurochem.
72 791–9
[287] Chen Y, Lu J and Pan H et al 2003 Association between genetic variation of CACNA1H
and childhood absence epilepsy Ann. Neurol. 54 239–43
[288] Heron S E, Khosravani H and Varela D et al 2007 Extended spectrum of idiopathic
generalized epilepsies associated with CACNA1H functional variants Ann. Neurol. 62
560–8
[289] Heron S E, Phillips H A and Mulley J C et al 2004 Genetic variation of CACNA1H in
idiopathic generalized epilepsy Ann. Neurol. 55 595–6
[290] Chen Y H, Li M H and Zhang Y et al 2004 Structural basis of the α1-β subunit interaction
of voltage-gated Ca2+ channels Nature 429 675–80
[291] Taviaux S, Williams M E, Harpold M M, Nargeot J and Lory P 1997 Assignment of
human genes for β2 and β4 subunits of voltage-dependent Ca2+ channels to chromosomes
10p12 and 2q22-q23 Hum. Genet. 100 151–4
10-36
[292] Escayg A, De Waard M and Lee D D et al 2000 Coding and noncoding variation of the
human calcium-channel β4-subunit gene CACNB4 in patients with idiopathic generalized
epilepsy and episodic ataxia Am. J. Hum. Genet. 66 1531–9
[293] Bertrand D, Galzi J L, Devillers-Thiery A, Bertrand S and Changeux J P 1993 Stratification
of the channel domain in neurotransmitter receptors Curr. Opin. Cell Biol. 5 688–93
[294] Dani J A 1989 Open channel structure and ion binding sites of the nicotinic acetylcholine
receptor channel J. Neurosci. 9 884–92
[295] Karlin A 2002 Emerging structure of the nicotinic acetylcholine receptors Nat. Rev.
[296] Unwin N and Fujiyoshi Y 2012 Gating movement of acetylcholine receptor caught by
plunge-freezing J. Mol. Biol. 422 617–34
[297] Albuquerque E X, Pereira E F, Alkondon M and Rogers S W 2009 Mammalian nicotinic
acetylcholine receptors: from structure to function Physiol. Rev. 89 73–120
[298] Dani J A and Balfour D J 2011 Historical and current perspective on tobacco use and
nicotine addiction Trends Neurosci. 34 383–92
[299] Fasoli F and Gotti C 2015 Structure of neuronal nicotinic receptors Curr. Top Behav.
Neurosci. 23 1–17
[300] Lewis A S and Picciotto M 2013 High-affinity nicotinic acetylcholine receptor expression
and trafficking abnormalities in psychiatric illness Psychopharmacology 229 477–85
[301] McGehee D S and Role L W 1995 Physiological diversity of nicotinic acetylcholine
receptors expressed by vertebrate neurons Annu. Rev. Physiol. 57 521–46
[302] McKay B E, Placzek A N and Dani J A 2007 Regulation of synaptic transmission and
plasticity by neuronal nicotinic acetylcholine receptors Biochem. Pharmacol. 74 1120–33
[303] Papke R L 2014 Merging old and new perspectives on nicotinic acetylcholine receptors
Biochem. Pharmacol. 89 1–11
[304] Dani J A and Bertrand D 2007 Nicotinic acetylcholine receptors and nicotinic cholinergic
mechanisms of the central nervous system Annu. Rev. Pharmacol. Toxicol. 47 699–729
[305] Unwin N 2013 Nicotinic acetylcholine receptor and the structural basis of neuromuscular
transmission: insights from Torpedo postsynaptic membranes Q. Rev. Biophys. 46 283–322
[306] Hilf R J and Dutzler R 2008 X-ray structure of a prokaryotic pentameric ligand-gated ion
channel Nature 452 375–9
[307] Lee W Y and Sine S M 2005 Principal pathway coupling agonist binding to channel gating
in nicotinic receptors Nature 438 243–7
[308] Sine S M and Engel A G 2006 Recent advances in Cys-loop receptor structure and function
Nature 440 448–55
[309] Dani J A, Radcliffe K A and Pidoplichko V I 2000 Variations in desensitization of nicotinic
acetylcholine receptors from hippocampus and midbrain dopamine areas Eur. J.
Pharmacol. 393 31–8
[310] Giniatullin R, Nistri A and Yakel J L 2005 Desensitization of nicotinic ACh receptors:
shaping cholinergic signaling Trends Neurosci. 28 371–8
[311] Quick M W and Lester R A 2002 Desensitization of neuronal nicotinic receptors
J. Neurobiol. 53 457–78
[312] Wooltorton J R, Pidoplichko V I, Broide R S and Dani J A 2003 Differential desensitiza-
tion and distribution of nicotinic acetylcholine receptor subtypes in midbrain dopamine
areas J. Neurosci. 23 3176–85
10-37
[313] Aridon P, Marini C and Di Resta C et al 2006 Increased sensitivity of the neuronal nicotinic
receptor α2 subunit causes familial epilepsy with nocturnal wandering and ictal fear Am. J.
Hum. Genet. 79 342–50
[314] Trivisano M, Terracciano A and Milano T et al 2015 Mutation of CHRNA2 in a family
with benign familial infantile seizures: Potential role of nicotinic acetylcholine receptor in
various phenotypes of epilepsy Epilepsia 56 e53–7
[315] Cho Y W, Motamedi G K and Laufenberg I et al 2003 A Korean kindred with autosomal
dominant nocturnal frontal lobe epilepsy and mental retardation Arch. Neurol. 60 1625–32
[316] Fedi M, Berkovic S F and Scheffer I E et al 2008 Reduced striatal D1 receptor binding in
autosomal dominant nocturnal frontal lobe epilepsy Neurology 71 795–8
[317] Hirose S, Iwata H and Akiyoshi H et al 1999 A novel mutation of CHRNA4 responsible
for autosomal dominant nocturnal frontal lobe epilepsy Neurology 53 1749–53
[318] Phillips H A, Scheffer I E, Berkovic S F, Hollway G E, Sutherland G R and Mulley J C
1995 Localization of a gene for autosomal dominant nocturnal frontal lobe epilepsy to
chromosome 20q 13.2 Nat. Genet. 10 117–8
[319] Saenz A, Galan J and Caloustian C et al 1999 Autosomal dominant nocturnal frontal lobe
epilepsy in a Spanish family with a Ser252Phe mutation in the CHRNA4 gene Arch. Neurol.
56 1004–9
[320] De Fusco M, Becchetti A and Patrignani A et al 2000 The nicotinic receptor β2 subunit is
mutant in nocturnal frontal lobe epilepsy Nat. Genet. 26 275–6
[321] Phillips H A, Favre I and Kirkpatrick M et al 2001 CHRNB2 is the second acetylcholine
receptor subunit associated with autosomal dominant nocturnal frontal lobe epilepsy Am.
J. Hum. Genet. 68 225–31
[322] Rempel N, Heyers S, Engels H, Sleegers E and Steinlein O K 1998 The structures of the
human neuronal nicotinic acetylcholine receptor β2- and α3-subunit genes (CHRNB2 and
CHRNA3) Hum. Genet. 103 645–53
[323] Conti V, Aracri P and Chiti L et al 2015 Nocturnal frontal lobe epilepsy with paroxysmal
arousals due to CHRNA2 loss of function Neurology 84 1520–8
[324] Poduri A 2017 When should genetic testing be performed in epilepsy patients? Epilepsy
Curr. 17 16–22
10-38
IOP Publishing

Chapter 11
Neuroimaging in Parkinson’s disease
Pearl S Rodrigues, Pravin P Kale and Lokesh K Bhatt
Parkinson’s is a neurodegenerative disease causing motor and non-motor disorder

which hampers normal lifestyle. As it is possible to give only a symptomatic
treatment, early detection for Parkinson’s is necessary. Various diagnostic methods
have been developed to detect PD. Biomarkers help in diagnosis of a disease.
Neuroimaging techniques provide a more prominent data and help in giving new
insights about the neuronal function. Different methods like MRI, molecular
imaging, transcranial sonography, positron emission tomography, single photon
emission computed tomography, diffusion tensor magnetic resonance imaging and
proton magnetic resonance spectroscopy are a few methods of diagnosing
Parkinson’s disease, which are discussed below.
11.1 Introduction
Parkinson’s disease is the second most common neurodegenerative disorder which
includes multiple motor and non-motor neural circuits. The two main causes of the
disease involve the loss of dopamine neurons in the substantia nigra pars compacta
and the accumulation of Lewy bodies. Lewy body comprises of misfolded aggre-
gations of the protein α-synuclein. Both these causes eventually lead to degradation
of neurons [1–4].
Initially it was considered that this disease occurs due to aging, but several studies
have shown that certain risk factors can also lead to the progression of the disease.
Environmental factors include pesticides, herbicides and heavy metals; dietary
factors include fats and fatty acids and genetic factors include mutation in certain
genes like α-syn, parkin, Dj-1, LRRK2, PINK1, etc. Apart from these factors
traumatic brain injury is also one of the leading causes for PD [4, 5].
In industrialized countries, it is estimated that the prevalence of PD is about 1% of
the entire population in people with age group of above 60 years. Studies have
reported that there is a higher risk of PD in males than females. The estrogens in
females are said to have a neuroprotective effect which explains why males have a

higher risk than females. Studies have also suggested that the ratio of mortality was
found to be 1.5 and 2.7, respectively [4, 6].
11.2 Neuroimaging biomarkers in Parkinson’s disease

A biomarker is rightly defined by the National Institute of Health (NIH) as ‘a
characteristic that is objectively measured and evaluated as an indicator of normal
biological processes, pathogenic processes or pharmacological responses to a
therapeutic intervention’. In simple terms, a biomarker is used as a tool to help
the scientists and researchers for the diagnosis and identification of disease
progression [7].
There are diverse biomarkers involved in PD (figure 11.1). They are classified as
clinical, genetic, imaging and biochemical. One of the drawbacks of Parkinson’s
disease is that it is diagnosed far too late. Hence, we need to validate certain
biomarkers which would help in the diagnosis of the disease at an early stage [8, 9].
Development of neuroimaging techniques which can provide higher resolution
data has been the main objective. It helps in supporting the clinical diagnosis of PD.
Modern neuroimaging techniques give us new insights to understand neuronal
function and PD pathophysiology and the treatment complications [7, 10].
11.3 Molecular imaging of Parkinson’s disease

With molecular imaging it is possible to detect the changes or events happening in
the body at molecular or cellular levels. In PD, it helps to measure the abnormal
levels of neurotransmitters in the neurons. This process involves an imaging device,
an imaging agent and a probe. It is a painless, non-invasive and safe technique for
diagnosis of the disease [9, 11].
Figure 11.1. Detailed classification of biomarkers in Parkinson’s disease which includes clinical, biochemical,
genetic, imaging.
11-2
11.3.1 Dopamine
Dopamine imbalance is the basic cause in the pathogenesis of PD (figure 11.2). To
detect the imbalance in this neurotransmitter we will need to perform DAT imaging.
DAT imaging refers to the studies related to the level of membrane dopamine
transporter (DAT) [11]. The radiotracer used for DAT imaging is 2-beta-[11C]
carbomethoxy-3-beta-4-flurophenyltropane (CFT) which reflects DAT.
C-dihydrotetrabenazine (DTBZ) a radio tracer has shown to be useful as a
biomarker for dopamine neuron integrity. It works by reflecting vesicular mono-
amine transporter type 2 (VMAT2). It was used in the assessment of striatal
dopamine nerve terminal in patients with mild cognitive impairment. The results
proved to be useful in detecting the depletion of striatal dopamine [7, 10].
SPECT imaging is done with the help of radio ligand 123I-FP-CIT. This radio
ligand can detect DAT by reversibly binding to the striatal presynaptic DAT.
Clinical studies carried out in different age groups for this radio ligand showed that
striatal structures were involved in contributing to neurodegeneration in the older
patients [10, 12].
11.3.2 Serotonin
The PET ligand for serotonergic transporter (SERT), is [11C]-DASB, helps to assess
in vivo serotonergic terminals [13]. In PD patients, buspirone a serotonin receptor
type 1A agonist has been used to improve the condition of levodopa induced
dyskinesia’s (LIDs). The increase in the SERT to DAT ratio has been found to be
associated with the progression of PD in consequence of which the patients
experience LIDs [11].
In early stages of PD, 123I-FP-CIT SPECT has been used to determine the clinical
correlates of brainstem raphe serotonergic dysfunction. Within 2 years of PD, it was
found that serotonergic raphe nuclei complex was already dysfunctional. The
binding values of raphe region were found to be decreased which reflected the
reduced serotonin transporter availability. Studies have suggested that there appears
be a link between raphe serotonergic dysfunction and severe resting tremors [11, 14].
Figure 11.2. Imbalance of neurotransmitters in Parkinson’s disease. In Parkinson’s disease levels of DA

decrease, which leads to imbalance between DA and ACh. DA: dopamine; ACh: acetylcholine.
11-3
11.3.3 Cholinergic dysfunction

The pathogenesis of PD itself involves an imbalance in dopamine and acetylcholine
levels. Due to cholinergic dysfunction there is an excessive increase in the levels of
acetylcholine. This leads to Parkinson like condition [15]. Recent studies suggested
the role of cholinergic switch between large scale functions of the brain was found to
be was found to be involved in PD pathogenesis [10].
Cholinergic system was also found to be involved in causing gait difficulties,
postural instability, freezing of gait, slow gait speed conditions in PD. Hence,
researchers are trying to find new ways to assess the diagnosis of PD in the GIT by
using 11C-donepezil marker for PET scanning. SPECT study for detection of
cholinergic dysfunction is carried out by using [123I]-QNB 11C-PMP as marker for
the muscarinic cholinergic receptor [10, 16, 17].
A decrease in the levels of acetylcholine esterase (AChE) occurred in PD patients.
This was detected with the help of PET study using 11C-methylpiperidine-4-yl
acetate (MP4A) and 11C-methyl-4-piperinyl propionate (PMP) [16].
11.3.4 Noradrenergic function

Impairment in the noradrenergic function can also play an important role in the
pathogenesis of PD. 11C-MeNER is a radio ligand for noradrenaline transporter.
The binding of this radio ligand to the noradrenergic receptors was found to be
reduced in PD patients. This correlated with their cognitive performance, REM
sleep behavioral disorders (RBDs), orthostatic hypotension [10].
11
C-RT123 binds to dopamine and noradrenaline membrane transporters and
can be detected by PET scan. As PD patients have low dopaminergic innervation it
is likely that 11C-RT123 binds to the noradrenergic transporters. This signals to the
role of noradrenaline dysfunction in causing PD [16, 18].
11.4 Imaging midbrain structural changes in PD

11.4.1 Magnetic resonance imaging techniques
There are different types of MR imaging techniques available such as fMRI and
dMRI, which help in the studies of functional disorders, movement disorders and
the neuronal pathway disorder. Elevated water levels were found in the substantia
nigra in PD patients. These findings were done by dMRI studies in PD patients [10,
19, 20]. In the neural tracts water flows normally. The use of diffusion tensor (DTI)
or diffusion weighted (DWI) was found to be helpful to quantify anisotropic
(directionality) loss or an increase in the amplitude of water diffusion and also in
the demonstration of neural tracts disruption. The detection of fractional anisotropy
(FA) in the regions of substantia nigra using diffusion tensor imaging increased the
MRI sensitivity to study the structural changes [11, 16, 21].
However, fMRI has been found to be useful in atypical PD, multiple system
atrophy (MSA), and progressive supranuclear palsy (PSP) [22]. And it cannot
differentiate between atypical PD and PD. Also this technique cannot diagnose early
stage Parkinsonism. Tasked based fMRI shows progressive changes in the regions of
11-4
putamen and motor cortex for patients with PD [10, 23]. Patients with PD have an
increased level of iron in the substantia nigra. MRI can detect these increased iron
levels using T2/T2* relaxometry. Here, as a replacement of increased iron concen-
tration we measure a change in the relaxation time constants using T2/T2* imaging.
Ferritin and neuromelanin are more sensitive to MRI techniques [9].
Neuromelanin (NM) is an autophagic end product of oxidation of catechol-
amines and subsequent reactions. NM pigments the dopaminergic neurons in the
substantia nigra as its level increases in the substantia nigra and locus coeruleus. Iron
binds to these pigmented neurons and forms NM–iron complexes [24, 25]. Now, iron
present as ferritin in brain is in high content but NM-MRI cannot detect it as it is not
paramagnetic, whereas NM–iron complexes in pigmented neurons are highly
paramagnetic. The reason is that NM–iron complexes have chelating groups which
bears a free stable radical and they also contain aliphatic chains, which helps in the
macromolecular free pool exchange. Ferritin on the other hand lacks these aliphatic
chains and the chelating groups in ferritin are amino acids [26].
11.4.2 Transcranial sonography (TCS)

TCS is a non-invasive, sensitive, inexpensive and potentially useful tool for screen-
ing. It is performed with no need of anesthesia. It is available widely as B-mode
neuroimaging technique that helps to assess the brain echogenicity using ultrasound.
Hyperechogenicity is said to be an echo which results due to deposits of fats in an
organ. Studies suggested that it might be able to detect the nigrostriatal disruptions
in individuals developing PD. Hence, it can be used as a diagnostic biomarker [9, 27,
28].
It is said that hyperechogenicity was associated with an increase in the iron
concentration. Elevated levels of iron is found in most of the PD patients. This is
because elevated iron level will lead to oxidative stress and then cause an injury to
the dopaminergic neurons. The main cause leading to PD. The echogenicity changes
occur due to functional and morphological changes in the brain. An ultrasound
device Sonos 5500 is widely used to measure the area of SN [9, 29, 30].
Across the transtemporal bone window, near the preauricular site, a transducer is
placed with 2.5 MHz phased-array to carry out TCS. The basal ganglia can be
visualized in the plane of the third ventricle. The SN, which appears as a butterfly
shaped structure with low echogenicity is identified within the brainstem surrounded
by basal cisterns with hyperechogenicity [28].
The signals of TCS do not help in detecting the progression of the disease as the
signals remain static for over a period of five years. It said that mid brain
hyperechogenicity is a trait in PD patients and not a marker. It also reflects the
iron deposition in midbrain rather than dopaminergic cell loss. The autopsy studies
suggested that even in the presence of nigral degeneration, the abnormal midbrain
signals were not detected by TCS in conditions like progressive supranuclear palsy
and multiple system atrophy [16, 31].
This technique has a number of limitations which includes: (1) TCS may not be a
feasible option for patients with insufficient bone window due to inability to image
11-5
the SN. (2) In healthy individuals it was found to show a high false positive rate
(9%). (3) There appeared to be lack of sensitivity towards progression of the disease.
(4) There is huge dependence on operator skill [27, 32].
11.5 Positron emission tomography (PET)

PET is an in vivo functional technique for neuroimaging that requires the use of
radionuclides. Radionuclides help to elucidate the integrity of the cerebral metab-
olism, dopaminergic system, the pathological accumulation of proteins, and brain
inflammation. 18F-dopa and 11C-raclopride are examples of radiotracers used to
image the integrity of the post and presynaptic nigrostriatal projections [11, 28].
11.5.1 Presynaptic and postsynaptic dopaminergic imaging

Very helpful data and diagnostic information is generated by PET imaging of the
nigrostriatal dopaminergic system. The most commonly known PET tracer,
18
F-dopa helps to measure the density of axons in the presynaptic nigrostriatal
region, and also measures the activity of the nigrostriatal enzyme aromatic amino
acid decarboxylase (AADC) protein. AADC helps in conversion of 18F-dopa to
18
F-dopamine. And in this way, it helps to estimate the dopaminergic storage pool
indirectly. It was found that the progression of PD was associated with uptake of
18
F-dopa [28, 33, 34].
F-dopa is actually a fluorinated form of L-dopa i.e. levodopa. The uptake of
18
F-dopa helps in reflecting the amount of dopamine produced by the dopaminergic
neurons through AADC. After the radiotracer is injected in the blood it quickly
crosses the blood-brain barrier (BBB) and then reaches the dopaminergic cells. Now,
it acts as a precursor of dopamine for dopaminergic integrity assessment and also
shows the monoamine disturbances accurately. Hence, this test can be used
specifically to distinguish PD from other neurodegeneration diseases [9, 35].
11
C-DTBZ is used as a radiotracer to label the vesicular monoamine transporter
type-2 (VMAT2). VMAT2 is involved in storing and packaging of monoamines into
the synaptic vesicles. 11C-DTBZ PET imaging studies showed that a decrease in the
striatal VMAT2 binding was found in PD patients which reflected the nigrostriatal
degeneration. Membrane dopamine transporter (DAT) 11C-methylphenidate (11C-MP)
is used as a ligand. The binding of this radio ligand to DAT is reduced to a greater
extent as compared to 11C-CTBZ. This shows that there is a marked down-regulation of
DAT activity in the striatal region [28]. Another radioligand which can be used for
DAT scan is CFT [7] and the 11C-raclopride is a radiotracer which bind to striatal post-
synaptic D2 receptors (figure 11.3) [28, 36, 37].
11.5.2 Assessment of cerebral glucose metabolism

For assessment of cerebral glucose metabolism, 11F-labelled fluorodeoxyglucose
(18F-FDG) is used as a radiotracer very commonly. It helps to reveal the preserved
glucose metabolism in the thalamus and the lentiform nucleus (LN). Also, in the
premotor and supplementary motor regions. It helps to distinguish PD patients from
multiple system atrophy (MSA) and progressive supranuclear palsy (PSP). Because
11-6
Figure 11.3. Binding of radio ligands to the receptors near the dopaminergic nerve terminal. (a) 11C-DTBZ
radioligand binds to vesicular monoamine transporter type-2 and is used in PET imaging. (b) 18F-dopa is a
radioligand helps to detect dopamine content by binding to 18F-DA. (c) 11C-MP radioligand binds to
membrane dopamine transporters and helps in detection of neurotransmitter imbalance. It is used in DAT
imaging.
PSP patients show hypometabolism of glucose and MSA patients show impaired
glucose metabolism instead of preserved glucose metabolism. An asymmetric
hypometabolism in the thalamus, frontoparietal cortical regions and the basal
ganglia was found to be the most common feature in corticobasal syndrome (CBS)
patients [28, 38].
11.6 Single photon emission computed tomography in

Parkinsonian disorders (SPECT)
SPECT imaging is one of the commonly used imaging techniques used in Parkinson
syndrome. It is used to keep a check on the cerebral perfusion changes in vivo and also
to measure the integrity of dopaminergic system in the nigrostriatal region. This
method involves the use of selective gamma-emitting cocaine analogs. The examples
are 123I-N-ω-fluoropropyl-2β-carbomethoxy-3β-(4-iodophenyl) nortropane (123I-FP-
CIT or 123I-ioflupane), they bind to DAT and another example is 123I-(S)-(−)-2-
hydroxy-3-iodo-6-methoxy-N-([1-ethyl-2-pyrrolidyl]-methyl) benzamide (123I-IBZM)
used as a ligand for dopamine D2 receptors.
The reuptake of dopamine from the synaptic cleft is mediated by DAT which is a
sodium-coupled transmembrane protein. D2 striatal dopamine receptors come
under G-protein-coupled inhibitory receptors that are expressed at both presynaptic
11-7
nigrostriatal axon terminals as well as postsynaptic dopamine target cells. Post

synaptic striatal neurons functioning is evaluated by using SPECT imaging for D2
receptors. One of the most widely used ligands for DAT is 123I-FP-CIT to quantify
DAT density. It is used widely because of its greater selectivity, faster kinetics, and
its compatibility with levodopa treatment. The changes in cerebral perfusion can be
measured by SPECT imaging with the help of lipophilic radiotracers [28].
DAT ligands non-selectively bind to the transporters of noradrenaline, serotonin
and dopamine. As dopaminergic transporters are in the majority in the striatum; due
to this it can provide a useful marker for reduction in the striatal uptake of such
tracers [16].
11.7 Diffusion tensor magnetic resonance imaging (DTI)

DTI in Parkinsonism is an in vivo tractography technique which indirectly allows
one to quantify the microstructural integrity of the brain by analysis of the degree of
displacement in space, i.e. fractional anisotropy and by overall displacement of
water molecules, i.e. mean diffusivity. As water is found in the brain along the
underlying microstructures of the tracts of white matter it can prove to be a useful
physical property to measure fractional anisotropy and mean diffusivity [21, 39].
Due to decrease in the flow of water along the damaged tracts of white matter
there is an increase in mean diffusivity and decrease in the fractional anisotropy. The
association of decreased fractional anisotropy in the SN in PD patients was found to
be unclear with respect to diseased severity. However, the use of high-resolution DTI
showed 100% sensitivity and specificity in distinguishing PD and controls. In the
olfactory tracts there was an increase in the mean diffusivity and decrease in the
fractional anisotropy among the PD patients, which reflects the olfactory distur-
bances usually occurring in PD [16].
11.8 Proton magnetic resonance spectroscopy (1H-MRS)

1
H-MRS is an in vivo imaging technique which is a non-invasive technique. It uses
the proton’s resonance frequency to measure the total number of biochemical
molecules in the brain. On a neurospectrograph the peaks of the relative concen-
trations of metabolites that contain protons appear near their characteristic
resonance frequencies. Different metabolites are used for MRS of neurodegenerative
diseases [40].
The levels of NAA or NAA/Cr was reduced in the lentiform nucleus, i.e. in the
putamen and globus pellidus, pre-SMA (supplementary motor area), temporo-
parietal and posterior cingulate cortex. With the help of 3D MRS it was found that
NAA/Cr ratio was increased in the rostal SN for patients with atypical PS and
controls. This was the reverse pattern of PD patients. Hence, it can be suggested that
there is a loss of neurons in the rostal SN in patients with PD. The levels of NAA
were found to be reduced in conditions like PD, PSP, MSA and CBS in varying
degrees as compared to controls [16, 41–43].
11-8
References
[1] Galvan A and Wichmann T 2008 Pathophysiology of Parkinsonism Clin. Neurophysiol. 119
1459–74
[2] Reichmann H 2010 Clinical criteria for the diagnosis of Parkinson’s disease Neurodegener.
Dis. 7 284–90
[3] Barber T R, Klein J C, Mackay C E and Hu M T M 2017 Neuroimaging in pre-motor
Parkinson’s disease NeuroImage Clin. 15 215–27
[4] Rizek P, Kumar N and Jog M S 2016 An update on the diagnosis and treatment of
Parkinson disease Can. Med. Assoc. J. 188 1157–65
[5] Tysnes O-B and Storstein A 2017 Epidemiology of Parkinson’s disease J. Neural Transm.
124 901–5
[6] de Lau L M and Breteler M M 2006 Epidemiology of Parkinson’s disease Lancet Neurol. 5
525–35
[7] Delenclos M, Jones D R, McLean P J and Uitti R J 2016 Biomarkers in Parkinson’s disease:
Advances and strategies Parkinsonism Relat. Disord. 22 S106–10
[8] Picillo M, Moccia M, Spina E, Barone P and Pellecchia M T 2016 Biomarkers of Parkinson’s
disease: recent insights, current challenges, and future prospects (New York: Dove), pp 1–13
[9] Lotankar S, Prabhavalkar K S and Bhatt L K 2017 Biomarkers for Parkinson’s disease:
recent advancement Neurosci. Bull. 33 585–97
[10] Strafella A P et al 2018 Imaging markers of progression in Parkinson’s disease Mov. Disord.
Clin. Pract. 5 586–96
[11] Politis M 2014 Neuroimaging in Parkinson disease: from research setting to clinical practice
Nat. Publ. Gr. 10 708–22
[12] Christopher L, Marras C, Duff-canning S and Koshimori Y 2015 Combined insular and
striatal dopamine dysfunction are associated with executive deficits in Parkinson’s disease
with mild cognitive impairment Brain 137 565–75
[13] Houle S, Ginovart N, Hussey D, Meyer J H and Wilson A A 2000 Imaging the serotonin
transporter with positron emission tomography: initial human studies with [11C] DAPP and
[11C] DASB Eur. J. Nucl. Med. 27 1–4
[14] Pasquini J et al 2018 Progression of tremor in early stages of Parkinson’s disease: a clinical
and neuroimaging study Brain 141 811–21
[15] Rizzi G and Tan K R 2017 Dopamine and acetylcholine, a circuit point of view in
Parkinson’s disease Front. Neural Circuits 11 110
[16] Brooks D J and Pavese N 2011 Imaging biomarkers in Parkinson’s disease Prog. Neurobiol.
95 614–28
[17] Pet A C et al 2017 Decreased intestinal acetylcholinesterase in early Parkinson disease
Neurobiology 88 775–81
[18] Remy P, Doder M, Lees A, Turjanski N and Brooks D 2005 Depression in Parkinson’s
disease: loss of dopamine and noradrenaline innervation in the limbic system Brain 128
1314–22
[19] Ofori E et al 2015 Increased free water in the substantia nigra of Parkinson’s disease: a single-
site and multi-site study Neurobiol. Aging 36 1097–104
[20] Kordower J H et al 2013 Disease duration and the integrity of the nigrostriatal system in
Parkinson’s disease Brain 136 2419–31
11-9
[21] Vaillancourt D E et al 2009 High-resolution diffusion tensor imaging in the substantia nigra
of de novo Parkinson disease Neurology 72 1378–84
[22] Quattrone A et al 2019 Refining initial diagnosis of Parkinson’s disease after follow-up: A
4-year prospective clinical and magnetic resonance imaging study Mov. Disord. 34 487–95
[23] Filippi M, Sarasso E and Agosta F 2019 Resting-state functional MRI in Parkinsonian
syndromes Mov. Disord. Clin. Pract. 6 104–17
[24] Isaias I U et al 2016 Neuromelanin imaging and dopaminergic loss in Parkinson’s disease
Front. Aging Neurosci. 8 196
[25] Kuya K, Shinohara Y, Miyoshi F, Fujii S, Tanabe Y and Ogawa T 2016 Correlation
between neuromelanin-sensitive MR imaging and (123)I-FP-CIT SPECT in patients with
Parkinsonism Neuroradiology 58 351–6
[26] Sulzer D et al 2018 Neuromelanin detection by magnetic resonance imaging (MRI) and its
promise as a biomarker for Parkinson’s disease NPJ Parkinsons Dis. 4 11
[27] Gitelman D and Simuni T 2012 Imaging studies as biomarkers of Parkinson’s disease
Imaging Med. 4 263–6
[28] Saeed U et al 2017 Imaging biomarkers in Parkinson’s disease and Parkinsonian syndromes:
current and emerging concepts Transl. Neurodegener. 6 8
[29] Bouwmans A E P, Vlaar A M M, Mess W H, Kessels A and Weber W E J 2013 Specificity
and sensitivity of transcranial sonography of the substantia nigra in the diagnosis of
Parkinson’s disease: prospective cohort study in 196 patients BMJ Open 3 e002613
[30] Ayton S and Lei P 2014 Nigral iron elevation is an invariable feature of Parkinson’s disease
and is a sufficient cause of neurodegeneration Biomed. Res. Int. 2014 581256
[31] Berg D et al 2002 Echogenicity of the substantia nigra: association with increased iron
content and marker for susceptibility to nigrostriatal injury Arch. Neurol. 59 999–1005
[32] He R, Yan X, Guo J, Xu Q, Tang B and Sun Q 2018 Recent advances in biomarkers for
Parkinson’s disease Front. Aging Neurosci. 10 1–19
[33] Nurmi E et al 2001 Rate of progression in Parkinson’s disease: a 6-[18F]fluoro-L-dopa PET
study Mov. Disord. 16 608–15
[34] Broussolle E et al 1999 The relation of putamen and caudate nucleus 18F-Dopa uptake to
motor and cognitive performances in Parkinson’s disease J. Neurol. Sci. 166 141–51
[35] Niccolini F and Politis M 2016 A systematic review of lessons learned from PET molecular
imaging research in atypical Parkinsonism Eur. J. Nucl. Med. Mol. Imaging 43 2244–54
[36] Lee C S et al 2000 In vivo positron emission tomographic evidence for compensatory changes
in presynaptic dopaminergic nerve terminals in Parkinson’s disease Ann. Neurol. 47 493–503
[37] Brighina L et al 2013 Analysis of vesicular monoamine transporter 2 polymorphisms in
Parkinson’s disease Neurobiol. Aging 34 1712.e9–13
[38] Eckert T et al 2005 FDG PET in the differential diagnosis of Parkinsonian disorders
Neuroimage 26 912–21
[39] Geng D-Y, Li Y-X and Zee C-S 2006 Magnetic resonance imaging-based volumetric analysis
of basal ganglia nuclei and substantia nigra in patients with Parkinson’s disease
Neurosurgery 58 256–62 discussion 256–62
[40] Firbank M J, Harrison R M and O’Brien J T 2002 A comprehensive review of proton
magnetic resonance spectroscopy studies in dementia and Parkinson’s disease Dement.
Geriatr. Cogn. Disord. 14 64–76
11-10
[41] Abe K et al 2000 Proton magnetic resonance spectroscopy of patients with Parkinsonism
Brain Res. Bull. 52 589–95
[42] Klietz M et al 2019 Altered neurometabolic profile in early Parkinson’s disease: a study with
short echo-time whole brain MR spectroscopic imaging Front. Neurol. 10 777
[43] Camicioli R M et al 2004 Posterior cingulate metabolic changes occur in Parkinson’s disease
patients without dementia Neurosci. Lett. 354 177–80
11-11

ILMA - Volume V - Neurological Disorders and Imaging Physics - App

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ILMA - Volume V - Neurological Disorders and Imaging Physics - App

Uploaded by

Copyright:

Available Formats

Neurological Disorders and

Imaging Physics, Volume 5

IOP Publishing, Bristol, UK

ISBN 978-0-7503-2723-7 (ebook)

Published by IOP Publishing, wholly owned by The Institute of Physics, London

IOP Publishing, Temple Circus, Temple Way, Bristol, BS1 6HG, UK

1 Majority vote of machine learning methods for 1-1

2 Delineation of epileptogenic zone 2-1

2.3.1 Introduction 2-4

3 Dyslexia: behavioral and biological correlates and treatment 3-1

3.3 Phenotypes (behavioral markers) and genetic bases of dyslexia 3-3

4 Cholesterol and oxidized cholesterol derivatives: 4-1

5 Computer assisted diagnosis of gait dynamics 5-1

6 Rett syndrome 6-1

7 Knowledge about epilepsy in university health students 7-1

8 Current methods and new trends in signal processing 8-1

9 ‘They labelled me ignorant’: the role of neuroscience 9-1

9.2.1 The participants 9-4

10 Advances in epilepsy: from gender to genetics 10-1

11 Neuroimaging in Parkinson’s disease 11-1

11.4 Imaging midbrain structural changes in PD 11-4

Neurological Disorders and Imaging Physics, Volume 5

This research represents the practical application of machine learning in healthcare

doi:10.1088/978-0-7503-2723-7ch1 1-1 ª IOP Publishing Ltd 2020

1.1.1 Epilepsy and epileptic seizure

1.1.1.1 Epilepsy seizure prediction

A system controlling seizures may be an effective anticonvulsive agent in the

Figure 1.1. SPH and SOP.

with no seizure during SOP are classiﬁed as false predictions (Winterhalder et al

1.1.2 EEG signals related to epileptic seizures

1.2 Literature review

Figure 1.2. Interictal, preictal and ictal signals.

solutions on this same topic. It is a research in which some certain signal

artiﬁcial neural networks with the back-propagation algorithm; support vector

between preictal/ictal and interictal EEG stages is implemented. The Freiburg

1.3.2 EEG signal processing techniques

1.3.2.1 Discrete wavelet transformation (DWT)

Table 1.1. CHB-MIT subjects—details.

coefﬁcients, by the process of scaling the ‘mother’ wavelet function. A base

[i ] = who[i ] = ∑x[k ]l [2i − k ]k (1.3)

1.3.2.2 Wavelet packet transform (decomposition)—WPT (WPD)

1.3.2.3 Multiscale principal component analysis—MSPCA

1.3.3 Meta heuristic methods

1.3.3.1 Naive Bayes

1.3.3.2 Decision tree and decision tree classifiers

1.3.3.3 k-Nearest neighbor

1.3.3.4 Support vector machine (SVM)

P = {(xi , )∣xi ∈ Rk , yi ∈ { −1, 1}} i = 1n (1.5)

1.3.3.5 Artificial neural networks (ANNs)

1.4 Experimental setup

1.4.2 Data preparation with de-noising and feature extraction

• Ratio of absolute mean values:

1.4.3 Training the data—Matlab GUI

Figure 1.3. Matlab GUI.

1.4.4 Data testing, processing and seizure prediction

Figure 1.4. Example of predicted seizure.

Figure 1.5. No seizure detected.

Figure 1.4 shows an example of positive prediction. Each rectangle represents an

Table 1.3. Comparison table for seizure prediction on CHB-MIT database.

Author Accuracy (%)

Khan et al (2012) 91.8

Figure 1.6. Seizure prediction for seizure 1 (CHB05).

Figure 1.7. Seizure prediction for seizure 2 (CHB05).