2019 Book TheHandbookOfCufflessBloodPres

Josep Solà · Ricard Delgado-Gonzalo
Editors
The Handbook
of Cuffless
Blood Pressure
Monitoring
A Practical Guide for Clinicians,
Researchers, and Engineers
The Handbook of Cuffless Blood Pressure
Monitoring
Josep Solà • Ricard Delgado-Gonzalo
Editors
The Handbook of Cuffless

Blood Pressure Monitoring
A Practical Guide for Clinicians, Researchers,
and Engineers
Editors
Josep Solà Ricard Delgado-Gonzalo
Aktiia SA Swiss Center for Electronics and
Neuchâtel, Switzerland Microtechnology (CSEM, Centre Suisse
d’Electronique et de Microtechnique)
Neuchâtel, Switzerland
ISBN 978-3-030-24700-3 ISBN 978-3-030-24701-0 (eBook)

https://doi.org/10.1007/978-3-030-24701-0
© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims
in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Let us start with a typical scenario from clinical practice today.

During an annual physical exam:
143/97, Mr. Vailet, I have the feeling you are developing hypertension.
The day after, at Mr. Vailet’s home:
134/88
Two days after:
129/84
No further measurements until the day before the next physical exam, at Mr. Vailet’s
home:
148/95
Back to the physician’s office:
Mr. Vailet, where are the weekly blood pressure readings I asked you to do?
Measuring arterial blood pressure is a fundamental action when checking a
patient’s overall health. Nowadays, physicians and patients alike still rely on a
century-old technology that requires the inflation of a cuff around the arm. This
leads to intermittent, sparse, or nonexistent monitoring of a patient’s health status
over long periods of time. As a result, millions of individuals are being wrongly
treated, underdiagnosed, or simply not diagnosed at all.
For more than a decade, an assembly of scientists, physicians, researchers, and
engineers have been working on the development and validation of novel technolo-
gies to overcome the burden of cuff-based blood pressure measurements. This book
is a tribute to this collective effort.
In 2017, we started crafting “The Handbook of Cuffless Blood Pressure
Monitoring: A Practical Guide for Clinicians, Researchers, and Engineers” that
you have now in your hands. From the beginning, we aimed at creating the first
v
vi Preface
comprehensive publication providing an overview of the emerging field of cuffless

blood pressure monitoring. We gathered the most knowledgeable authors around the
world who could summarize the basics, the medical context, the potential, and the
technical challenges of cuffless blood pressure monitoring. This work is now done.
Dear reader, if you are a researcher, clinician, engineer, journalist, investor, or
student who wants to get into the field of cuffless blood pressure, we are convinced
you will enjoy diving into these chapters.
Neuchâtel, Switzerland Josep Solà

Ricard Delgado-Gonzalo
Acronyms
AAMI Association for the Advancement of Medical Instrumentation

ABPM Ambulatory blood pressure monitoring
AI/AIx Augmentation index
ANSI American National Standards Institute
AP Augmented pressure
APG Acceleration plethysmogram
BCG Ballistocardiography
BHS British Hypertension Society
BP Blood pressure
CT Computed tomography
DBP/DIA Diastolic blood pressure
DIA Diastolic blood pressure
DL Deep learning
DNN Deep neural network
DPTI Diastolic pressure-time index
DT Diastolic time
ECG Electrocardiography
EHS European Home Systems
EMAT Electromechanical activation time
ESH European Society of Hypertension
ESP End-systolic pressure
FDA Food and Drug Administration
FF Form factor
FQP Four-quadrant plot
GCP Good clinical practice
GTF Generalized transfer function
HR Heart rate
ICG Impedance cardiography
IDE Investigational device exemption
IEEE Institute of Electrical and Electronics Engineers
IPG Impedance plethysmography
vii
viii Acronyms
IR Infrared
ISO International Organization for Standardization
IVCT Isovolumic contraction time
LED Light-emitting diode
LSTM Long short-term memory
LVET Left ventricular ejection time
MAD Mean absolute differences
MAP Mean arterial pressure
MAPD Mean absolute percentage difference
MDBP Mean diastolic blood pressure
ML Machine learning
MSBP Mean systolic blood pressure
NIBP Noninvasive blood pressure
NN Neural network
NPMA N-point moving average
PAT Pulse arrival time
PCG Phonocardiography
PDA Pulse decomposition analysis
PEP Pre-ejection period
PMA Premarket approval
PP Pulse pressure
PPG Photoplethysmography
PTT Pulse transit time
PWA Pulse wave analysis
PWD Pulse wave decomposition
PWTT Pulse wave transit time
PWV Pulse wave velocity
RI Reflection index
RMSE Root mean square error
RNN Recurrent neural network
SBP Systolic blood pressure
SCG Seismocardiography
SD Standard deviation
SEVR Subendocardial viability ratio
SI Stiffness index
SPTI Systolic pressure-time index
SVM Support vector machine
SVR Support vector regression
SYS Systolic blood pressure
Contents
1 Cuffless Blood Pressure Monitoring and the Advent of a

New Era in Medicine and Society�� 1
Alberto Avolio, Fatemeh Shirbani, Isabella Tan, and Mark Butlin
2 Clinical Relevance of Continuous and Cuffless
Blood Pressure Monitoring �� 9
Gianfranco Parati
3 A Historical Journey on the Physiology of Blood
Pressure Monitoring�� 15
Audrey Adji and Michael F. O’Rourke
4 The Definition and Architecture of Cuffless Blood
Pressure Monitors�� 31
Josep Solà
5 Pulse Arrival Time Techniques �� 43
Marshal S. Dhillon and Matthew J. Banet
6 Pulse Wave Velocity Techniques�� 61
Jim Li
7 Pulse Decomposition Analysis Techniques�� 75
Martin C. Baruch
8 Pulse Wave Analysis Techniques�� 107
Martin Proença, Philippe Renevey, Fabian Braun, Guillaume Bonnier,
Ricard Delgado-Gonzalo, Alia Lemkaddem, Christophe Verjus,
Damien Ferrario, and Mathieu Lemay
9 Machine Learning Techniques�� 139
Xiaorong Ding
ix
x Contents
10 Initialization of Pulse Transit Time-Based Blood

Pressure Monitors�� 163
Ramakrishna Mukkamala and Jin-Oh Hahn
11 Key Regulatory Aspects and the Importance of
Consensus Standards in Bringing Devices to Market�� 191
Carole C. Carey
12 Design of Clinical Trials to Validate Cuffless
Blood Pressure Monitors �� 203
Willem J. Verberk
13 Cuffless Blood Pressure Monitoring: The Future
for the Evaluation and Management of Hypertension �� 225
George S. Stergiou
Index�� 231
Contributors
Audrey Adji St Vincent’s Clinical School, University of New South Wales, Sydney,

NSW, Australia
Department of Biomedical Sciences, Faculty of Medicine and Health Sciences,
Macquarie University, Sydney, NSW, Australia
Alberto Avolio Department of Biomedical Sciences, Faculty of Medicine and
Health Sciences, Macquarie University, Sydney, NSW, Australia
Matthew J. Banet toSense, Inc., San Diego, CA, USA
Martin C. Baruch Caretaker Medical LLC, Charlottesville, VA, USA
Guillaume Bonnier Swiss Center for Electronics and Microtechnology (CSEM,
Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland
Fabian Braun Swiss Center for Electronics and Microtechnology (CSEM, Centre
Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland
Mark Butlin Department of Biomedical Sciences, Faculty of Medicine and Health
Sciences, Macquarie University, Sydney, NSW, Australia
Carole C. Carey C3-Carey Consultants, LLC, Fulton, MD, USA
Ricard Delgado-Gonzalo Swiss Center for Electronics and Microtechnology
(CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel,
Switzerland
Marshal S. Dhillon toSense, Inc., San Diego, CA, USA
Xiaorong Ding Institute of Biomedical Engineering, Department of Engineering
Science, University of Oxford, Oxford, UK
Damien Ferrario Swiss Center for Electronics and Microtechnology (CSEM,
xi
xii Contributors
Jin-Oh Hahn Department of Mechanical Engineering, University of Maryland,

College Park, MD, USA
Mathieu Lemay Swiss Center for Electronics and Microtechnology (CSEM,
Alia Lemkaddem Swiss Center for Electronics and Microtechnology (CSEM,
Jim Li Global Medical Affairs, Omron Healthcare, Inc., Lake Forest, IL, USA
Ramakrishna Mukkamala Department of Electrical and Computer Engineering,
Michigan State University, East Lansing, MI, USA
Michael F. O’Rourke St Vincent’s Clinical School, University of New South
Wales, Sydney, NSW, Australia
Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia
Gianfranco Parati Department of Medicine and Surgery, University of Milano-
Bicocca, Milan, Italy
Department of Cardiovascular, Neural and Metabolic Sciences, Istituto Auxologico
Italiano, IRCCS, San Luca Hospital, Milan, Italy
Martin Proença Swiss Center for Electronics and Microtechnology (CSEM,
Philippe Renevey Swiss Center for Electronics and Microtechnology (CSEM,
Fatemeh Shirbani Department of Biomedical Sciences, Faculty of Medicine and
Health Sciences, Macquarie University, Sydney, NSW, Australia
Josep Solà Aktiia SA, Neuchâtel, Switzerland
George S. Stergiou Hypertension Center STRIDE-7, National and Kapodistrian
University of Athens, Third Department of Medicine, Athens, Greece
Isabella Tan Department of Biomedical Sciences, Faculty of Medicine and Health
Sciences, Macquarie University, Sydney, NSW, Australia
Willem J. Verberk Microlife AG, Widnau, Switzerland
CARIM School for Cardiovascular Diseases, Maastricht University, Maastricht,
The Netherlands
Christophe Verjus Swiss Center for Electronics and Microtechnology (CSEM,
Chapter 1
Cuffless Blood Pressure Monitoring
and the Advent of a New Era in Medicine
and Society
Alberto Avolio, Fatemeh Shirbani, Isabella Tan, and Mark Butlin
Abstract Measurement of arterial blood pressure (BP) by the brachial cuff sphyg-
momanometer has been a cornerstone of modern medicine, and notwithstanding its
limitations of intermittent BP monitoring, the cuff sphygmomanometer has not been
surpassed by any other noninvasive methodology. However, advances in sensor
technology for arterial pulse detection have paved the way for the potential develop-
ment of devices for cuffless measurement of BP, with the prospect of continuous
monitoring. Just as the cuff sphygmomanometer could be considered as having pro-
vided a significant step in establishing elevated BP as a major factor of cardiovascu-
lar risk since the turn of the twentieth century, cuffless BP monitoring might be
considered a disruptive technology for continuous BP monitoring in healthy indi-
viduals during daily living and could become an integral component of modern
digital health platforms in the twenty-first century. This volume of the Handbook of
Cuffless Blood Pressure Monitoring addresses the significant challenges faced by
this methodology. The underlying physiology and hemodynamic principles involved
in the generation of the arterial pressure pulse are complemented by the clinical
relevance of the novel cuffless methodologies in relation to the conventional cuff
sphygmomanometer. The underlying pressure dependency of arterial properties,
which is at the base of the cuffless technique employing pulse transit measurement
as a surrogate measure of BP, is addressed in association with instrumentation, mea-
surement techniques, calibration procedures, device validation and regulatory
requirements. This book provides a timely and comprehensive platform on how to
approach the critical question of whether cuffless is the future of BP monitoring.
Keywords Arterial stiffness · Pulse wave velocity · Pulse transit time ·

Sphygmomanometer · Blood pressure · Cuffless techniques · Peripheral pulse ·
Arterial hemodynamics · Big data
A. Avolio (*) · F. Shirbani · I. Tan · M. Butlin

Department of Biomedical Sciences, Faculty of Medicine and Health Sciences,
Macquarie University, Sydney, NSW, Australia
e-mail: alberto.avolio@mq.edu.au
© Springer Nature Switzerland AG 2019 1

J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure
Monitoring, https://doi.org/10.1007/978-3-030-24701-0_1
2 A. Avolio et al.
Introduction
There is a notable coincidence between the accepted dates marking the publication
of the first clinical use of the cuff sphygmomanometer by Scipione Riva-Rocci
(15 December 1896) in Turin, Italy and the first projection of cinema to a paying
audience by brothers Auguste and Louis Lumière (28 December 1895) in Paris,
France. In addition to the remarkable proximity of the official dates, only 1 year
apart, the two milestone events have a common origin through the work of French
physiologist Étiene-Jules Marey in his studies of the circulation of blood. His evolv-
ing attempts to obtain quantitative representation of the force of the arterial pulse as
detected and registered by graphic means paved the way for the invention of the
sphygmograph, This led to the construction of devices for achieving time-lapsed
photography to capture movement, culminating in series of ground-breaking publi-
cations and in the coining of the term chronophotography, a technique used to study
movement in flight [1] and which influenced the seminal flight experiment of the
Wright brothers in 1903 at Kitty Hawk, marking the birth of modern aviation.
Although the arterial pulse has been used in medical diagnoses and described
qualitatively since antiquity, the use of the sphygmograph enabled the first ever
quantitative registration of any physiological parameter—a cuffless measure of arte-
rial pressure, and long before the registration of the electrocardiogram. It was the
efforts of many scientists and inventors such as Marey in the search for mechanical
devices capable of producing an image of the varying morphology of the arterial
pulse, coupled with advances in photography, that enabled the evolution and conver-
gence of sphygmography and chronophotography. The one resulting in the advent
and clinical use of the cuff sphygmomanometer by the Italian physician Riva-Rocci
and the other in the promotion of cinema by French scientists and entrepreneurs
such as the Lumière brothers in the late nineteenth century. With the dawn of the
twentieth century, technological advances in the form of the cuff sphygmomanom-
eter and cinema produced respective fundamental changes in the way medicine was
practiced and in the way the moving image was used for enhanced communication
and enrichment of the human experience. Both have had an immeasurable impact
on human health and culture in society.
uffless Monitoring of Blood Pressure: A Disruptive Step

C
in Health Management
The above advances were not made in isolation, but were integral component parts
of a movement of progressive development of ideas and concepts that revolution-
ized scientific thinking and technological applications in the eighteenth and nine-
teenth centuries. History is rich with iconic milestones such as James Watt’s
development of the steam engine (cc 1780s) that was a major driving force for the
industrial revolution. Perhaps one of the most significant fundamental unifying
1 Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine… 3
concepts in human history, the description of electromagnetic radiation by James

Maxwell in 1865, has extended the application of human endeavor beyond the
Newtonian view of the world to Einstein’s concepts of relativity and to quantum
mechanics. In modern parlance, and given the significant impact these develop-
ments have made on society as a whole, they might be described as highly “disrup-
tive” in comparison to what came before.
And it is precisely the notion of “disruption” that is at the center of the transition
from blood pressure being measured intermittently as a clinical parameter by the
cuff sphygmomanometer to blood pressure being monitored continuously by cuff-
less devices. Just as the intellectual and scientific pathways that led to the develop-
ment of the cuff sphygmomanometer were intertwined with advances and pitfalls in
other avenues of scientific and technological endeavor in the nineteenth century, the
development of cuffless blood pressure technology will tread similar ground in the
twenty-first century. The difference being that the disruptive elements in the cuffless
journey are supported and boosted by the immense power of all that is digital—the
Internet, global positioning systems, satellite communication, cloud computing,
miniaturization of computer chips and electronic circuits, intelligent algorithms,
developments in sensor technology supported by intelligent materials. The differ-
ence will also extend to the use of the devices. Whereas the cuff sphygmomanom-
eter, virtually unchanged since its inception, has been essentially used for detection
of markers associated with pathology (hypertension), cuffless monitoring of blood
pressure will be incorporated in wearable devices, thus enabling collection of data
in normal daily living of individuals in society, in the form of “big data.” It is yet not
known in what way this form of data collection will modify the practice of medicine
and how it will affect the understanding of blood pressure profiles with daily habits
or lead to phenotypic classifications of normal individuals in the context of genetic
analysis and precision medicine. However, there is a precedent. When brachial cuff
blood pressure values were systematically collected using pen and paper by life
insurance companies in the early twentieth century [2], it led to the understanding
of how asymptomatic high blood pressure can be one of the most powerful factors
of cardiovascular risk.
Underlying Principles of Cuffless Methodology
An important overarching concept underlying cuffless measurement of blood pres-

sure is the fundamental relationship between transmural pressure and mechanical
properties of the arterial wall which influence wave propagation phenomena [3].
This is the pressure dependency of the material stiffness of all blood vessels. This
property is present in all species with pressurized circulatory systems and is a fun-
damental evolutionary property of arterial design [4]. Since, in any physical system,
wave propagation is determined by the bulk modulus of the material, the speed of
any disturbance (due to cardiac contraction and ejection) that travels along the arte-
rial wall is directly related to vessel stiffness. And given the essential relationship of
4 A. Avolio et al.
arterial stiffness and distending pressure, any measure of the velocity of the travel-
ling pulse wave would be a measure of arterial pressure. This is an important con-
sideration because pulse wave velocity (PWV) is directly related to blood pressure
and so an accurate measurement of PWV should theoretically deliver a measure of
blood pressure, provided that the relationship of PWV and blood pressure of the
specific arterial path length is known. Changes in blood pressure would then be
registered as changes in PWV, or more specifically, for a fixed distance, as changes
in pulse transit time (PTT) [3]. Hence, on theoretical considerations, cuffless mea-
surement of blood pressure using PTT could be considered to be a more robust
measure of intra-arterial pressure than that obtained by using indirect surrogate sig-
nals such as appearance of a distal pulse (palpatory method), Korotkoff sounds (aus-
cultatory method) or features of the envelope of the oscillation of cuff pressure
during cuff deflation (oscillometric method).
The Handbook of Cuffless Blood Pressure Monitoring
This volume of The Handbook of Cuffless Blood Pressure Monitoring, the first of its
kind, promises to be a guide in the modern twenty-first century journey of novel
developments of methods and technology for monitoring of blood pressure, meth-
odologies and techniques which go beyond the traditional cuff sphygmomanometer.
It is a timely endeavor, and its publication will be an important milestone in this new
field. It is both important and essential. Wearable, mainly wrist-worn devices are
being produced that purport to measure blood pressure. Being connected to the
Internet, they transmit data to databanks that log blood pressure values for individu-
als during daily activities. Being consumer devices, there is generally little or no
regulatory requirement for the blood pressure measurement, other than device
safety. With increasing use of these devices, it is conceivable that blood pressure
data will be mined from big data sources and potentially “useful” knowledge will be
extrapolated. However, there is no guarantee that the information from such data has
any reliable relation to continuous blood pressure or any potential physiological
significance. This is because of the inherent complexity in translating the noninva-
sive cuffless surrogate signal to a physiological arterial pressure. That is, the reli-
ability of the basic methodology of calibration [5, 6]. This is in contrast to the cuff
sphygmomanometer, where the cuff pressure is actually measured with high preci-
sion, and specific levels of cuff pressure are associated with arterial phenomena,
such as appearance and disappearance of Korotkoff sounds associated with systolic
and diastolic pressure, although to a varying degree of association and correlation.
The chapters in this book are comprehensive and cover a wide range of topics, all
of which are highly relevant to the scientific, technological, industrial, and clinical
aspects of the field of cuffless and continuous measurement of arterial blood pres-
sure. The material presented in all the chapters offers a robust platform on which to
launch this new field and which builds on the large amount of published work based
on basic concepts of the circulation involved in the genesis of the pressure pulse,
sensor technology, device development and fabrication, signal processing, deep

learning strategies, modelling, calibration procedures, in vivo, in vitro, and in silico
experimentation, and methods of validation.
This book addresses critical aspects relevant to the monitoring of blood pressure
using cuffless techniques in 12 chapters. The following 11 chapters can be divided
into three broad topic sections:
Topic Section 1 (Chaps. 2 and 3) describes the underlying physiology and hemo-
dynamic principles involved in the generation of the arterial pressure pulse and the
clinical relevance of the novel cuffless methodologies in relation to the conventional
cuff sphygmomanometer. The important distinction between the new and old tech-
niques for blood pressure monitoring is that the cuff sphygmomanometer is usually
associated with the detection, treatment and management of hypertension, thus a
pathological condition. Novel, cuffless techniques, embedded in wearable devices,
will be associated with monitoring continuous changes in blood pressure predomi-
nantly in normal individuals during daily activities. This will undoubtedly have an
impact on the meaning of such things as blood pressure thresholds that separate the
“normal” state from the “pathological” state. This suggests significant “disruption”
in the actual understanding of what constitutes “hypertension” in association with a
plethora of other health data obtained by wearable devices and interpreted by data
analytics.
Topic Section 2 (Chaps. 4–9) describes the areas of cuffless devices involving
instrument components, measurement techniques and calibration procedures. This
Topic Section is critical to the fundamental understanding of the essential compo-
nents of what constitutes a cuffless device for monitoring of arterial blood pressure.
The basic requirement is the detection of the arterial pulse, with sensor modality
being key for reliable pulse detection. An important distinction between sensors
used for cuffless devices is that they do not necessity need to provide a quantitative
measure of the force of the arterial pulse, but rather a fiducial signal which is in
synchrony with the beating heart. The measurement of arterial pressure is based in
temporal characteristics, so there is no requirement for absolute grading of mea-
sured signals. This enhances the capacity of available sensors for signal detection
and the scope for instrumentation to be adapted to wearable devices. However,
while sensors may be sufficiently robust for signal detection, the conversion of a
measured time delay (irrespective of the degree of accuracy) to a physiologically
relevant blood pressure remains one of the most challenging aspects of this field. To
date, calibration issues remain unresolved, particularly in terms of stability, fre-
quency of performing calibrations, procedures involved in obtaining adequate
ranges of blood pressure and the differences in relationships between pulse transit
time and arterial pressure for different arterial sites, or even at different levels of
blood pressure. The limitations related to inconsistent physiological relationships
can be addressed using data-driven procedures involving machine learning tech-
niques [7]. However, it is not yet known how different databases can be sufficiently
regulated to produce reliable training platforms for a broad range of algorithms
applied to machine learning approaches for blood pressure monitoring.
6 A. Avolio et al.
Topic Section 3 (Chaps. 10–12) addresses the important area of device validation
in the context of regulatory requirements and standards. To a certain extent, there is
a conceptual overlap between “cuffless” and “continuous” techniques for blood
pressure monitoring which is not adequately addressed by the current standards.
Hence, there is a need for clear and definite guidance for evaluation from regulatory
authorities so as to provide sufficient certainty for the medical devices industry.
Finally, with appropriate validation, future projections are made on the expectations
of the use of cuffless devices for diagnosis and management of hypertension.
However, this will need to be in the context of the potential disruption to the concept
of hypertension and blood pressure thresholds.
All chapters are comprehensive and authoritative. They cover a broad range of
relevant topics that enable and enhance understanding of the field of cuffless blood
pressure monitoring at a considerable depth. While there will be some necessary
and unavoidable repetition of some sections in different chapters, this is a positive
aspect as it highlights their relative importance and contributes to placing this new
field in the historical, technological and social continuum.
Is “Cuffless” the Future of Blood Pressure Monitoring?
The comprehensive nature of this volume brings out the advances as well as the
formidable challenges facing the journey of device development for reliable and
continuous monitoring of blood pressure using cuffless techniques. The “state of the
art” suggests that some milestones have already been achieved. Sensor technology
for pulse detection is highly evolved and robust. Advances in component miniatur-
ization and complex chip design have enabled the explosion of wearable devices
incorporating measurement and processing of physiological signals, providing
information on metrics that can guide decisions on health management. However,
while some of these metrics are highly reliable (e.g., heart rate, blood oxygen levels,
and many others) and have been available for some time, accurate, reliable continu-
ous, cuffless monitoring of blood pressure has presented insurmountable challenges.
There have been many patent submissions, start-up companies, and scientific pub-
lications, but to date there is no device that is universally accepted by the wider
community beyond research laboratories and company boardrooms.
This book will make a significant contribution to providing an informed answer
to this important question.
References
1. Marey E-J. Le vol des insectes étudié par la chronophotographie. La Nat. 1892;20(1):135–8.
2. Fischer JW. The diagnostic value of the sphygmomanometer in examinations for life insurance.
JAMA. 1914;63:3.
3. Mukkamala R, Hahn JO, Inan OT, Mestha LK, Kim CS, Töreyin H, Kyal S. Toward ubiquitous
blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng.
2015;62:1879–901.
4. Shadwick RE. Mechanical design in arteries. J Exp Biol. 1999;202:3305–13.
5. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio AP. Cuffless estimation of blood
pressure: importance of variability in blood pressure dependence of arterial stiffness across
individuals and measurement sites. IEEE Trans Biomed Eng. 2018;65:2377–83.
6. Ding X, Zhang Y, Tsang HK. Impact of heart disease and calibration interval on accuracy of
pulse transit time-based blood pressure estimation. Physiol Meas. 2016;37:227–37.
7. Khalid SG, Zhang J, Chen F, Zheng D. Blood pressure estimation using photoplethysmog-
raphy only: comparison between different machine learning approaches. J Healthc Eng.
2018;2018:1548647.
Chapter 2
Clinical Relevance of Continuous
and Cuffless Blood Pressure Monitoring
Gianfranco Parati
Abstract The assessment of dynamic features of blood pressure, which represent a

response of cardiovascular control mechanisms to environmental stimulations and
to daily life challenges, not only offers important insights into cardiovascular regu-
lation patterns but also carries clinically relevant information. Availability of tools
for continuous blood pressure monitoring represents a key step to make such an
assessment possible, but its implementation in daily practice requires noninvasive,
simple and minimally intrusive methods. These methods are expected to overcome
the well-known limitations characterizing the conventional approach to blood pres-
sure measurement based on discontinuous blood pressure readings obtained through
repeated arm cuff inflations. In such a perspective, techniques able to provide con-
tinuous blood pressure monitoring without the need of a cuff inflation would be
welcome.
Keywords Continuous blood pressure monitoring · Blood pressure variability ·

Arm cuff inflation · Cuffless blood pressure measurement technology
Introduction
Blood pressure (BP) is one of the most dynamic physiologic variables among those
routinely measured in clinical practice, consisting of a series of pulse waves con-
tinuously changing in terms of both frequency and amplitude. BP is indeed charac-
terized by continuous and significant changes occurring over different time windows,
with beat-by-beat oscillations being intertwined in a complex manner with
G. Parati (*)
Department of Medicine and Surgery, University of Milano-Bicocca, Milan, Italy
Department of Cardiovascular, Neural and Metabolic Sciences, Istituto Auxologico Italiano,
IRCCS, San Luca Hospital, Milan, Italy
e-mail: gianfranco.parati@unimib.it

10 G. Parati
fluctuations occurring from minute to minute and from hour to hour, at the time of
the different behaviors which occur over the day and night, Further complexity is
added by additional and more long-lasting fluctuations which may be observed over
different days, over different seasons, and at the time of consecutive visits by a phy-
sician even in a time window of several years [1].
In physiological conditions these BP variations largely represent a response of
cardiovascular control mechanisms to environmental stimulations and to daily life
challenges, aimed at maintaining the so-called cardiovascular “homeostasis,” being
additionally influenced by food, physical exercise, or sleep patterns.
Moreover, sustained increases in BP variability may also reflect intrinsic altera-
tions in cardiovascular regulatory mechanisms or the effect of underlying pathologi-
cal conditions responsible for a dysregulation of neural and humoral factors involved
in modulating cardiac and vascular functions.
The interest in the assessment of these BP variations comes from the evidence
that they may have important clinical significance and prognostic implications, as
demonstrated by a series of experimental, clinical, and population studies. Indeed,
enhanced BP fluctuations have been associated with a higher risk of subclinical
organ damage, cardiovascular events, and cardiovascular and all-cause mortality
independently and on top of what determined by elevated average BP values
[2–4].
Thus, theoretically there is no doubt that the most accurate and detailed approach
to the assessment of BP behavior in daily life would benefit from the possibility of
performing continuous beat by beat recording of BP over 24 h.
However, such continuous recordings are not easy to implement in daily prac-
tice, where discontinuous BP measurements obtained through repeated arm cuff
inflations still represent the method routinely applied. In fact, BP assessment in
daily practice is currently based on two main methods. The first is the auscultatory
method, based on the occlusion arm cuff-based technique introduced in 1896 by
Scipione Riva-Rocci, coupled with use of Korotkoff sounds. The second is the oscil-
lometric method, which makes use of the application of repeated automated cuff
inflations to directly measure mean BP and then systolic and diastolic BP values are
estimated through proprietary algorithms.
It is by means of these measurements that in year 2019 we are still currently
managing hypertensive and cardiovascular patients in daily practice. Although these
approaches have been shown to carry important clinical diagnostic and prognostic
information when applied either in a clinic environment or in daily life conditions
out of a clinic setting, they are not free from major limitations, which include the
inherent inaccuracy of the techniques used, the inability to account for the dynamic
nature of blood pressure and for its fluctuations over time and the difficulty in
obtained undisturbed readings that might reflect subjects’ actual BP patterns both
during wake and sleep.
In fact, the arm cuff inflation is often itself responsible for difficulties, as it may
be responsible for discomfort and sometimes even pain in patients in whom high
levels of air pressure might be required for a proper BP measurement. These prob-
lems are in particular relevant when focussing on nighttime BP measurement, that
2 Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring 11
is, on the measurements of BP obtained during sleep which have been shown to
carry highly relevant clinical information. Nocturnal cuff inflations may indeed
significantly disturb sleep of the subjects in whom they are performed, and may
importantly interfere with the BP levels they are aimed at assessing. This is exem-
plified by the fact that conventional 24 h ambulatory BP monitoring, which is based
on discontinuous and frequent BP measurements (typically every 15–30 min), in
spite of its acknowledged clinical value is poorly accepted by many patients due to
discomfort associated with repeated cuff inflations, which during the night may
significantly worsen the quality of sleep [5]. Even if in many patients this does not
affect BP levels [6], in some, an increase in nocturnal BP can be induced, leading
to an artifactual reduction in BP dipping pattern with a loss of its prognostic signifi-
cance, as reported previously for untreated hypertensive patients with significant
sleep deprivation during 24 h BP monitoring [7].
These methodological problems with currently available techniques for BP
assessment make it difficult to clarify a number of still pending clinical issues, such
as the definition of the actual link between obstructive sleep apnea and hyperten-
sion; the understanding of why heart attacks and stroke mostly happen during sleep,
especially in early morning; the definition of whether periodic leg movements and
snoring might actually cause a BP increase; and the demonstration of the real ability
of continuous positive air pressure (CPAP) ventilation to reduce BP levels in hyper-
tensive patients with obstructive sleep apnea.
To avoid such inconveniences and to get rid of the related clinical difficulties,
techniques able to provide continuous BP monitoring without need of a cuff infla-
tion would be required.
Currently, a few techniques for continuous BP recordings without cuff inflation
are indeed available, but are rarely employed for difficulties in their applicability.
The first of such techniques, available since many years, is the possibility of intra-
arterial beat-by-beat BP recordings, which have allowed to make the first important
progress in understanding BP variability phenomena and the dynamics of cardio-
vascular control mechanisms [8–10]. This method however is based on invasive
recordings, and even if it was extremely useful in former experimental studies, it
cannot obviously be proposed for a clinical application. Another approach to nonin-
vasive continuous cuffless beat by beat BP recordings is based on the volume-clamp
method described by Jan Penaz and implemented in the Finapres© and Portapres©
devices by Karel Wesseling. This approach has turned out to be useful in research
[11] but has not found an application in clinical practice due to methodological dif-
ficulties in its daily use, to problems in calibration of the BP values provided and to
its high cost. Some other recent approaches have been proposed making use of
mobile health technologies. Indeed, several smartphones apps have been developed
to measure BP. However, validation studies for most of these smartphone-based BP
measurement techniques have not been conducted. To date, mobile health BP moni-
tors have shown poor accuracy compared with oscillometric readings [12–15].
Other approaches are based on assessment of pulse transit time, which can derive
BP levels from pulse transit time (PTT) measurement on the basis of a stretch–strain
relationship model, calibrated with BP from a single initial conventional BP mea-
12 G. Parati
surement [16–18]; or on pulse wave analysis, and are currently under investigations
and development [19]. There is increasing interest towards thee modern approaches,
because the cuffless and thus less interfering nature of the devices may render them
particularly useful in patients who poorly tolerate traditional ambulatory BP moni-
toring and whenever undisturbed assessment of sleep BP is of importance, for
example, in patients with sleep disordered breathing.
There is therefore a strong need for research and development of reliable and
accurate continuous cuffless blood pressure measuring technologies, able to face
the yet unmet need of accurate and at the same time noninterfering repeated BP
measurements.
The present book offers an interesting and up-to-date insight into the progress
made by technology in this stimulating and highly dynamic field.
References
1. Parati G, Ochoa JE, Lombardi C, Bilo G. Assessment and management of blood-pressure vari-
ability. Nat Rev Cardiol. 2013;10(3):143–55.
2. Parati G, Pomidossi G, Albini F, Malaspina D, Mancia G. Relationship of 24-hour blood pres-
sure mean and variability to severity of target-organ damage in hypertension. J Hypertens.
1987;5(1):93–8.
3. Stevens SL, Wood S, Koshiaris C, Law K, Glasziou P, Stevens RJ, et al. Blood pressure vari-
ability and cardiovascular disease: systematic review and meta-analysis. BMJ. 2016;354:i4098.
4. Parati G, Ochoa JE, Bilo G. Blood pressure variability, cardiovascular risk, and risk for renal
disease progression. Curr Hypertens Rep. 2012;14(5):421–31.
5. O’Brien E, Parati G, Stergiou G, Asmar R, Beilin L, Bilo G, et al. European Society of
Hypertension position paper on ambulatory blood pressure monitoring. J Hypertens.
2013;31(9):1731–68.
6. Parati G, Pomidossi G, Casadei R, Malaspina D, Colombo A, Ravogli A, Mancia
G. Ambulatory blood pressure monitoring does not interfere with the haemodynamic effects
of sleep. J Hypertens Suppl. 1985;3:S107–9.
7. Verdecchia P, Angeli F, Borgioni C, Gattobigio R, Reboldi G. Ambulatory blood pres-
sure and cardiovascular outcome in relation to perceived sleep deprivation. Hypertension.
2007;49:777–83.
8. Mancia G, Parati G, Pomidossi G, Casadei R, Di Rienzo M, Zanchetti A. Arterial baroreflexes
and blood pressure and heart rate variabilities in humans. Hypertension. 1986;8(2):147–53.
9. Parati G, Saul JP, Di Rienzo M, Mancia G. Spectral analysis of blood pressure and heart
rate variability in evaluating cardiovascular regulation. A critical appraisal. Hypertension.
1995;25(6):1276–86.
10. Mancia G, Ferrari A, Gregorini L, Parati G, Pomidossi G, Bertinieri G, Grassi G, Di Rienzo
M, Pedotti A, Zanchetti A. Blood pressure and heart rate variabilities in normotensive and
hypertensive human beings. Circ Res. 1983;53:96–104.
11. Parati G, Casadei R, Groppelli A, Di Rienzo M, Mancia G. Comparison of finger and intra-
arterial blood pressure monitoring at rest and during laboratory testing. Hypertension.
1989;13:647–55.
12. Muntner P, et al. Measurement of blood pressure in humans: a Scientific Statement From the
American Heart Association. Hypertension. 2019;73(5):e35–66.
2 Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring 13
13. Kumar N, Khunger M, Gupta A, Garg N. A content analysis of smartphone based applications
for hypertension management. J Am Soc Hypertens. 2015;9:130–6. https://doi.org/10.1016/j.
jash.2014.12.001.
14. Cortez NG, Cohen IG, Kesselheim AS. FDA regulation of mobile health technologies. N Engl
J Med. 2014;371:372–9. https://doi.org/10.1056/NEJMhle1403384.
15. Bruining N, Caiani E, Chronaki C, Guzik P, van der Velde E, Task Force of the e-Cardiology
Working. Acquisition and analysis of cardiovascular signals on smartphones: poten-
tial, pitfalls and perspectives: by the Task Force of the e-Cardiology Working Group of
European Society of Cardiology. Eur J Prev Cardiol. 2014;21(2 Suppl):4–13. https://doi.
org/10.1177/2047487314552604.
16. Schmalgemeier H, Bitter T, Bartsch S, Bullert K, Fischbach T, Eckert S, et al. Pulse transit
time: validation of blood pressure measurement under positive airway pressure ventilation.
Sleep Breath. 2012;16:1105–12.
17. Gesche H, Grosskurth D, Kuchler G, Patzak A. Continuous blood pressure measurement
by using the pulse transit time: comparison to a cuff-based method. Eur J Appl Physiol.
2012;112:309–15.
18. Bilo G, Zorzi C, Oghoa Munera JE, Torlasco C, Giuli V, Parati G. Validation of the

Somnotouch™ NIBP non-invasive continuous blood pressure monitor according to the
European Society of Hypertension International Protocol revision 2010. Blood Press Monit.
2015;20(5):291–4.
19. Baruch MC, Warburton DE, Bredin SS, Cote A, Gerdt DW, Adkins CM. Pulse decomposition
analysis of the digital arterial pulse during hemorrhage simulation. Nonlinear Biomed Phys.
2011;5(1):1.
Chapter 3
A Historical Journey on the Physiology
of Blood Pressure Monitoring
Audrey Adji and Michael F. O’Rourke
Abstract The arterial pulse has been the most basic sign of life for centuries. The
radial pulse palpation has been pictured in the crest of the Royal Academy College
of Physicians since 1628. The history of the arterial pulse entails the discovery of
pulse, blood pressure and/or flow, and their measurements. This chapter begins with
a review the description of the pulse and the related discoveries of pulse and blood
pressure and/or flow since the ancient period until the late 1970s where the concept
of haemodynamics and importance of pressure and flow pulsatility as well as meth-
ods to analyse the pulse in both time and frequency domains gained wider accep-
tance. Human aging is associated with an increase in blood pressure, particularly
systolic and pulse pressures, and this is attributable to the loss of distensibility of the
human aorta of which its function is to cushion pulsation from the ejecting heart.
Stiffening of the major elastic arteries due to aging will cause the speed of the trav-
elling pulse to be higher, and the reflected pulse wave from periphery to occur ear-
lier, therefore will increase the amplitude of pressure. To understand how arterial
haemodynamics is altered by the ageing process and cardiovascular disease is vital
and this involves accurate measurement of central (or aortic) pressure. Finally, the
chapter briefly considers the demand and technology to develop cuffless blood pres-
sure measuring devices. This development could allow a device that can measure
blood pressure accurately, with ease, comfortably and continuously.
Keywords Systolic pressure · Diastolic pressure · Mean pressure · Pulse pressure

· Pressure pulse wave · Sphygmocardiography · Aging · Hypertension · Arterial
stiffness
A. Adji (*)
St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia
Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie
University, Sydney, NSW, Australia
M. F. O’Rourke
St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia
Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia

16 A. Adji and M. F. O’Rourke
From Pulse Wave to Systolic and Diastolic Pressures
Many clinicians and researchers think of blood pressure only in terms of the bra-
chial arterial cuff systolic pressure (SP) and diastolic pressure (DP) derived there-
from. But this is just a convenient way of getting two numbers for the top and
bottom of the pressure wave in the brachial artery.
Though usually now generated from mathematical algorithms in an office device,
the vast amount of blood pressure data obtained in the past was from the ausculta-
tory method where SP represents the first Korotkov sounds heard as the cuff pres-
sure is reduced from above, and DP the cuff pressure at which Korotkov sounds are
no longer heard. These are usually rounded off to the nearest 2 or 5 mmHg, so that
SP measured at 120 mmHg may be from 110 to 130 mmHg, DP from 70 to
90 mmHg, and Pulse Pressure (PP = SP − DP) within an even wider range. For
electronic devices, one still has the issue of respiratory and cardiac variability with
respiration alone which is easily up to 5 mmHg in sinus rhythm with respiration, or
15 mmHg in atrial fibrillation.
We are told by James Mackenzie (Father of Cardiology in the English-speaking
world), to strike an average SP or DP, and by the American Heart Association guide-
lines [1, 2] to do the same—but there is likely to be bias (“experience”) in writing
two numbers between hearing the K sounds, and writing in the medical record or
entering two numbers into a computer. Clinicians (doctors and nurses) are aware of
these practical issues, but non-medical persons (including scientists and patients)
often are not. One needs take time and care to measure blood pressure, and taking
as many recordings as possible.
The heart is a periodic fluid pump; therefore it must generate a pulsatile blood
flow pattern. The pressure generated by intermittent flow is determined by both the
cardiac performance in ejecting blood out of the heart and the properties of the arte-
rial system. The arterial system acts as a conduit to maintain continuous delivery of
oxygenated blood to the tissues, and as a cushion to dampen the pressure and flow
pulsatility and convert this oscillation into a near continuous flow at the capillaries.
With each beat of the heart, a pulse wave of blood is expelled from the heart into the
circulation. These waves can be felt as pulsations in arteries close to the skin sur-
face, such as those in the wrist and the neck.
The Ancient Times
The arterial pulse felt at the wrist has been used by physicians from ancient times to
the modern epoch, for diagnosing illnesses. Throughout centuries, the arterial pulse
was considered the most basic sign of life and was considered to contain an abun-
dance of information on the health or disease of a patient. Graphic recordings of the
pulse waveform at the wrist were first made in the late nineteenth century and were
quickly embraced by physicians as a useful clinical tool, along with the stethoscope.
The pulse waveform was recognised as providing information on elevated arterial
pressure and on effects of arterial stiffening with age. Interest in the pulse lapsed
3 A Historical Journey on the Physiology of Blood Pressure Monitoring 17
with introduction of the cuff sphygmomanometer solely to measure peak and trough
of the arterial pressure wave. Presentation these top and bottom numbers provides a
veneer of science, which the method of measuring cuff blood pressure does not
deserve. Ironically, as shown in modern SphygmoCor reports (Figs. 3.1 and 3.2),
variation in waveforms values (2–3% in figures shown) is far less than variation in
values of SP and DP taken with a cuff sphygmomanometer.
Fig. 3.1 An example of a young male with spurious systolic hypertension. His brachial SP was
taken as 154 mmHg; hence, he would be diagnosed as hypertensive based on this number alone.
His aortic SP, however, was calculated as 123 mmHg, which was normal. The elevated brachial SP
is due to distortion in pulse wave transmission from the heart to upper limb, associated with young
elastic arteries, resulting in sharp narrow systolic peak recorded in the radial pulse
Fig. 3.2 An example of an elderly female with true isolated systolic hypertension. Her brachial SP
was taken as 220 mmHg, and her aortic SP, was calculated as 209 mmHg, appropriately diagnosed
as hypertensives. The elevated brachial SP is due to stiffening of the central elastic arteries, where
the pulse wave transmitted faster from the heart to the periphery, resulting in the reflected pulse
wave superimposed in the ejected pulse wave earlier, thus higher aortic SP
The Mercury Manometer
Stephen Hales performed the first measurement of blood pressure [3]. As described
in his book “Statical Essays: Containing Haemastatics” in 1733, Hales recorded
damped systolic arterial pressure using a tall glass tube in a horse from the height of
a blood column and determined the effects of exsanguination on arterial pressure
(Fig. 3.3). He reported that the flowing blood exerts a pressure on the blood vessel’s
wall while the blood circulation obeys hydrostatic laws [4]. Jean Poiseuille (1799–
1869) advanced Hales’ method of measuring blood pressure by replacing the long
glass tube with a U-tube mercury manometer (called haemodynamometer), pub-
lished in his doctoral dissertation in 1828. Poiseuille calibrated his manometer to
Fig. 3.3 An attempt to measure blood pressure in a horse. An artist’s depiction of Hales’ experi-
ment to measure blood pressure in a horse. From [3]
record arterial pressure level in mmHg, and was able to calculate the force of the
heart beat by observing the oscillation of mercury with his manometer [5]. He also
found that there is no significant difference in mean pressure between central and
tiny peripheral arteries, observed fluctuations in pressure with each heartbeat, and
demonstrated that arterial pressure was maintained in smaller arteries [6]. Further
modification of the Poiseuille manometer was made by Julius Hérisson of Paris in
1831/1833 by sealing the end of the mercury containing glass tube with a thin mem-
brane which rested on the artery, thus enabled him to obtain crude reading of blood
pressure from intact arteries [4]. In his book of 1835, Hérisson stated that he had no
knowledge of Poiseuille’s manometer at the time he developed his, and further
claimed his manometer was not similar to Poiseuille’s. Utilising Hérisson’s device,
the pulse can be studied by its force, regularity and rhythm.
The Kymograph and the Sphygmograph
In 1847, Carl Ludwig (1816–1895) made another modification to Poiseuille’s

manometer by adding a float, thus invented the kymograph (or “wave writer”) to
depict arterial pressure [3, 4], hence set out modern methods of graphic representa-
tion of the pulse [5]. Ludwig’s invention started from his interest on physiological
relation between respiration rate and blood pressure. About the same time as
Ludwig, Karl van Vierordt in 1855 suggested that an indirect, non-invasive tech-
nique to measure pulse might be generated from the counter pressure required to
cause arterial pulsation to cease [3]. Vierordt’s instrument was large, had two heav-
ily weighted levers which converted the unequal rise and fall of the pulse into a
symmetrical curve [5].
Etienne Jules Marey (1830–1904), frustrated by the bulk of Vierordt’s instru-
ment, developed a simpler and more accurate device to illustrate and measure the
arterial pulse [5], by combining a sphygmograph and a kymograph [3] (Fig. 3.4).
Marey’s device was introduced in the Lancet in 1860 with a notice “It may be
doubted whether these instruments, though very ingenious, will ever prove actually
useful in practice” [5]. Marey did the first formal study on arterial pulse waves to
demonstrate the difference in the pulse of elderly and young adults (Fig. 3.5).
Marey’s sphygmograph made apparent the asymmetrical nature of pulse waves and
their variation under different physiological and pathological condition [5]. Marey
wrote a chapter on arterial blood velocity in his popular textbook of medicine “Le
circulation du sang à l’état physiologique et dans des maladies (The Circulation of
Blood in the Physiological State and in Diseases)” in 1881.
While Marey’s sphygmograph was accepted widely for recording and studying
the pulse, it was still too complicated to use routinely in practice, hence the
sphygmograph went through further modifications in England. John Burden-

Sanderson, as described in his “Handbook of the Sphygmograph” in 1867–1868,
together with Francis Anstie, modified Marey’s sphygmograph by fixing the pres-
sure to be 300 g at the centre button and adding a screw to raise and lower the
spring’s free end by a measured amount at the distal end of the device.
Fig. 3.4 Von Basch sphygmomanometer and stand, invented circa 1881. Von Basch’s sphygmo-
graph incorporated Marey’s sphygmograph tambour—see also Fig. 3.5. From [3]
Independent from the development of sphygmography, the first truly accurate

estimation of human blood pressure was done by Faivre in 1856 [3]. He connected
an artery to a mercury manometer during a surgical operation and was able to obtain
direct readings. Faivre found brachial artery pressure to be between 115 and
120 mmHg [3]. Non-invasive blood pressure measurement was pioneered by von
Basch in 1880, where he used an inflatable rubber bag filled with water, then the
edges of the bag were tightly drawn up around the neck of a mercury manometer
bulb to record pressure indirectly from the radial artery [3].
In 1875, William Henry Broadbent delivered a series of lectures on the pulse,
which was published later in 1890 in his book “The Pulse”. Broadbent advocated
careful and systematic evaluation of the pulse, started with frequency and regularity
of beats, then the force or strength, finally the character (i.e. rise, duration and fall)
of the pulse [7]. He tested his clinical skills against pulse waveforms recorded by
sphygmography, and encouraged his students and examinees to do the same. (He
was chief censor (examiner) for the Royal College of Physicians of London for
many years.)
Frederick Mahomed (1849–1884), one of Broadbent’s students, made further
modification to Marey’s sphygmograph and published his work in 1872. His modi-
fication included placing over the centre of the spring a vertical eccentric turned by
a graduated screw, as well as replaced the side wires with ivory bars [8]. Mahomed
recorded the pulse graphically, thus permitting the pulse to be studied as a wave [9].
Fig. 3.5 Marey sphygmograph and the recorded pulse. Marey’s sphygmograph and his study on
arterial pulse, demonstrating the difference in the pulse of elderly and young adults. From this
sphygmogram, it is apparent the asymmetrical nature of pulse wave and their variation under dif-
ferent physiological and pathological condition. From [3]
The use of sphygmogram to detect elevated pressure was further emphasised by

William Osler, where senile or physiological arteriosclerosis and high blood pres-
sure were diagnosed on the basis of pulse wave contour—from high and prolonged
pressure in late systole, “the tidal wave is prolonged and too much sustained”, and
“the pulse (of high tension) is slow in its ascent, enduring, subsides slowly, and in
the interval between the beats, the vessel remains full and firm”. He stressed that the
(radial) “pulse of high tension” is similar to the pulse in arteriosclerosis but that a
test could be used to distinguish between the two (which are often combined) “(if)
when the radial (artery) is compressed with the index finger, the artery can be felt
beyond the point of compressions, its walls are sclerosed” [10].
Robert Ellis Dudgeon (1820–1904) invented a new compact sphygmograph,
which he exhibited in 1881 and published in 1882 [8]. Dudgeon made significant
improvements to Marey’s and Mahomed’s sphygmograph [8]. Dudgeon’s
sphygmograph was later used by James Mackenzie, founder of Cardiology as a
specialist discipline in the English-speaking world.
Sir James Mackenzie (1853–1925) is closely associated with non-invasive inves-
tigation of normal and pathological cardiovascular phenomena [8]. He was one of
the first observers to be interested in the regularity or otherwise of the pulse, also
laid the foundation of modern concept of heart failure, which was being studied by
Frank in Germany and by Starling in England at this time [8].
While the introduction of the sphygmomanometer into clinical medicine in the
early 1900s was accepted by some practitioners as a valuable aid for diagnosis, the
“British Medical Journal” in 1905 stated in relation to use of clinical tools that “we
pauperize our senses and weaken clinical acuity” [3].
The Sphygmomanometer
The next development in blood pressure measurement was the introduction of blood
pressure measurement by palpation. Scipione Riva-Rocci in 1896 published two
papers titled “Un Nuovo Sfigmomanometro” in Gazetta Medica di Torino. His
sphygmomanometer was based on established principles of Vierordt and further
improvement by Marey and von Basch. Advantages of the Riva-Rocci device were
the ease of its application, rapid action, precision and harmlessness [3, 11]. The
Riva-Rocci technique involved compression of the arm around its whole circumfer-
ence by inflating a rubber bag with air. Using the palpation of the radial artery,
systolic blood pressure can be determined [3]. One flaw of this technique was that
the cuff was narrow; thus the reading of pressure can be inaccurate [3]. Later, von
Recklinghausen in 1901 fixed this defect by substituting the armband by a wider
one [3].
Further advancement in pressure measurement was made by Nikolai Korotkoff,
where in 1905, he reported that by placing a stethoscope over the brachial artery
below the air-pressure cuff, tapping sounds—the sounds of the column of blood—
could be heard as the cuff was deflated and blood flowed through the artery [11].
Korotkoff concluded that a perfectly constricted artery under normal conditions
does not emit any sounds, and he introduced the auscultatory method to measure
maximum and minimum level of blood pressure [11]. This auscultatory method was
described in the Imperial Military Medical Academy in St Petersburg in 1905.
Popularity of the sphygmomanometric cuff arose from its use in the Life Insurance
industry which utilised its value of identifying apparently normal people who had
markedly reduced age expectancy. In 1906, Dr. J. W. Fisher, medical director of the
Northwestern Mutual Life Insurance Co., initiated the inclusion of a blood pressure
reading in every routine examination by their examiners [12] (Fig. 3.6). By 1918
most insurance companies were measuring cuff blood pressures. Reliance on blood
pressure measured from the brachial artery began since then, and cuff-based sphyg-
momanometry is widely used to date because it is easy to perform, despite merely
describing the pressure as its peak (systolic) and nadir (diastolic) value.
Clinical Practice Today
To a great extent pressure metrics have pervaded medicine, since virtually every
diagnostic clinic visit is preceded by the ritual of taking the blood pressure. The
conventional office cuff pressure measurement remains the “gold standard” for
screening, diagnosis and assessment of arterial pressure level. Unfortunately, mea-
surement of cuff pressure sometimes is performed “in the sloppiest manner” [13,
14]. These are strong words, but were made by respected, and usually reserved,
clinical scientists. Cuff blood pressure is an ever-changing hemodynamic index; it
varies with time, body position, heartbeat and many more physiological factors.
Preoccupation with the sphygmomanometric “numbers” persists in the diagnosis
and management of elevated blood pressure. At the turn of this (twenty-first) cen-
tury, attention was transferred from diastolic to systolic pressure as the major index
of risk in adult hypertension [15, 16] and to the importance of large arteries both in
determining SP and PP and as the target of damage when pressure is high. This has
been aided by new methods for determining and interpreting pulsatile arterial pres-
sure, flow and diameter non-invasively [10].
Hypertension as a Risk Factor in Cardiovascular Disease
Blood pressure will increase with age, even in apparently healthy individuals; it is
acknowledged as a feature of human aging [17, 18] and has been demonstrated
repeatedly including by the longitudinal study of Framingham [17]. In the past, risk
evaluation was based on brachial DP, led by a misguided therapeutic proposition of
James Orr, the literary executor of James MacKenzie following his death [19].
During 1910s, the use of sphygmomanometers gained popularity, especially within
life insurance companies, who were able to relate blood pressure level to mortality
outcomes [20] (Fig. 3.6). It was not until the 1960s when the investigators from the
Framingham Heart Study began to recognise high brachial SP as an important car-
diovascular risk factor [21]. This recognition was then followed by two major
reports from the Veteran Affairs Cooperative Study on Antihypertensive Agents that
lowering blood pressure resulted in significant reduction in major adverse events,
including stroke and heart failure [22, 23]. The Framingham Heart Study [24] went
Fig. 3.6 Inclusion of blood pressure measurement for insurance examination as published in
JAMA 1914 by Dr. Fisher. From [12]
on to conclude the importance of systolic hypertension in the middle-aged and

elderly subjects, as opposed to DP. Many other clinical scientists continue these
epidemiological studies in association with the US National Heart, Lung and Blood
Institute and American Heart Association [25].
The significance of SP was finally confirmed with the publication of the Systolic
Hypertension in the Elderly Project (SHEP) [26]. Subsequently, the (brachial) SP
and PP emerged as the (modifiable) key risk factor in cardiovascular disease follow-
ing longitudinal studies of Framingham Heart Study [16, 27]. They found that coro-
nary heart disease risk is more related to pulsatile stress due to large artery stiffness
during systole, thus emphasised the consequence of isolated systolic hypertension
and the importance of SP and/or PP level in adults. Franklin et al. went further and
concluded that the two measures of pressure are superior to any single pressure
component in predicting cardiovascular risk because these pressures assessed both
arterial stiffness and resistance [28]. The latest publication from the Framingham
Heart Study cohort again emphasised the importance of pulse pressure as the pulsa-
tile load in increasing brachial blood pressure with age, at least for those over the
age of 60 [17].
Increased arterial stiffness due to age is the main determinant of high blood pres-
sure. Elevated blood pressure increases the load on the heart and stress on the arter-
ies, hence assessment of arterial function should incorporate monitoring of blood
pressure levels to gauge the ill-effect of arterial stiffness. It is now widely accepted
that stiffening of the major (central) arteries with ageing affects the propagation of
pressure pulse along the arterial tree, increases blood pressure, changes the ejection
pattern of the heart, and eventually causes cardiac failure. Progressive stiffening of
the large elastic arteries is attributable to loss of their distensibility with increasing
age, characterised as faster travelling of the pulse generated by the heart (measur-
able through “aortic” pulse wave velocity) and higher central systolic pressure.
Stiffening of the central arteries has been recognised as a cause of adverse cardio-
vascular outcomes, including a higher amplitude of central PP and an increase in the
transmission of pulsatile flow into the microcirculation [2]—published as a Scientific
Statement by the American Heart Association, and by the European Societies of
Cardiology and Hypertension [29]. Most of the non-invasive studies published on
central aortic pressure and related indices have shown clear superiority of these in
determining clinical end points. Central aortic SP and aortic PP are increasingly
important because these are not affected by amplification of pressure between aorta
and upper limb (Figs. 3.1 and 3.2). They are more relevant to left ventricular load
than SP and PP measured in the upper limb. SP varies among arterial segments due
to the phenomenon of arterial pulse amplification [30]. Therefore, the common
practice of using conventional brachial cuff sphygmomanometric measurements as
pressure load imposed on the heart and on central circulation (including coronary
and cerebrovascular arteries) may be erroneous [30]. Indices derived from the aortic
pressure wave, including pressure pulse amplification to the upper limb, its augmen-
tation from peripheral wave reflection, its integral during the periods of systole and
diastole would provide more information on vascular load, cardiac function, and
ventricular–vascular interaction.
Management of elevated high blood pressure has evolved since cuff blood pres-
sure was introduced clinically 120 years ago. The conventional office blood pres-
sure recording remains the “gold standard” for screening, diagnosis and management
of elevated blood pressure [29, 31]. At present, out-of-office blood pressure mea-
surements are increasingly recommended to confirm the diagnosis of elevated blood
pressure and to gauge the effectiveness of therapy.
The Need for Cuffless Blood Pressure Measuring Devices
There is now a demand and a sufficiently advanced technology to develop cuffless

blood pressure measuring devices. This development could allow for a device that
can measure accurately, with ease, comfort and continuity. Current devices use
modified tonometry or volume clamp/plethysmography of pulse wave transit time/
velocity to measure pressure.
A cuffless blood pressure would utilise optical sensors, similar to fitness trackers.
These wrist and finger sensors use beat-to-beat variability to compute systolic and
diastolic readings by mathematical modelling. The benefits of these sensors include
the ability to monitor blood pressure continuously and avoid sleep-disrupting cuff
inflations when recordings are performed in the evening. These devices also will
particularly be useful in the elderly or person with limited mobility. However, the
accuracy of these devices is still questioned. Despite validation, measurements can
vary as much as 20 mmHg when compared to blood pressure measured using stan-
dard brachial cuff, especially those with elevated blood pressure [32, 33].
Mobile or mHealth is increasingly suggested as a solution to poor blood pressure
awareness and control. At present, most adults now own smartphones equipped with
cameras and light/motion sensors. Concerns remain with this type of technology,
where data safety is uncertain, and the information provided may be misleading or
inaccurate [34].
With improvements in sensor technology, we have the ability to acquire high-
fidelity pulse waveform signals from extremities to obtain various physiological
parameters. These have led to development of several physiological parameters that
correlate with blood pressure [35]. Pulse transit time (the amount of time from left
ventricular contraction to pulse waveform acquisition at the distal site) is being
looked at as an independent variable to predict blood pressure [36]. Central, not
brachial, pulse wave velocity has long been known to be a predictor of vascular
stiffness, and better than the brachial blood pressure [37]. Pulse tonometry (counter
pressure to measure arterial distension) and vascular unloading (measuring counter
pressure required to maintain constant finger blood volume) are two other tech-
niques being attempted to get continuous non-invasive cuffless measurements. The
challenge faced by most of these technologies has been calibration of local pressure
measurement site. With the present technology, it is possible to implement these
pulse-based systems that accurately predict the trends of blood pressure instead of
measuring pressure values itself. Considerable fluctuations in this trend can be used
as a warning signal for users to monitor their blood pressure and continuous moni-
toring of this variation can be helpful in clinical management of certain cases.
With the advancing technology, trends will continue in development of cuffless
blood pressure monitoring. Devices that can monitor blood pressure continuously
and provide additional and comprehensive hemodynamic information will allow us
to establish better correlations with clinical outcomes and open up new possibilities
for cardiovascular and renal disease risk prediction. The importance of this initiative
is well worth investigating.
References
1. Pickering TG, Hall JE, Appel LJ, Falkner BE, Graves J, Hill MN, Jones DW, Kurtz T, Sheps
SG, Roccella EJ. Subcommittee of Professional and Public Education of the American Heart
Association Council on High Blood Pressure Research. Recommendations for blood pres-
sure measurement in humans and experimental animals: part 1: blood pressure measurement
in humans: a statement for professionals from the Subcommittee of Professional and Public
Education of the American Heart Association Council on High Blood Pressure Research.
Hypertension. 2005;45:142–61.
2. Townsend RR, Wilkinson IB, Schiffrin EL, Avolio AP, Chirinos JA, Cockcroft JR, Heffernan
KS, Lakatta EG, McEniery CM, Mitchell GF, Najjar SS, Nichols WW, Urbina EM, Weber
T. American Heart Association Council on Hypertension. Recommendations for improv-
ing and standardizing vascular research on arterial stiffness: a Scientific Statement From the
American Heart Association. Hypertension. 2015;66:698–722.
3. Booth J. A short history of blood pressure measurement. Proc R Soc Med. 1977;70:793–9.
4. Wakerlin GE. From Bright toward light: the story of hypertension research. Circ Res.
1962;11:131–6.
5. Lawrence C. Physiological apparatus in the Wellcome museum: 1. The Marey sphygmograph.
Med Hist. 1978;22:196–200.
6. O’Rourke MF. Frederick Akbar Mahomed. Hypertension. 1992;19:212–7.
7. Fye WB. William Henry Broadbent. Clin Cardiol. 1990;13:62–4.
8. Lawrence C. Physiological apparatus in the Wellcome museum: 2. The Dudgeon sphygmo-
graph and its descendants. Med Hist. 1979;23:96–101.
9. Ghasemzadeh N, Zafari AM. A brief journey into the history of the arterial pulse. Cardiol Res
Pract. 2011;2011:164832.
10. Nichols WW, O’Rourke MF, Vlachopoulos C. McDonald’s blood flow in arteries. 6th ed.
London: Arnold Hodder; 2011.
11. Lewis WH Jr. The evolution of clinical sphygmomanometry. Bull N Y Acad Med.

1941;17:871–81.
12. Fisher JW. The diagnostic value of the sphygmomanometer in examinations for life insurance.
JAMA. 1914;58:1752–4.
13. Messerli FH, White WB, Staessen JA. If only cardiologists did properly measure blood
pressure. Blood pressure recordings in daily practice and clinical trials. J Am Coll Cardiol.
2002;40:2201–3.
14. Kaplan NM. Commentary on the sixth report of the Joint National Committee (JNC-6). Am
J Hypertens. 1998;11:134–6.
15. Franklin SS, Gustin W, Wong ND, Larson MG, Weber MA, Kannel WB, Levy D. Hemodynamic
patterns of age-related changes in blood pressure: the Framingham Heart Study. Circulation.
1997;96:308–15.
16. Staessen JA, Gasowski J, Wang JG, Thijs L, Den Hond E, Boissel JP, Coope J, Ekbom T,
Gueyffier F, Liu L, Kerlikowske K, Pocock S, Fagard RH. Risks of untreated and treated
isolated systolic hypertension in the elderly: meta-analysis of outcome trials. Lancet.

2000;355:865–72. Erratum in: Lancet. 2001;357:724.
17. Cheng S, Xanthakis V, Sullivan LM, Vasan RS. Blood pressure tracking over the adult life
course: patterns and correlates in the Framingham Heart Study. Hypertension. 2012;60:1393–9.
18. Gurven M, Blackwell AD, Rodríguez DE, Stieglitz J, Kaplan H. Does blood pressure inevi-
tably rise with age?: longitudinal evidence among Forager-Horticulturalists. Hypertension.
2012;60:25–33.
19. O’Rourke MF. From theory into practice: arterial haemodynamics in clinical hypertension.
J Hypertens. 2002;20:1901–15.
20. Postel-Vinay P. A century of arterial hypertension 1896–1996. New Jersey: Wiley; 1996.
21. Kannel WB, Dawber TR, Kagan A, Revotskie N, Stokes IJ. Factors of risk in the development
of coronary heart disease—six-year follow-up experience: the Framingham Study. Ann Intern
Med. 1961;55:33–50.
22. Veterans Affairs Cooperative Study Group. Effects of treatment on morbidity in hypertension:
results in patients with diastolic blood pressures averaging 115 through 129 mmHg. JAMA.
1967;202:1028–34.
23. Veterans Affairs Cooperative Study Group. Effects morbidity of treatment on in hypertension:
II. Results in patients with diastolic blood pressure averaging 90 through 114 mmHg. JAMA.
1970;213:1143–52.
24. Kannel WB, Gordon T, Schwartz MJ. Systolic versus diastolic blood pressure and risk of coro-
nary heart disease: the Framingham study. Am J Cardiol. 1971;27:335–46.
25. Greenland P, Peterson ED, Gaziano JM. Progress against cardiovascular disease: putting the
pieces together. JAMA. 2014;312:1979–80.
26. Systolic Hypertension in Elderly Program Trial. Prevention of stroke by antihypertensive drug
treatment in older persons with isolated systolic hypertension: final results of the systolic
hypertension in the elderly program (SHEP). JAMA. 1991;265:3255–64.
27. Franklin SS, Khan SA, Wong ND, Larson MG, Levy D. Is pulse pressure useful in predicting
risk for coronary heart disease?: the Framingham Heart Study. Circulation. 1999;100:354–60.
28. Franklin SS, Lopez VA, Wong ND, Mitchell GF, Larson MG, Vasan RS, Levy D. Single versus
combined blood pressure components and risk for cardiovascular disease: the Framingham
Heart Study. Circulation. 2009;119:243–50.
29. Williams B, Mancia G, Spiering W, Agabiti Rosei E, Azizi M, Burnier M, Clement DL, Coca
A, de Simone G, Dominiczak A, Kahan T, Mahfoud F, Redon J, Ruilope L, Zanchetti A, Kerins
M, Kjeldsen SE, Kreutz R, Laurent S, Lip GYH, McManus R, Narkiewicz K, Ruschitzka F,
Schmieder RE, Shlyakhto E, Tsioufis C, Aboyans V, Desormais I; ESC Scientific Document
Group. 2018 ESC/ESH Guidelines for the management of arterial hypertension. Eur Heart
J 2018;39:3021–3104.
30. Roman MJ, Devereux RB. Association of central and peripheral blood pressures with interme-
diate cardiovascular phenotypes. Hypertension. 2014;63:1148–53.
31. Whelton PK, Carey RM, Aronow WS, Casey DE Jr, Collins KJ, Dennison Himmelfarb C,
DePalma SM, Gidding S, Jamerson KA, Jones DW, MacLaughlin EJ, Muntner P, Ovbiagele B,
Smith SC Jr, Spencer CC, Stafford RS, Taler SJ, Thomas RJ, Williams KA Sr, Williamson JD,
Wright JT Jr. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH /ASPC/NMA/PCNA
Guideline for the prevention, detection, evaluation, and management of high blood pressure
in adults: a Report of the American College of Cardiology/American Heart Association Task
Force on Clinical Practice Guidelines. Hypertension. 2018;71:e13–e115.
32. Guggiari C, Bula C, Iglesias K, Waeber B. Measurement with an automated oscillometric wrist
device with position sensor leads to lower values than measurements obtained with an auto-
mated oscillometric arm device from the same manufacturer in elderly persons. Blood Press
Monit. 2014;19:32–7.
33. Pereira T, Correia C, Cardoso J. Novel methods for pulse wave velocity measurement. J Med
Biol Eng. 2015;35:555–65.
34. Goldberg DM, Levy PD. New approaches to evaluating and monitoring blood pressure. Curr
Hypertens Rep. 2016;18:49.
35. Pandit JA, Batlle D. Snapshot hemodynamics and clinical outcomes in hypertension: precision
in the measurements is key. Hypertension. 2016;67:270–1.
2015;62:1879–901.
37. Sharma M, Barbosa K, Ho V, Griggs D, Ghirmai T, Krishnan SK, Hsiai TK, Chiao JC, Cao
H. Cuff-less and continuous blood pressure monitoring: a methodological review. Technologies.
2017;5:21.
Chapter 4
The Definition and Architecture of Cuffless
Blood Pressure Monitors
Josep Solà
Abstract This chapter introduces a systematic framework to navigate through the

universe of cuffless blood pressure technologies. The chapter is organized in three
parts. Initially, a glossary is provided to anchor the discussion and formalize the
terms involved in cuffless blood pressure monitoring. Then, a classification of exist-
ing blood pressure monitoring technologies is proposed together with a formal defi-
nition of a cuffless blood pressure monitor. Finally, an generic system architecture
is presented organizing all the elements that any cuffless blood pressure monitor
shall contain.
Keywords Basic glossary of cuffless blood pressure monitoring · Definition of a

cuffless blood pressure monitor · Architecture of a cuffless blood pressure monitor
· Transducer layer · Processing layer · Initialization layer · Pulsatility sensor ·
Pulsatility-based algorithm · Initialization algorithm · Calibration parameters
Scope
After covering the clinical and physiological aspects of cuffless blood pressure tech-
nologies, this chapter introduces the generic architecture of a cuffless blood pres-
sure monitor. This architecture is intended to establish a systematic framework that
will help the reader to navigate through the jungle of cuffless blood pressure tech-
nologies, and that will support the comparison of legacy and novel approaches.
The chapter is organized as follows: initially a basic glossary to understand cuf-
fless blood pressure technologies is provided. Then, a definition of cuffless blood
pressure monitors is proposed, followed by the generic architecture of the elements
that any cuffless blood pressure monitor shall present. Finally, illustrative examples
of implementation of cuffless monitors are briefly enumerated.
J. Solà (*)
Aktiia SA, Neuchâtel, Switzerland
e-mail: josep@aktiia.com

32 J. Solà
Terms and Definitions
Over the last several decades, the community of clinicians, scientists, engineers, man-
ufacturers and users of cuffless blood pressure devices has incrementally introduced a
list of terms in the field that describe the different units and functions of a cuffless
blood pressure monitor. The goal of this section is to summarize the most relevant
items in a compiled list of physiological terms (Table 4.1) and technological terms
(Table 4.2) adopted in the field of blood pressure monitoring. The sorting of terms is
intentionally not alphabetical, with the goal of incrementally educating the reader.
Table 4.1 Physiological terms adopted in the field of cuffless blood pressure monitoring
Term Definition
Arterial-Blood The pressure within an artery, commonly expressed in mmHg or kPa.
Pressure (BP) Because the blood pressure continuously evolves during a cardiac cycle it
is typically summarized in two values: the maximum value (Systolic),
and the minimum value (Diastolic)
Systolic Blood The maximum blood pressure value during a cardiac cycle
Pressure (SBP)
Diastolic Blood The minimum blood pressure value during a cardiac cycle
Pressure (DBP)
Pulse Pressure (PP) The difference between systolic and diastolic blood pressure values
during a cardiac cycle
Central artery An artery close to the heart, typically the ascending aorta and the
descending aorta
Peripheral artery An artery further (downstream) away from the heart. The brachial artery
is a peripheral artery
Vasomotion The change of tone of an artery created by the contraction/relaxation of
its smooth muscles
Muscular artery An artery whose wall contains layers of smooth muscles, and that is
affected by vasomotion
Elastic artery An artery whose wall contains few smooth muscles and that is not
severely affected by vasomotion. A central artery is an elastic artery
Perfusion The delivery of (arterial) blood into an organ or tissue
Perfusion Index (PI) A quantified measurement of pulsatility, typically used in pulse oximetry
(Arterial) Pressure A pressure impulse generated at the heart during the opening of the aortic
wave valve that propagates along the entire arterial tree dilating the walls of each
artery. An arterial pressure wave does not indicate the movement of blood
Forward pressure An arterial pressure wave traveling from the heart towards the periphery
wave
Backward pressure An arterial pressure wave traveling from the periphery towards the heart.
wave A backward pressure wave is generated when a forward pressure wave is
reflected at a given reflection site
(Arterial) Pressure The time series of the superposition of the forward and backward pressure
waveform waves at a given location of the arterial tree during one cardiac cycle
Mean Arterial Blood The average of the arterial blood pressure waveform during a cardiac
Pressure (MAP) cycle
(continued)
4 The Definition and Architecture of Cuffless Blood Pressure Monitors 33
Table 4.1 (continued)
Term Definition
Pulsatility The ability of an artery (or arterial bed) to pulsate, that is, to change its
characteristics (e.g., diameter) at the arrival of an arterial pressure wave
Pulsatility waveform The time series of the pulsatility of an artery (or arterial bed) during one
cardiac cycle
Blood velocity The velocity at which blood circulates within an artery. While blood
velocity in central arteries is in the range of 0.2–0.4 m/s, in peripheral
arteries it is decreased
Pulse Wave Velocity The velocity at which an arterial pressure wave propagates along the
(PWV) walls of the arterial tree. While PWV in a central artery is in the range of
5–15 m/s, in peripheral arteries it is higher
Central pulse wave The pulse wave velocity of a central artery
velocity
Peripheral pulse The pulse wave velocity of a peripheral artery or a peripheral segment of
wave velocity the arterial tree
Local pulse wave The pulse wave velocity measured punctually on a given arterial site, or
velocity along a homogenous segment of the arterial tree (along one single artery)
Regional pulse wave The pulse wave velocity measured along a heterogeneous segment of the
velocity arterial tree (along a concatenation of arteries)
Pulse Transit Time The time that an arterial pressure wave requires to propagate along the
(PTT) walls of a given segment of the arterial tree
Pulse Arrival Time The time at which an arterial pressure wave arrives at a certain point of
(PAT) the arterial tree. Typically, PAT measurements consider as initial time
reference the R-Wave of an electro-cardiogram
Pre-ejection period The time lapse between the R-wave of an electro-cardiogram and the
(PEP) opening of the aortic valve
Table 4.2 Technological terms adopted in the field of cuffless blood pressure monitoring
Term Definition
Blood Pressure A device to determine blood pressure
Monitor (BPM)
Invasive blood A blood pressure monitor that requires the puncture of an artery in order
pressure monitor to directly measure intra-arterial pressure
Noninvasive blood A blood pressure monitor that does not require the puncture of an artery
pressure monitor
Arterial-occlusion A noninvasive blood pressure monitor that requires the partial or
blood pressure complete occlusion of an artery
monitor
Cuff-based blood An arterial-occlusion blood pressure monitor that requires the inflation
pressure monitor of a cuff around a limb
Full occlusion blood An arterial-occlusion blood pressure monitor whose pressure exceeds
pressure monitor the systolic blood pressure leading to complete arterial occlusion
Semi occlusion An arterial-occlusion blood pressure monitor whose pressure does not
exceed the systolic blood pressure
No arterial-occlusion A blood pressure monitor that does not require the occlusion of an
blood pressure artery
monitor
(continued)
34 J. Solà
Term Definition
Cuffless blood A no arterial-occlusion blood pressure monitor that does not require the
pressure monitor inflation of a cuff around a limb
Continuous blood A blood pressure monitor that determines the entire pressure waveform
pressure monitor
Beat-to-beat blood A blood pressure monitor that determines one systolic and a diastolic
pressure monitor value at each cardiac cycle
Intermittent blood A blood pressure monitor that determines one systolic and one diastolic
pressure monitor value over a series of cardiac cycles
Manual blood A blood pressure monitor that requires the intervention of a skilled
pressure monitor operator in order to perform a measurement. The most common manual
blood pressure technique used in practice is auscultation
Automated blood A blood pressure monitor that does not require the intervention of a
pressure monitor skilled operator in order to perform a measurement
Arterial line (A-line) An invasive blood pressure monitoring method that requires insertion of
a catheter into an artery via an arterial puncture
Auscultation A noninvasive, cuff-based, intermittent manual determination of blood
pressure that requires the expert operation of a cuff and a stethoscope.
During the manual deflation of the cuff an operator identifies the presence
Korotkoff sounds using the stethoscope put on top of the measured artery
Oscillometry A noninvasive, cuff-based, intermittent automated determination of
blood pressure that uses an electronic-driven pneumatic cuff and
requires no particular intervention from an operator. Oscillometric
devices are typically presented in the form of upper-arm or wrist cuffs
Office Blood Pressure A manual or automated blood pressure measurement performed at a
Measurement physician’s office or in clinics. Auscultation and oscillometric monitors
(OBPM) are typically used
Home Blood Pressure An automated blood pressure measurement performed by a patient at
Measurement home under controlled conditions. The measurement is typically
(HBPM) triggered by the patient at rest. Oscillometric monitors are typically used
Ambulatory Blood An automated blood pressure measurement performed on a patient
Pressure under uncontrolled conditions. The measurement is typically
Measurement automatically triggered at 15–30 min intervals during the day and the
(ABPM) night without any intervention from the patient. Oscillometric monitors
are typically used
Volume clamp A noninvasive, cuff-based, continuous automated determination of
blood pressure that relies on the vascular unloading technique.
Volume-clamp monitors perform an optical measurement of arterial
volume at a phalanx while continuously applying a counter pressure by
means of a servo-controlled finger-cuff
Tonometry A noninvasive continuous automated determination of blood pressure that
consists on the application of a force to a superficial artery in order flatten
its walls and to measure its pulsatility. Tonometric monitors rely on
different sensing technologies to quantify force and arterial displacement
Pulse wave analysis The analysis of a pressure waveform in order to extract cardiovascular
(PWA) information. PWA can also be referred to as pulse wave decomposition
(PWD) or pulse contour analysis (PCA)
(continued)
Term Definition
Initialization The process of determining subject- or condition-specific parameters
that are required to determine a blood pressure value in a blood pressure
monitor. Initialization is typically required by cuffless blood pressure
monitors, and involves the use of a cuff-based monitor
Calibration function/ The set of subject- or condition-specific parameters that are required to
parameters initialize the blood pressure monitor
Reinitialization The process of updating the calibration function/parameters of a blood
pressure monitor
Defining the Perimeter of Cuffless Blood Pressure Monitors
While reading the existing literature and commercial claims around cuffless moni-
toring, it can be unclear what a cuffless blood pressure monitor actually is. The goal
of this section is to provide a formal definition of a cuffless blood pressure monitor,
and to determine its perimeter within the large catalog of existing blood pressure
technologies.
The arena of the most relevant blood pressure technologies in use today is cate-
gorized by the illustration in Fig. 4.1.
The proposed classification starts, at its root, by splitting blood pressure moni-
tors between invasive and noninvasive monitors. Invasive Blood Pressure monitors
require the puncture of an artery (by means of an arterial catheter) and are out of
scope of this book. The reader can find more information in [1].
Focusing on the family of noninvasive blood pressure monitors (monitors that do
not require an arterial puncture), the proposed classification further splits between
monitors that measure a physical pressure on the body and monitors that estimate
blood pressure without applying any pressure to the body. Arterial-occlusion Blood
Pressure monitors typically require the use of a cuff (or an actuator) that applies a
certain pressure to the body in order to perform a measurement. These monitors can
further be differentiated between those that create a prolonged supra-systolic occlu-
sion of an artery (including manual auscultation techniques, and automated oscil-
lometric techniques) and those that create only partial or an intermittent occlusion
of an artery (including volume-clamp techniques, and tonometric techniques).
These monitors are also out of scope of the current book. The reader can find more
information in [1].
The perimeter of cuffless blood pressure monitors commences at the second
branch of noninvasive monitors: a cuffless blood pressure monitor is defined as a
device or a technology that noninvasively determines the blood pressure of an
individual without creating any arterial occlusion.
Furthermore, within the vast family of cuffless monitors one can differentiate
between two categories:
36
Invasive A-line
Auscultation
Full-occlusion
Arterial Oscillometry
Blood Pressure
monitors occlusion
Volume-clamp
Semi-
occlusion
Tonometry
Non-Invasive
Pulse Arrival Time
Regional
Pulse Transit Time
Pulse Wave
Velocity
Central Pulse Transit Time
Local
No arterial Peripheral Pulse Transit Time
occlusion
Feature-based
Pulse Wave
Analysis
Machine Learning
Cuffless Blood Pressure monitors
Fig. 4.1 Visual classification of the most relevant blood pressure technologies, with a proposed delimitation of the perimeter of Cuffless Blood Pressure
monitors
J. Solà
• Monitors based on the principle of pulse wave velocity (PWV), measuring local
or regional PWV values by means of at least two pulsatility sensors and/or addi-
tional cardiovascular sensors. This family includes pulse arrival time (PAT) tech-
niques, covered by Chap. 5 of this book, and pulse transit time (PTT) techniques,
covered by Chap. 6 of this book.
• Monitors based on the principle of pulse wave analysis (PWA), performing the
analysis of pressure waveform by means of one single pulsatility sensor. This
family includes Feature-based techniques, covered by Chap. 7 of this book, and
machine learning-based techniques, covered by Chap. 8 of this book.
The Architecture of a Cuffless Blood Pressure Monitor
Following the formal identification of what attributes define a cuffless blood pres-
sure monitor, a generic architecture of the components of a cuffless solution is fur-
ther proposed.
The main characteristic of a cuffless blood pressure monitor is that it creates no
arterial occlusion during the process of determining a blood pressure value. In practi-
cal terms, this means that a cuffless monitor does not apply any force to the body. The
monitor simply extracts information from the user’s skin surface that is further pro-
cessed to determine a blood pressure value. As explained later, most cuffless moni-
tors happen to rely on information related to arterial pulsatility (the change of
diameter of an artery or arterial bed occurring during a cardiac cycle). In some cases,
a cuffless monitor might still apply some pressure to the skin surface in order to
obtain reliable readings. However, this exerted pressure is related to the need of accu-
rately measuring pulsatility signals, rather than related to the need of creating a phys-
ical interaction between a sensor and the arterial bed underlying the skin surface. The
lack of applied force to the body leverages advantages and creates limitations. On
one side, the obvious advantages of a cuffless monitor compared to an arterial-occlu-
sion monitor are an increase in user comfort and the possibility of performing long-
term continuous blood pressure monitoring. On the other side, the main disadvantage
is that because of the lack of mechanical interaction between the monitor and the
underlying arteries, the implemented sensor can only retrieve information related to
the pulsatility of an artery, but cannot collect its actual pressure values. In practice,
while arterial-occlusion monitors rely on pressure transducers to determine a physi-
cal blood pressure quantity, cuffless monitors rely on pulsatility sensors that have no
access to physical pressure quantities. Consequently, while arterial-occlusion moni-
tors are capable of outputting absolute blood pressure readings that are expressed in
pressure units (kPa or mmHg), cuffless monitors are limited to output blood pressure
readings that are expressed in non-pressure units (see Table 4.3).
In order to prepare the introduction of a generic architecture for cuffless blood
pressure monitors, Figs. 4.2 and 4.3 start by illustrating the typical architecture of
invasive and arterial-occlusion monitors. The reader will note that both approaches
rely on an actual pressure transducer in order extract a pressure waveform. For the
38 J. Solà
Table 4.3 Sensing principles and output units of different blood pressure monitoring technologies
Blood pressure monitoring Sensing
technology principle Output
Invasive Pressure Blood pressure in mmHg
transducer
Noninvasive with arterial Pressure Blood pressure in mmHg
occlusion transducer
Noninvasive without arterial Pulsatility Blood pressure in non-pressure units, expressed
occlusion (cuffless) sensor in mmHg after an initialization procedure
Hydraulic
pressure in Pressure
catheter Pressure waveform Peak detection Systolic BP
transducer algorithm Diastolic BP
Body Transducer layer Processing layer Output
Invasive blood pressure monitor
Fig. 4.2 Typical architecture of an invasive blood pressure monitor that relies on the puncture of
an artery to measure a hydraulic pressure in a catheter. A pressure transducer generates than a pres-
sure waveform that is further processed by an algorithm to generate an absolute blood pressure
determination expressed in mmHg
Pressure
Pneumatic Pressure command
pressure actuator
in cuff Pressure-
Systolic BP
based
Pressure Diastolic BP
waveform algorithm
Pressure
transducer
Body Transducer layer Processing layer Output
Non-invasive arterial occlusion blood pressure monitor
Fig. 4.3 Typical architecture of an arterial occlusion blood pressure monitor that relies on a pneu-
matic cuff (or another actuator) to generate an occlusion to the body and to measure a pressure. A
pressure transducer generates then a pressure waveform that is further processed by an algorithm
to generate an absolute blood pressure determination expressed in mmHg
invasive monitors, the pressure transducer captures its signals directly from the
hydraulic pressure that a catheter transmits from the interior of the artery (arterial
puncture) towards the transducer. For the arterial-occlusion monitors, the pressure
transducer captures its signals from the pneumatic pressure that a cuff transmits
from the periphery of the artery towards the transducer (Fig. 4.3). In the case of
tonometry blood pressure monitors, the pneumatic cuff is replaced by a pressure (or
displacement) sensor, and the pressure transducer requires additional initialization
information to translate force or displacement into pressure (typically an initializa-
tion value provided by an oscillometric blood pressure monitor). In both Figs. 4.2
and 4.3, the pressure waveform is further processed by an algorithm that determines
absolute Systolic and Diastolic Blood Pressure values. The main feature of these
approaches is that the pressure waveform is already expressed in pressure units,
being either mmHg or kPa.
As described previously, the common characteristic of all cuffless monitors is
that they rely on pulsatility sensors instead of pressure transducers, and therefore
the output of their processing layer is a blood pressure determination expressed in
arbitrary units (not in mmHg). Because of this fundamental limitation, cuffless
blood pressure monitors have historically introduced an additional processing layer
that transforms the non-pressure (uncalibrated) blood pressure values into absolute
blood pressure readings, by a so-called initialization procedure.
With this information in hand, a generic architecture for a cuffless blood pressure
monitor is introduced in Fig. 4.4. A detailed description of the different layers of the
architecture is provided in the following section.
Non-initialized BP
Pulsatility Pulsatility
Pulsatility-
energy Pulsatility waveform Initialization Systolic BP
based
sensor algorithm Diastolic BP
algorithm
Additional Calibration
cardio-synchronous Parameters
information
Body Transducer layer Processing layer Initialization layer Output
Non-invasive cuffless blood pressure monitor
Fig. 4.4 Generic architecture of a blood pressure monitor that requires no arterial occlusion, a
so-called cuffless blood pressure monitor. The monitor first relies on a pulsatility sensor (trans-
ducer layer) that generates a pulsatility waveform expressed in arbitrary units. The pulsatility
waveform contains information on the change in diameter of the artery occurring during at least
one cardiac cycle. By analyzing the pulsatility waveform, a pulsatility-based algorithm (process-
ing layer) generates uncalibrated determinations of blood pressure. Finally, an initialization algo-
rithm translates the uncalibrated blood pressure values into absolute blood pressure determinations
expressed in mmHg (initialization layer)
40 J. Solà
xamples of Implementation of Cuffless Blood Pressure

E
Monitors
A more detailed description of the generic architecture of Fig. 4.4 is provided in this

section, including practical examples on what technologies are known to implement
each layer of the architecture:
1. Transducer layer: At a given body location, and for each cardiac cycle, the
arrival of an arterial pressure wave generates a change in diameter of the arterial
bed (including arteries and arterioles underlying the skin surface) that further
induces a general displacement of the surrounding materials and anatomical
structures. This so-called pulsatility energy can be captured from the skin surface
at most body locations by a large variety of sensor modalities, as depicted in
Table 4.4. The common feature of any pulsatility sensor is thus that it generates
an electrical signal that codes a pulsatility waveform: a waveform which contour
describes the change of diameter of the underlying arterial bed.
2. Processing layer: This second layer of the architecture analyzes the pulsatility
waveform generated by the transducer layer in order to generate a non-initialized
blood pressure determination. Because the pulsatility waveform is generated by
the change in diameter of the arterial bed, and because the change in diameter is
generated by the arrival of a pressure wave at the arterial bed, pulsatility signals
are known to contain relevant information on the pressure waveform in time,
amplitude, and frequency domains. This information has been shown to be use-
ful to generate estimates of blood pressure, and/or estimates of changes of blood
pressure. Accordingly, Table 4.5 summarizes the most common strategies in use
to process pulsatility waveforms in order to extract blood pressure-related infor-
mation. Note that some pulsatility-based algorithms require the acquisition of
additional cardio-synchronous signals in order to perform denoising tasks, or in
order to introduce into the calculation information on triggering events, such as
Table 4.4 Examples of sensing modalities implemented in the transducer layer of cuffless blood
pressure monitors
Sensor technology Described in this book
Ballistocardiography (BCG) In Chaps. 5, 6, and 10
Electrical impedance tomography (EIT) Not covered, see Sola [2]
Impedance cardiography (ICG) In Chaps. 5, 6, and 10
Implantable photoplethysmography Not covered, see Theodor [3]
Phonocardiography (PCG) In Chap. 5
Photoplethysmography (PPG) In Chaps. 5, 6, and 8–10
Radar Not covered, see Solberg [4]
Seismocardiography (SCG) In Chap. 5
Tonometry In Chaps. 5, 6, and 8
Ultrasound (US) In Chap. 10
Video plethysmography (VPG) In Chaps. 6 and 8
Table 4.5 Examples of Pulsatility-based algorithms Described in this book

pulsatility-based algorithms
Regional PAT In Chaps. 5, 9, and 10
and additional cardio-
synchronous signals Regional PTT In Chaps. 6, 9, and 10
implemented in the Local central PTT In Chaps. 7 and 8
processing layer of cuffless Local peripheral PTT In Chaps. 6 and 9
blood pressure monitors Feature-based PWA In Chaps. 7, 8, and 10
Machine Learning-based PWA In Chaps. 6–9
Additional cardio-synchronous Described in this book
information
Electrocardiography (ECG) In Chaps. 5–7, 9, and 10
Impedance cardiography (ICG) In Chaps. 5, 6, and 10
Phonocardiography (PCG) In Chaps. 5 and 6
Table 4.6 Examples of Initialization algorithm Described in this book

algorithms and parameters
Parametric algorithm In Chaps. 5, 6, and 8–10
implemented in the
based on theoretical
initialization layer of
models
cuffless blood pressure
monitors Parametric algorithm In Chaps. 5, 6, and 8–10
based on empirical
models
Calibration parameters Described in this book
Person-specific In Chaps. 6 and 8–10
Population-based In Chaps. 6 and 8–10
Hybrid parameters In Chaps. 6 and 8–10
the opening of the aortic valve. The output of the processing layer of a cuffless
blood pressure monitor is typically a quantity expressed in non-pressure units
such as milliseconds, millivolts or Hz that correlates to blood pressure. In some
use-cases this output might be directly exploited to monitor blood pressure
trends, for instance to perform assessments on blood pressure variability.
However, a further processing layer is required to generate determinations of
systolic and diastolic blood pressure expressed in mmHg or kPa.
3 . Initialization layer: This final layer of a cuffless architecture is intended to
transform the non-initialized blood pressure determinations (expressed in non-
pressure units) into systolic and diastolic blood pressure values (expressed in
mmHg or kPa) that can be interpreted according to established clinical practices.
The additional information required to perform such initialization and the peri-
odicity of required initialization procedures largely vary according to the imple-
mented transducer and processing layers. Table 4.6 summarizes typical
implementations of this final layer.
42 J. Solà
Introduction to the Technical Chapters of this Book
The presented glossary and architecture support the technical chapters of this book.
In the following pages, the reader will find a review of the basics and the most
recent advances in the cuffless techniques of pulse arrival time (Chap. 5), pulse
wave velocity (Chap. 6), pulse wave decomposition (Chap. 7), pulse wave analysis
(Chap. 8), and machine learning developments (Chap. 9). Further, this book pres-
ents a detailed analysis of existing initialization techniques (Chap. 10) and the cur-
rent worldwide regulatory framework (Chap. 11). Finally, relevant aspects of
clinical trials for cuffless blood pressure monitors are presented (Chap. 12), con-
cluded by the expected next steps in the validation and implementation of cuffless
technologies in the field (Chap. 13).
Acknowledgements The author wants to specially thank Bastien Di Marco and Elisa Oliver for
the preparation of the infographics and the material of this chapter, Prof. Ramakrishna Mukkamala
and Prof. George Stergiou for the inputs on the terms and definitions, and the entire Aktiia team for
the revision work.
References
1. Geddes LA. Handbook of blood pressure measurement. Clifton: Humana Press; 1991. p. 1991.
2. Sola J. Continuous non-invasive blood pressure estimation, PhD Thesis No. 20093. Zurich:
ETHZ; 2011.
3. Theodor M. Subcutaneous blood pressure monitoring with an implantable optical sensor.
Biomed Microdevices. 2013;15(5):811–20. https://doi.org/10.1007
4. Solberg LE. Radar based central blood pressure estimation, PhD Thesis. Oslo: University of
Oslo; 2015.
Chapter 5
Pulse Arrival Time Techniques
Marshal S. Dhillon and Matthew J. Banet
Abstract For over 90 years, researchers and clinicians have worked on systems to
noninvasively measure pulse wave velocity, with the ultimate goal of measuring
continuous, cuffless blood-pressure. The design of multisensor systems to make two
or more measurements of the pulse wave along the human arterial tree has become
the hallmark of these efforts. Pulse arrival time measurements are traditionally
made by detecting a first fiducial point on the electrocardiogram (ECG) waveform,
and second fiducial point on a distal pulsatile waveform, such as the photoplethys-
mogram (PPG) or impedance cardiogram (ICG). By looking at the time difference
between these two fiducial points, an approximation for pulse wave velocity can be
calculated. Some other challenges exist in transforming the pulse arrival times into
absolute blood pressure, but these have been largely overcome, as evidenced by the
clinical-grade systems that are commercially available in the marketplace today.
Keywords Pulse arrival time · Pulse transit time · Pulse wave velocity · Blood
pressure · Continuous · Wearable · Cuffless · Noninvasive
Introduction
Methods for measuring physiological parameters using pulse arrival time (PAT) and
pulse transit time (PTT) have been researched and discussed for over 90 years. The
majority of work done in this time period has focused on PAT as a basis for a con-
tinuous, noninvasive blood pressure measurement. Efforts in this area have greatly
increased in the last 15 years, mostly because of technology advances (e.g., minia-
turization, cost reduction of components, module consolidation, and wearable com-
puting power) coupled with the unmet need for diagnosis and monitoring of
hypertensive patients worldwide.
M. S. Dhillon · M. J. Banet (*)

toSense, Inc., San Diego, CA, USA
e-mail: matt.banet@tosense.com

44 M. S. Dhillon and M. J. Banet
Definitions
PAT is generally defined as the time delay between the electrical activity of the
heart and a peripheral pulse measured further down the arterial tree. Although
the nomenclature is not consistent throughout the literature, PAT requires the
measure of the electrical activity of the heart (typically using an electrocardio-
gram (ECG)) and some measure of a mechanical activity of the pulse wave
(typically using a photoplethysmogram (PPG)). In contrast, PTT typically relies
on two mechanical measurements of pulse wave activity, rather than one
mechanical (e.g., PPG) and one electrical (e.g., ECG), as used in PAT measure-
ments. Despite these differences, both techniques attempt to accurately calcu-
late pulse wave velocity (PWV) within an individual patient, as this parameter
has a strong relationship with blood pressure.
Sensing Elements
To measure PAT, most methods utilize a fiducial from the QRS complex of the ECG
waveform (Q-point or R-point), and the onset of a pulse in the PPG waveform, as
shown in Fig. 5.1:
Fig. 5.1 ECG-PPG plot of PAT

5 Pulse Arrival Time Techniques 45
Transmissive and Reflective PPG Sensors
PPG sensors use an optical emitter at a specific wavelength and one or more corre-
sponding photodetectors to measure optical energy which is modulated by the
underlying vasculature. PPG sensors can be classified into two separate measure-
ment configurations: transmissive and reflective. Transmissive sensors have the
emitter(s) on one side of the tissue to be interrogated, and the photodetector(s) on
the opposite side of the tissue. In this configuration, the emitted light is transmitted
through the tissue and modulated by the underlying vasculature; the modulated
optical energy is then detected at the other side, as shown in Fig. 5.2. This configura-
tion is common in pulse oximeters designed to measure fingers and ear lobes.
Reflective sensors have the emitter(s) and the photodetector(s) on the same side
of the tissue, usually on the same plane. In this configuration, the emitted light pen-
etrates into the tissue, and some portion is modulated by the tissue and reflected
back to the photodetector(s), as shown in Fig. 5.3. Reflective sensors are common
for optical measurements on the forehead, arm, wrist, chest, and other areas where
a transmissive sensor is impractical due to the length of the optical path through the
tissue. Transmissive sensors provide a stronger pulsatile signal than reflective sen-
sors, often on the order of 10× to 100×.
With either configuration, the light sensed by the photodetector in a PPG sensor
contains a mixture of static (DC) and dynamic (AC) components. The static compo-
nent makes up the majority of the signal amplitude (>95% depending on the sensor
and measurement site), and is modulated by skin, muscle, fat, non-pulsatile blood,
and other tissue. The dynamic component is modulated of the heartbeat-induced
volumetric change in the vasculature in the light path with each cardiac cycle, which
yields a pulsatile signal [1].
Fig. 5.2 Transmissive
optical sensor
Fig. 5.3 Reflective optical

sensor
While many emitter wavelengths have been researched for systems that measure
PAT and PTT, the most popular wavelengths are infrared (~940 nm), red (~660 nm),
and green (~530 nm). Infrared (IR) and red are used in pulse oximetry sensors, and
light sources that generate them are readily available for experimentation and system
construction. Infrared is preferred over red due to its greater penetration depth, and
lack of variation with changes in blood oxygen content. The penetration depth of the
optical sensor is a subtle but important point, as greater penetration depth often yields
PPG pulses that are representative of larger artery blood flow, as opposed to capillary
or arteriole blood flow. Green emitters are used mainly in reflective-mode PPG sen-
sors, where they have been shown to yield stronger pulsatile waveforms than an IR
or red wavelength. This is due to the shorter penetration depth of the green wave-
length, which provides a stronger measurement of pulsatile blood flow near the sur-
face of the skin. While this is sufficient for measurements of pulse rate, the small
artery pulsatile signals are subject to morphology and timing changes due to vaso-
constriction and vasodilation (e.g., as a result of changes in skin temperature), and
can introduce error into a PAT-based blood pressure measurement.
Another consideration for PPG sensors is the applied pressure of the sensing
optics onto the skin of the subject, as this can have a dramatic effect on the morphol-
ogy of the measured pulse. Increasing the applied pressure, up to the level of the
subject’s mean arterial blood pressure, can increase the measured pulse amplitude
[2]. This is due to the principle of maximal vessel compliance during zero transmu-
ral pressure. However, this increase in amplitude can distort the timing information
within each PPG pulse, as well as cause discomfort to the patient, so it should be
approached with caution. In general, applied pressure is a bigger concern for trans-
missive PPG sensors, where a wrap or a spring-loaded housing can be used to impart
large amounts of force on the tissue being measured.
Impedance Plethysmography and Impedance Cardiology
Impedance plethysmography (IPG) and impedance cardiography (ICG) are other

well-researched methodologies for measuring distal pulsatile waveforms in a PAT-
measurement system. In this case, PAT is calculated as the time delay between the
QRS complex of the ECG waveform, and the zero crossing (also referred to as
“point B”) of the IPG/ICG waveform [3], as shown in Fig. 5.4.
IPG/ICG is the central technology in many commercially available bioimpedance
systems that noninvasively measure thoracic impedance (a proxy for body fluids), and
hemodynamic properties such as cardiac output, stroke volume, and systemic vascular
resistance. IPG/ICG systems traditionally use a multi-electrode system (as shown in
Fig. 5.5) where a high-frequency, low-amperage electrical current is applied to the
subject’s body through dedicated electrodes (impedance drive electrodes), and the
modulated signal is measured through the remaining electrodes (impedance sense
electrodes). These systems measure the impedance of the underlying tissue, and typi-
cally measure physiological activity at a greater depth than PPG sensors [4].
Fig. 5.4 ECG-IPG plot of PAT
Fig. 5.5 IPG multi-electrode diagram
The applied electrical current is conducted by the various tissues and organs in
the path of the impedance drive electrodes. Similar to PPG waveforms, the mea-
sured IPG/ICG waveform features a static (DC) component, and a dynamic (AC)
component. The static component makes up the majority of the measured signal,
and is affected by the amount of bone, muscle, fat, and other tissue in the measurement
path. The dynamic component is modulated by the pulsatile component of each

cardiac cycle. Due to the low resistance of blood, the applied current is well con-
ducted through the arterial vasculature between the impedance drive electrodes, and
the resulting dynamic signal is a good indicator of the pulsatile blood volume. Since
most IPG/ICG systems involve electrodes applied on the thorax, the dynamic IPG/
ICG component is also modulated by the respiration effort of the subject. The respi-
ratory modulation is usually 10X-100X greater than the cardiac modulation of the
IPG/ICG waveform, which can complicate the measurement of cardiac pulses.
However, since the frequencies of interest for respiration and cardiac pulses have
little overlap, they can be filtered out using analog or digital signal processing.
Other Sensing Technologies
Other less common methods for measuring PAT include those using ballistocardio-
gram (BCG), seismocardiogram (SCG), and phonocardiogram (PCG) waveforms.
BCG and SCG are both measurements of the vibration or recoil of the body due to
the force of each heartbeat-induced pulse ejected from the left ventricle into the
aorta. BCG is commonly measured using highly sensitive load sensors (e.g., load
cells). These are present in passive devices like weight scales, chairs, and beds [5].
The SCG is commonly measured using accelerometers, typically placed on or near
the sternum. Both measurements provide well-defined waveforms corresponding to
each cardiac pulse, although their morphology is not typical of standard pulsatile
waveforms, as shown in Fig. 5.6. The fiducial points are well described in literature
and could provide useful for future research efforts.
Fig. 5.6 Typical BCG and SCG waveforms

Fig. 5.7 S1/S2 heart sounds with ECG
PCG is a measurement of the acoustics of the cardiac cycle, and is essentially the
output of a digital stethoscope. In a healthy heart, a PCG sensor measures the sounds
of the mitral and tricuspid valves closing (S1) and the sounds of the pulmonary and
aortic valves closing (S2). These are the classic “lub dub” sounds heard with each
heartbeat through a stethoscope, as shown in Fig. 5.7.
In other cases, the sounds S3 and S4, which are faint and slightly precede S1 and
S2 respectively, can be measured with a PCG sensor. In some cases, the presence of
these sounds can indicate cardiac abnormalities or congestive heart failure. A PCG
sensor is also useful for identifying other sounds during the cardiac cycle which can
indicate issues with heart valves (e.g., heart murmurs), or conditions like pericardi-
tis. For PAT-based systems, the time delay between the fiducial of the ECG wave-
form and the S1/S2 markers can indicate how much of the cardiac cycle is spent in
systoli and diastoli, and could potentially be used to identify changes in blood pres-
sure over time.
Challenges of PAT-Based Blood Pressure Monitors
Underlying Physiology
Theoretical equations, such as Bramwell–Hill and Moens–Korteweg, describe the

relationship between PWV (e.g., the speed at which a pulse wave leaves the left
ventricle of the heart and travels down the arterial tree) to changes in blood pressure,
with some simplified accounting for arterial compliance. PAT-based methods
attempt to utilize this relationship by providing a noninvasive, continuous measure-

ment of PWV in order to calculate blood pressure. However, classic PAT, taken in
isolation, presents some challenges to comprehensive blood pressure measurement,
which are described below.
The Effect of Pre-ejection Period
The goal of most PAT-based systems is to measure and calculate a value that repre-
sents the PWV for a given cardiac pulse, and to relate that measured value to the
subject’s blood pressure. However, the common starting point of the time measure-
ment—a fiducial marker on the ECG—is not usually indicative of the time that the
mechanical pulse wave starts travelling from the left ventricle into the aorta. More
specifically, there is a delay from the electrical depolarization of the heart’s ventricles
(indicated by the ECG’s QRS complex), and the mechanical activation of the ven-
tricular muscle. This period is known as the electromechanical activation time
(EMAT). After EMAT is completed and the ventricular muscle is activated, there is
another delay where the ventricles have started contraction but blood has not started
to eject into the aorta. This time period is known as the isovolumic contraction time
(IVCT). Together, the EMAT and IVCT are often referred to as the pre-ejection period
(PEP). PAT measurements include both PEP and the PTT of the cardiac wave.
PAT = PEP + PTT (5.1)
PEP = EMAT + IVCT (5.2)
Another issue with the electromechanical delay is that it can change for various
reasons, often in a very short time period. Vasoactive drugs, hydration status, fluid
overload, venous return, and a patient’s posture can all cause changes to the electro-
mechanical delay, which can add a variable amount of error to measurement of PAT
[6, 7]. The most effective solutions to eliminate this source of error are methods that
directly measure the electromechanical delay or use additional sensors to directly
measure the PTT. Such sensors, for example, lack ECG-measuring systems, and
instead utilize sensors that measure waveforms such as PPG, IPG/ICG, BCG, SCG,
and PCG.
Motion Artifacts
Another challenge for PAT-measuring systems (and indeed all wearable systems) is
motion artifacts. The magnitude of motion artifacts is based upon the sensor’s loca-
tion and the activities undertaken by a subject during measurement. While it may
not be realistic to have a system that measures perfectly during vigorous exercise, a
reasonable goal is for a PAT-measuring system is to operate accurately during nor-

mal motions undertaken by patients in an ambulatory environment: brushing teeth,
walking, using the restroom, etc. There are two levels to dealing with motion arti-
fact. The first is accurately identifying motion-corrupted waveforms and not includ-
ing them in the calculation of blood pressure values. The second is using an active
measure of the corrupting motion (e.g., with an accelerometer or gyroscope), and
then using adaptive noise cancellation techniques to remove the motion artifact
from the corrupted waveforms so that an accurate measurement can be made.
The ECG waveform is relatively motion-tolerant, and thus the biggest challenge
related to motion artifacts for most PAT-measuring systems is from the sensor mea-
suring the distal waveform (e.g., the PPG waveform). For systems that utilize an
optical sensor on the wrist or finger, being able to ignore and possibly read through
periods of motion artifact is essential to avoiding erroneous blood pressure values
being calculated. Systems that utilize sensors on the thorax, the neck, or the head for
the distal waveform are less affected but are still well served by having a solution
for ameliorating motion artifacts.
For PAT-measuring systems that utilize a distal waveform sensor on the arm, the
wrist, or fingers, hydrostatic pressure error is another challenge that needs to be
addressed. When the sensor measuring the distal waveform is raised above the level of
the heart, the actual blood pressure at the sensor site is lower than the blood pressure at
heart level. Conversely, when the sensor is placed below the level of the heart, the blood
pressure at the sensor site is higher than the blood pressure at heart level. Both of these
changes in blood pressure are due to the hydrostatic effects of gravity on the blood in
the arterial tree. The change in PAT due to hydrostatic effects is significant enough that
efforts have been made to use this effect for calibration purposes [8]. For PAT-measuring
systems where hydrostatic error is a concern, an objective measure of the position of
the distal waveform sensor (e.g., using accelerometers) is recommended.
ositioning PAT-Based Monitors Among Conventional

P
Measurements of Blood Pressure
There are many existing methods for measuring blood pressure, namely because
blood pressure is an important clinical parameter and useful in many different
patient situations. The following section provides a brief overview of the existing
methods, and the advantages and disadvantages compared to a PAT-based monitor.
Auscultation
Auscultation is the classic and most widely accepted method for measuring blood
pressure. In addition to its use in clinical settings, auscultation is also accepted as a
reference method for validating blood pressure-monitoring technologies based on
Image 5.1 Auscultation example
PAT and PTT by regulatory agencies. It is based on the principle of occluding blood
flow in an artery with a manual cuff (usually the brachial artery), and then listening
for sounds indicating the return of turbulent blood flow, and laminar blood flow
(called Korotkoff sounds) that correspond to systolic and diastolic pressure respec-
tively [9]. Auscultation is accurate, clinically accepted, noninvasive and relatively
easy to perform; however, its measurements are not continuous, and can be uncom-
fortable (Image 5.1).
Oscillometry
Oscillometry is a widely accepted technology for measuring blood pressure that is

automated and relatively simple to deploy as compared to auscultation. Like auscul-
tation, it is based on the principle of occluding blood flow with an inflatable cuff.
Unlike auscultation, it measures the pulses transduced through the inflatable cuff,
and exploits the fact that the pulse amplitude is modulated by the difference between
mean arterial pressure (MAP) and the applied pressure in the cuff. The pulse ampli-
tudes increase when the applied pressure is between the patient’s diastolic blood
pressure (DBP) and MAP. The pulse amplitudes decrease when the applied pressure
is between MAP and systolic blood pressure (SBP). Variation of pulse amplitudes
with applied pressure yields a quasi-symmetric waveform from which SBP and
DBP can be determined using empirical formulas [10]. Oscillometric devices are
noninvasive, clinically accepted, easy to use, and readily available for patients to
Image 5.2 Oscillometry example
measure the blood pressure at home; however, like auscultation, oscillometric

devices are not continuous, can be uncomfortable, and can be inaccurate if used
improperly or not thoroughly validated (Image 5.2).
Invasive Measurements
Intra-arterial catheters (sometimes called “arterial lines”) are the standard of care
for critically ill patients where continuous blood pressure monitoring is necessary.
Arterial catheters are also recognized as a reliable reference method for validating
new technologies by regulatory agencies. This method utilizes a pressure transducer
incorporated with an invasive catheter in contact (through fluid) with a pulsating
artery (e.g., radial artery) to provide continuous, direct measurement of blood pres-
sure. Arterial catheters are accurate and clinically accepted, and provide continuous
measurement; however, they are invasive and require expert users to place and mon-
itor the catheter site, and patients must be relatively stationary during monitoring
(Image 5.3).
Volume Clamp
The volume clamp method, also known as vascular unloading or the method of
Penaz, provides a noninvasive, continuous measurement of blood pressure by using
a combination of sensors. In these systems, a finger cuff connected to a high-speed
Image 5.3 Arterial line example
servo pump is used to cancel out volumetric change of blood flow in the fingers with
each cardiac cycle. The lack of volumetric change is verified by a finger-mounted
PPG sensor, which senses a waveform lacking an AC component. The pressure
applied through the high-speed servo is essentially the inverse of the arterial blood
pressure waveform at the finger. Using an oscillometric calibration and empirical
models, a continuous blood pressure waveform representative of larger artery wave-
forms can be provided. Volume clamp methods are noninvasive and continuous;
however, they require a large, complex measurement apparatus. The finger cuff can
be uncomfortable for patients, and the system can be inaccurate if used improperly
(Image 5.4).
Tonometry
Arterial tonometry provides a noninvasive, continuous measurement of blood pres-

sure by applying a sensitive transducer directly above a pulsating artery located
above a rigid surface (e.g., bone). The transducer needs to be placed directly above
the pulsating artery (usually the radial artery), and pressure needs to be applied to
flatten the surface of the artery. Changes in arterial pressure with each cardiac cycle
are measured by the transducer, and can be analyzed to yield blood pressure. Arterial
tonometry is noninvasive and continuous; however, it is very sensitive to motion and
requires precise placement of the transducer, making it difficult to implement
(Image 5.5).
Image 5.4 Volume clamp example
Image 5.5 Tonometry example
PAT-Based Measurements
It is clear that while many methods exist for measuring blood pressure, each method
features nonideal aspects that need to be considered depending on the use case for
the blood pressure measurement. Furthermore, none of the existing methods for
noninvasive blood pressure measurement meet the needs of the growing hyperten-
sive population, both in the clinic and at home. An ideal solution for these environ-
ments would be accurate, wearable, passive, continuous, and easily accessible.
PAT-measuring techniques have the potential to satisfy all of these goals, which is a
reason so much effort has been invested in developing them.
Devices and History
Pulse wave velocity systems, using both PAT and PTT, have been studied regularly
for over 90 years. One of the earliest experiments involving a pulse wave velocity
system was conducted by Hickson and McSwiney in 1924 [11]. In this experiment,
two mechanical pulses were simultaneously recorded from the carotid and radial
arteries, using hot-wire sphygmographs. In this setup, a thin filament of wire is
heated to a very high temperature, using a battery or other energy source, and then
connected to a galvanometer. The hot wire is then placed in a funnel or tube on the
skin, above the artery of interest. Small perturbations of air in the funnel, caused by
each peripheral pulse, cause a slight cooling in the hot wire and a commensurate
current into the galvanometer. The authors calculated PWV using the two pulse
signals, and found a slowing of the pulse wave as each subject raised their arm, due
to hydrostatic pressure effects.
Another seminal experiment was conducted Thomas in 1955 [12]. In this experi-
ment, PAT was calculated using the time delay between an ECG waveform, and a
crystal microphone placed on the dorsalis pedis artery as a pulse transducer. Both
waveforms were recorded and pulse wave velocity was manually calculated from
the time delay, and shown to track changes in diastolic pressure.
In 1964, Weltman, Sullivan, and Bredon conducted an experiment which was
one of the first to utilize a computerized PWV measurement system [13]. They used
a standard ECG measurement setup, combined with a crystal microphone to trans-
duce a peripheral pulse. The computerized system used adjustable triggers to detect
the ECG R-peak and peripheral pulse, and then output an analog voltage that repre-
sented the time difference of the two signals. An image of the computerized system
is shown below in Image 5.6.
There are a handful of commercially available pulse arrival time systems as well.
Nihon Kohden has a series of multiparameter vital sign monitors that also measures
PAT to enhance their NIBP oscillometric measurement [14]. Using PAT measured
from the R-peak of the ECG waveform to the onset of the PPG pulse as measured
by a fingertip oximeter, the vital sign monitor can determine when more NIBP mea-
surements are needed for a patient, rather than only inflating the NIBP cuff at pre-
determined intervals. These systems from Nihon Kohden have been approved by
Japan’s PMDA and other regulatory agencies.
Sotera Wireless markets a body-worn multiparameter vital sign monitor, ViSi
Mobile, that features a measurement of continuous, noninvasive blood pressure
(cNIBP) using PAT [15]. ViSi Mobile has sensors deployed on a patient’s chest,
upper arm, wrist, and fingers. It measures PAT using the time delay between the
ECG measured at the chest and a pulsatile PPG waveform measured at the base of
the thumb or finger. In addition, the system contains a network of accelerometers
and impedance-based sensing to mitigate some of the challenges faced by conven-
tional PAT techniques. ViSi Mobile has been cleared by the US FDA and conforms
to the applicable governing standard (ISO 81060-2) for blood pressure accuracy.
Image 5.6 Computerized PWV system from 1964 [13]
Somnomedics has a body-worn multiparameter vital sign monitor (Somnotouch

NIBP, somnomedics.eu) that is similar to the ViSi Mobile from Sotera Wireless. The
Somnotouch NIBP system measures pulse arrival time using the time delay between
the ECG and a pulsatile PPG waveform measured at the fingertip, using a conven-
tional pulse oximetry probe [16]. The Somnotouch NIBP system has published
results from a clinical trial conducted using the European Society of Hypertension
International Protocol to quantify the accuracy of the blood pressure measurement.
Biopac has a modular system that can measure ECG and PPG waveforms, and
then process the waveforms with a proprietary software package that automatically
calculates PAT [17]. Biopac also provides a variety of reflective and transmissive
PPG sensors, IPG/ICG sensors, and accelerometers for multimodal data acquisition.
The Biopac system is for research purposes only and not cleared for use in patient
monitoring.
There are also a group of handheld or wrist-worn devices that incorporate a com-
bination of PPG and ECG sensors, and provide some measurement of absolute or
relative blood pressure. Examples of these devices include the CheckMe Pro [18],
Care Up [19], and the H2-BP [20]. These devices have been not widely accepted
clinically or commercially due to the lack of supporting data for the accuracy of
their blood pressure claims. While these devices are on the leading edge of acces-
sibility and usability, their slow adoption indicates the need for objective clinical
results to support blood pressure measurement accuracy for PAT-based systems.
Future Directions
The future of devices that calculate blood pressure using PAT and PTT lies in the
continued miniaturization and wearability of the systems, the accuracy of their
blood pressure algorithms, and their increasing accessibility for home and ambula-
tory use. Major electrical component vendors are reducing the size, cost, and power
consumption for many of the building blocks required for the sensing components
required by pulse arrival time systems. In addition, there continues to be an emer-
gence of specialized integrated circuits, combining multiple components into a
single package for biomedical sensing applications, that will lead to low-cost, effec-
tive systems. As wearable systems are able to measure more features related to
changes in blood pressure, artificial intelligence techniques are becoming a leading
method for combining and distilling the information into an accurate blood pressure
measurement. This shift requires more computation than traditional models that use
only a handful of features, and increases in the computing power and memory of
wearable systems is needed.
Many popular consumer watches and wearables have already incorporated ECG
and PPG sensors, making these technologies accessible to large amounts of con-
sumers outside of the hospital. Research efforts have also shown promise for modi-
fying the world’s most popular consumer device, the smartphone, into a simple and
accurate blood pressure monitor [21]. The prevalence of these consumer sensors,
and smartphone modifications, coupled with the technology improvements listed
above, will usher in the next-generation of pulse arrival time systems. These next-
generation systems will provide a new level of passive and accurate blood pressure
monitoring for use in both home and clinic environments.
References
1. Mukkamala R, Hahn JO, et al. Toward ubiquitous blood pressure monitoring via pulse transit
time: theory and practice. IEEE Trans Biomed Eng. 2015;62(8):1879–901.
2. Teng XF, Zhang YT. The effect of applied sensor contact force on pulse transit time. Physiol
Meas. 2006;27:675.
3. Bang S, et al. A pulse transit time measurement method based on electrocardiography and
bioimpedance. In: 2009 IEEE Biomedical Circuits and Systems Conference. 2009. p. 153–6.
4. Patterson R. Fundamentals of impedance cardiography. IEEE Eng Med Biol Mag. 1989;8:35–8.
5. Inan OT, et al. Robust ballistocardiogram acquisition for home monitoring. Physiol Meas.
2009;30:169.
6. Payne RA, et al. Pulse transit time measured from the ECG: an unreliable marker of beat-to-
beat blood pressure. J Appl Physiol. 2006;100:136–41.
7. Zhang G, et al. Pulse arrival time is not an adequate surrogate for pulse transit time as a marker
of blood pressure. J Appl Physiol. 2011;111:1681–6.
8. McCombie DB, et al. Motion based adaptive calibration of pulse transit time measurements to
arterial blood pressure for an autonomous, wearable blood pressure monitor. Conf Proc IEEE
Eng Med Biol Soc. 2008;2008:989–92.
9. Perloff D, et al. Human blood pressure determination by sphygmomanometry. Circulation.

1993;88:2460–70.
10. Alpert BS, et al. Oscillometric blood pressure: a review for clinicians. J Am Soc Hypertens.
2014;8:930–8.
11. Hickson SK, McSwiney BA. The effect of variation in blood pressure on pulse wave velocity
in the brachial artery in man. J Physiol. 1924;59:217–20.
12. Thomas JG. A method for continuously indicating blood pressure. J Physiol. 1955;129(3):75–6P.
13. Weltman G, Sullivan G, Bredon D. The continuous measurement of arterial pulse wave veloc-
ity. Med Electron Biol Eng. 1964;2:145–54.
14. “Innovative Technologies”, Nihon Kohden. https://ae.nihonkohden.com/en/innovativetechnologies/
pwtt.
15. “Why Visi”, Sotera Wireless. https://www.soterawireless.com/why-visi/.
16. “Somnotouch NIBP”, Somnomedics. https://somnomedics.eu/products/cardiology/24h-blood
pressure-24h-ecg/somnotouch-nibp/.
17. “Pulse Transit Time (PTT), Pulse Wave Velocity (PWV), and Pulse Wave Amplitude

(PWA)”, Biopac. https://www.biopac.com/application/plethysmography/advanced-feature/
pulse-transit-time-ptt-and-relative-bp-measurement/.
18. “CheckMe Pro”, Viatom. https://www.viatomtech.com/checkme-pro.
19. “Features”, Care Up. http://care-up.com/.
20. “H2-BP”, H2 Care. http://h2care.com/18.
21. Chandrasekhar A, Kim CS, Naji M, Natarajan K, Hahn JO, Mukkamala R. Smartphone-based
blood pressure monitoring via the oscillometric finger-pressing method. Sci Transl Med.
2018;10(431):eaap8674.
Chapter 6
Pulse Wave Velocity Techniques
Jim Li
Abstract Recently, pulse wave velocity (PWV), or its reciprocal pulse transit time
(PTT), has been intensively investigated as a promising technique for continuous,
cuffless, and noninvasive blood pressure (BP) monitoring. BP is mathematically
derived through PTT, or the “time delay” in propagation of pressure waves in the
vascular system, which can be easily derived from two pulse signals, including elec-
trocardiography (ECG) and pulse plethysmography (PPG) signals, together with
adequate calibration procedure. Practical steps in applying this method as well as
mathematical models in estimating BP were reviewed; while limitations of this
approach, such as the need for individual calibration and the need for a reasonably
stable condition were discussed. The future of this technology can be potentially
used in, but not limited to, continuous BP monitoring, BP change tracker, and trig-
ger for absolute BP measurement. Furthermore, with machine learning, the initially
extract surrogate cardiovascular indexes from physiological signals can be used to
train and adapt to the model to further improve the accuracy of BP prediction.
Keywords Blood pressure · Pulse wave velocity · Pulse transit time · Pulse arrival
time · Electrocardiogram · Pulse plethysmography · Wearable · Cuffless ·
Calibration · Machine learning
History Overview
Inspired by the invasive intra-arterial continuous blood pressure measurement

through clinical cannulation in 1949, researchers started exploring noninvasive
alternative for continuous blood pressure monitoring. In 1963, arterial tonometry
was used by Pressman et al. by applanating a superficial artery against a bone with
an external transducer until the artery is flattened, where the tangential arterial wall
tension no longer affects the vertical force measured [1]. Quickly, arterial tonometry
J. Li (*)
Global Medical Affairs, Omron Healthcare, Inc., Lake Forest, IL, USA
e-mail: jim.li@Omron.com

62 J. Li
was improved and widely used by the researchers in the measurement of pulse wave
velocity (PWV) for evaluation of arterial stiffness and pulse wave analysis. The
commonly applied sites include the radial artery, carotid artery, and femoral artery.
The validity of measuring arterial BP depends on the applanation of the artery, and
therefore encounters practical problems of sensor positioning, motion artifacts, cali-
bration, etc. To enhance usability, tonometry sensor array was used to ensure at least
one sensor among multiple ones is precisely positioned to capture good pulse wave
signal. AtCor SphygmoCor (AtCor Medical, Sydney, Australia) and Form/VP-2000
and HEM-9000AI (Omron Healthcare Co., Ltd, Kyoto, Japan) are example of com-
mercialized research tools that utilizing arterial tonometry. These tools were mainly
for pulse wave analysis, arterial stiffness, and central BP measurement. Even though
they are capable of providing continuous pulse waves for short period, they are not
a true noninvasive continuous BP monitor.
Another development in noninvasive continuous BP monitoring field is the use
of volume clamp on the finger technique proposed by Peñáz from Czech in 1973
[2]. The system has an inflatable finger cuff with a built-in photoplethysmograph
(PPG) sensor, and a closed loop servo system to apply a pulsating cuff pressure to
the finger arteries that is precisely opposite to the intra-arterial pressure. In equilib-
rium, when the cuff pressure equals the arterial pressure, the difference between the
intra-arterial pressure and external applied cuff pressure will become zero, which is
called the set point. The key to volume clamp method is to reach the set point. On
the basis of the volume clamp method, some research tools were commercialized,
including Finapres (Finapres Medical Systems, Enschede, The Netherlands) and
CNSystems (CNSystems Medizintechnik GmbH, Graz, Austria).
Despite being noninvasive continuous blood pressure measurement techniques,
both tonometry and volume clamp techniques are intrusive because both require the
application of external pressure or force on cuff during the entire course of the
monitoring, which leads to discomfort and motion artifacts. Another approach,
which is based using PWV or its reciprocal, pulse transit time (PTT), can be an
attractive alternative for measuring unobtrusive continuous BP. In principle, PWV
depends on the property of the arterial wall, which varies with the arterial pressure.
The usage of PTT can be dated back to 1959 when Weltman et al. designed the
PWV computer by utilizing the ECG and a pulse signal to define pulse transit time
over a known arterial length [3]. Recently, PTT became a popular way of achieving
cuffless continuous BP measurement through using wearable objects including
watches, rings, shirts, eyeglasses, smartphones, and cameras, as well as daily objects
such as sleeping cushions, chairs, and weighing scales.
Most of these methods are still in the research stage. The few commercially
available devices include Sotera ViSi Mobile (Sotera Wireless, San Diego,
USA) continuous, noninvasive BP (cNIBP) monitoring and SOMNOtouchTM-
NIBP (SOMNOmedics GmbH, Randersacker, Germany), which are based on the
PTT method. The ViSi’s cNIBP is determined on a beat-to-beat basis employing
PTT and calibration with automatic noninvasive BP method.
6 Pulse Wave Velocity Techniques 63
Physiological Description
PWV is usually assessed using the arrival time of a pressure wave propagating
through the arterial tree in a certain distance between the proximal and distal arterial
sites, in the form of PWV = L/PTT, where L is the distance between the proximal
and distal sites. Because of the complexity of distance measurement, PWV can be
indirectly approximated using PTT, which can be easily derived from two pulse
signals, including electrocardiography (ECG) and pulse plethysmography (PPG)
signals (Fig. 6.1). By using a calibration procedure, the measured PTT can be trans-
lated into arterial pressure by using an appropriate model.
Pulse transit time refers to the time it takes a pulse wave to travel between two
arterial sites. Like pulse wave velocity, PTT is a measure for arterial stiffness. When
blood pressure increases, the vascular tone increases and the arterial wall becomes
stiffer, causing the PTT to shorten. Conversely, when blood pressure falls, vascular
tone decreases and the arterial wall becomes less stiffer, and PTT increases.
Increased stiffness can be either structural—due to aging and atherosclerosis, or
functional—due to higher sympathetic activity or elevated blood pressure [4]. But
at the same time, the multiple factors besides blood pressure that influences vascular
stiffness mechanisms itself make the technique based on PWV or PTT sensitive but
not specific, particularly for peripheral/muscular arteries, and less prominent in cen-
tral/elastic arteries. Such influence presents a limitation of PWV or PTT for blood
pressure estimation in real life.
Fig. 6.1 Pulse wave, pulse transit time (PTT), and pulse arrival time (PAT)
64 J. Li
Another relevant concept and parameter is the pulse arrival time (PAT), as
described in the previous chapter, which measures the time difference between the
R-peak of ECG and a characteristic point of pulse plethysmography (PPG) wave-
form. PAT is the sum of PTT of the pressure wave and the pre-ejection period (PEP)
delay:
PAT = PTT + PEP (6.1)
PEP is the time needed to convert the electrical signal into a mechanical pumping
force and isovolumetric contraction to open the aortic valve, and can be calculated
by the delay between R-wave and impedance cardiography (ICG), as shown in
Fig. 6.1. PEP is a delay that changes with various factors such as stress, physical
activity, age and emotion [5]. The popularity of using PAT to estimate blood pres-
sure is based on the simplicity of obtaining PAT by referencing ECG R wave, which
is precise and easy to get. The disadvantage is the introducing the new variable of
PEP, especially when PTT is small when the distal measurement point is close to the
core body.
Studies showed that PEP accounts for 7% of the RR interval to approximately
20% of PTT measured at the finger tips at rest [6]. The impact of PEP decreases
with distance from the heart, but becomes more significant when heart rate lowered,
and therefore should be subtracted out to obtain PTT. It is generally thought that
PTT has higher correlation with SBP, DBP, and MAP than PAT, but with some stud-
ies stating PAT is a better indicator of SBP due to its dependency on both ventricular
contraction and vascular function.
Using PTT to effectively estimate blood pressure needs to be handled in a rea-
sonably stable condition, where smooth muscle contraction is minimal and viscous
effects are negligible. Such conditions are best met by measuring PTT through cen-
tral arteries. It also needs to be individually calibrated to associate the PTT to the
absolute blood pressure values. Furthermore, aging or diseases can potentially
change the arterial elasticity characteristics; therefore, periodic recalibration is nec-
essary to ensure the close correlation between PTT and blood pressure values dur-
ing long-term BP monitoring.
Practical Approach
The practical PTT-based cuffless BP monitoring involves a three-step approach:

Step 1—obtain the proximal and distal arterial waveforms;
Step 2—calculate PTT from the waveforms, either from foot-to-foot, or peak-to-
peak of the waveforms;
Step 3—calibration of PTT (in units of ms) to BP (in units of mmHg)
There are various challenges along the way during implementation of the above
steps. First challenge is to reliably obtain good quality arterial waveforms in a
r easonably convenient way. Generally speaking, the more convenient the waveform
measurement is, the more noise contamination can be resulted. For example, when
the proximal and distal sites are apart, the signal-to-noise ratio will be higher, but
such design will lead to two different sensors, and compromise on the convenience.
On the other hand, getting the two sites next to each other can make a great case on
convenience, but the signal-to-noise ratio might be an issue given too small PTT
between the two sites.
Second challenge is to calculate the PTT with suboptimal waveforms obtained
by applying algorithm to enhance the signal-to-noise ratio. It is necessary to strike
a balance between waveform quality and convenience, and rely on signal processing
algorithm to optimize the suboptimal waveforms.
The third and the most difficult challenge is to choose the right calibration and
appropriate calibration intervals. The calibration curve relating PTT to BP is depen-
dent on quite a few factors, including individual specific characteristics such as
body height, relevant artery cross-sectional area, and vascular tree compliance.
Therefore, obtaining an individual specific calibration curve through multiple blood
pressure readings which covers the target blood pressure range is optimal. However,
constructing such a curve requires cuff BP measurements from the subject, and the
calibration curve need to be updated at a rate faster than the arteriosclerotic process
(e.g., up to a few years at a time).
A final major challenge is the independent determination of both systolic BP and
diastolic BP at the same time. This challenge is important, as isolated systolic
hypertension often occurs in the elderly. However, conventionally estimated PTT
may only be a marker of diastolic BP.
Mathematical Modeling
PTT based BP estimation relies on two basic equations for arterial wave propaga-
tion. The first is the relationship between the Young’s modulus (E) and the arterial
pressure (P):
E = Eo e aP (6.2)

where a is a parameter that is related to the vessel and Eo is the Young’s modulus for
zero arterial pressure. The formula estimates the arterial pressure if a and Eo are
adjusted for subject (age, gender, health condition, etc.) on the elasticity due to the
change in the wall composition.
The second is the Moens–Korteweg equation in which the elasticity of arteries
determines the propagation speed, which is the pressure’s pulse wave velocity
(PWV).
66 J. Li
hE
PWV = (6.3)
ρd

where h is the thickness in an elastic artery, d is the diameter, and ρ is the blood
density.
Combining Eqs. (6.2) and (6.3), the relationship between P and PWV can be
derived by the Bramwell–Hill equation
hEo e aP
PWV = L / ∆t = (6.4)
ρd

where L is the length the pulse wave passed through, and ∆t is PTT, or the time
delay [7]:
This equation indicates that an increase in pressure will lead to an increase in
PWV and decrease in time delay.
Various mathematical models were used to approximate the relationships
between BP and the PTT including the following:
1. Logarithmic model—Rearranging the Bramwell–Hill equation, we will have the
logarithmic relationship between BP and PTT in the form of [8]:
 1   ρ dL 
2
 2
P =  −  ln ( ∆t ) +   ln  
 a  a   hEo 
Or in a simplified version as:
BP = A ∗ ln ( ∆t ) + B (6.5)

Here, A and B are subject-specific constants and they can be obtained through a
regression analysis between the reference BP and the corresponding PTT, or ∆t—
the time delay [9].
The above mathematical relationship is the theoretical basis between PTT and
BP. The more generalized relationship can be summarized as
BP = A ∗ f ( ∆t ) + B (6.6)

where f(∆t) is a specific function of the PTT, or ∆t—the time delay. Among various
approximation of BP from PTT, the following models were used besides the loga-
rithmic model shown in (6.5):
2. Linear model—Assuming there is a negligible change in the arterial thickness
and diameter with pressure variation, BP and the time delay can be linearly
related [10]:
BP = A ∗ ∆t + B (6.7)
3. Inverse Square Model
BP = A / ∆t 2 + B (6.8)
where A and B are constants that are related to subject’s biometrics such as height
and blood density [6]
4. Inverse Model—an inverse relationship between BP and PTT [11]
BP = A / ∆t + B (6.9)
While PTT is the most important parameter for estimating BP, other factors, espe-
cially those that are related to cardiovascular activity can be added to enhance the
robustness of the model. Specifically, heart rate (HR) represents the cardiac cycle
and influences the heart’s preload and the cardiac output (CO), which positively
impact BP as the pressure on the arterial walls. HR is proportional to the volume of
blood ejected. HR and BP are regulated by the autonomic nervous system which has
been found to be inversely related, depending on the baroreflex activity. HR is cal-
culated from the RR interval in ECG signals and has been incorporated in several
algorithms to estimate BP, contributing some improvement in accuracy.
BP = A ∗ f ( ∆t ) + B ∗ g ( HR ) + C (6.10)

where A, B, and C are constants, and ∆t and HR are two variables [12].
In order to apply a mathematical model, one needs to vary the BP over a consid-
erable range to obtain the curve that can relate PTT or PAT closely to BP. Models
under different conditions, generally exercising, hydrostatic posture, Valsalva
maneuver, and medication, have been commonly used [13–16].
Technical Details on Measurement Methods
The most direct way is to measure the time delay between two arterial sites. In
research, it is easy to handle such way, but in consumer oriented application, it is
challenging, because the device needs to be simple and compact, which leads to
short PTT due to the two testing sites being quite close to each other quite sensitive
to error.
Alternative way is to calculate the difference between two PATs at two different
sites: for example, one is at the finger tip, but the other is at the wrist along the same
arm. The approach cancels the same pre-ejection period PEP. The disadvantage of
this approach is that the user needs to wear sensors at two different sites.
68 J. Li
The focus of the PPT measurement is on conveniently measuring waveforms,

contact or noncontact, most of which are relevant waveforms instead of blood pres-
sure waveforms.
PPG, bioimpedance, electrocardiography (ECG), ballistocardiography (BCG),
and video plethysmography (VPG) are among the most popular methods. These are
based on either optical or electrical principles.
PPG method Uses optical transmittance or reflectance to measure waveforms of
proximal and distal blood volumes—the same technique as used in pulse oximeters
to measure arterial oxygen saturation [17]. Transmission-mode and reflectance-
mode PPG offer different advantages and limitations.
The reflectance-mode PPG is less restrictive in measurement locations (including
forehead, forearm, supraorbital artery, under the legs, and the wrist), greater signal
amplitude and lower motion artifact [18]. The disadvantage includes lower signal-to-
noise ratio, motion artifact, and variation due to relative distance between the light
source and the sensor caused by the shorter green wavelength that is only able to
penetrate the skin and reflects skin blood flow rather than deeper larger arteries.
The transmission-mode PPG measurements are limited in locations: such as the
earlobe, fingertip, and toe for maximizing the signal-to-noise ratio. Infrared (IR)
and red optical wavelengths are widely used for transmission-mode PPG due to
higher tissue penetration depth. IR is less sensitive to the oxygen content of hemo-
globin and thus yields waveforms that are more stable over time. Transmission-
mode PPG at the fingertip has been the most widely used method for obtaining the
distal waveform.
Bioimpedance method Bioimpedance can be measured using skin surface elec-
trode pairs placed on skin surface at both proximal and distal sites. Micro amp level
electrical current with high frequency is applied onto the outer electrodes; and the
resultant differential voltage is measured across the inner electrodes. The measured
differential voltage has both an AC component for the pulsatile blood and the DC
component for other stable tissue components (bone, fat, muscle, and interstitial
fluid). For PTT measurements, bioimpedance plethysmography is used with four
electrodes in two pairs, positioned locally along the same artery in the limb (such as
the lower leg or forearm) to measure the differential voltage when blood is flowing
from one set of electrodes to the other [19, 20].
ECG method This method provides a time reference with ECG, an easy to mea-
sure and resistant to noise and artifact. The time delay between the ECG waveform
and a distal arterial waveform is the pulse arrival time (PAT). The advantage is, ECG
is easy to measure, but the disadvantage is, PAT does not have as tight relationship
to BP as PTT, due to the pre-ejection period PEP [21].
Ballistocardiography (BCG) method It measures the reaction forces of the body in
response to cardiac ejection of blood into the aorta using daily common objects such
as chairs, beds, weighing scales, and on-body accelerometers [22]. BCG provides a
proximal waveform, but it can be applied at a distal location, such as at the feet with a
weighing scale or at the wrist with an accelerometer.
Video plethysmography (VPG) method A noncontact method uses camera such
as that on a smartphone to measure arterial waveforms from the skin [23]. It uses
ambient/external light serving as the excitation source. Both proximal and distal
waveforms can be obtained from two sites, such as the face, finger, or hand. Infrared
thermal imaging with a highly sensitive camera can be used to measure arterial
waveforms from the skin, based on the principle that the measured skin temperature
changes with pulsatile blood flow due to heat exchange between vessels and sur-
rounding tissue.
The actual measurement involves following measurement steps:
(a) Sampling and filtering the proximal and distal waveforms.
(b) Detecting the beats in the waveforms.
(c) Detecting the feet or other features within the beats.
(d) Calculating PTT as the time delay between the features. Handling artifacts in
the waveforms (e.g., due to motion) is also crucial in practice yet is often not
mentioned.
Calibration of PTT to BP
PTT is a relative parameter that correlates with blood pressure. It needs calibration
to have absolute value of blood pressure. Calibration can be done with automatic
blood pressure monitors. Calibration process is to set up the necessary parameters
for the model to establish one to one mapping relationship between the actual BP
readings and the PTT measurements.
There are several aspects of calibration: the first is how to calibrate, the second
one is how frequent to calibrate, and lastly is if the calibration can be done by popu-
lation instead of by individuals.
The goal is to construct a calibration curve to map PTT measurement to absolute
value of BP readings. The actual steps of calibration involve the following steps:
1. Define a mathematical model to relate PTT to BP, either using physical models
or empirical regression models. Most of the physical models are based on the
Moens–Korteweg and Bramwell–Hill equations with an assumed function to
relate the elastic modulus or compliance to BP.
2. Measure multiple pairs of PTT and BP values from a subject during interven-
tions that change the BP in a wide range. Commonly employed interventions to
change BP include exercise (e.g., climbing steps, cycling on an ergometer), pos-
tural changes (i.e., seated, supine, standing), sustained handgrip, and the Valsalva
maneuver. Large BP changes can be achieved using anesthesia induction, sur-
gery, and ICU therapies but are limited to hospitalized subjects and are therefore
not valid for chronic hypertension management.
70 J. Li
3. Estimate the parameters by fitting the model to the PTT-BP multiple measure-
ments. The accuracy of the parameter estimation generally improves with the
ratio of the number of data pairs to the number of parameters. Least squares
regression has been commonly used for parameter estimation.
Calibration Frequency The calibration curve can be constructed either one-time
at the beginning, or periodically throughout a period [24, 25]. For the studies require
periodic calibration, the period between calibrations was within 2 h. Such frequent
calibration is to account for body system changes such as vascular tone change or
smooth muscle contraction.
Shorter PTT is directly linked to the increased arterial stiffness, which is driven
by either higher sympathetic nerve activity or elevated blood pressure, so the sym-
pathetic nerve activity can also change the individual’s arterial stiffness “black
box,” and therefore such change will require the calibration again.
Besides individualized calibration, generalized calibration approach can also be
utilized. The process is to use a population average value for one model parameter
while estimating the other parameter from cuff BP measurement [26]. The advan-
tage of this approach is to avoid BP perturbation. The disadvantage of using popula-
tion averages is the less accurate BP values. To implement this method, the following
steps would have to be performed: (a) collection of training data comprising pairs of
PTT estimates and BP values during a set of BP varying interventions per subject
from a vast number of diverse subjects; (b) estimation of the parameters of a calibra-
tion model for each subject; and (c) regression of these parameters on simple sub-
ject information. Collecting the necessary training data is a serious endeavor but
may be the best way to popularize the PTT-based BP monitoring approach.
There are challenges to have independent determination of systolic and diastolic
BP, if these two BP values do not vary in the same direction (e.g., isolated systolic
hypertension). This problem could be addressed by including additional simple
covariates, such as heart rate, in the calibration model or by estimating multiple
PTT values per beat via arterial modeling. Note that PTT estimates should correlate
better with BP values than these covariates do to offer any real value. Limited by
length and depth, this chapter is just aimed to serve as a general overview of key
elements in PWV or PTT related cuffless continuous BP monitoring. Interested
readers who want to study on mathematical models, calibration, and practical
approaches should further reference to relevant literature in this field [27–29].
The Future of Technology
With the clear advantage of this PWV or PTT based technology being less obstruc-
tive and therefore continuous, there are three potential applications for such tech-
nology to play meaningful roles in the real world:
1. Continuous BP monitoring—This is the most active field in which lots of startup

companies or research interests reside. In order to provide BP readings on a con-
tinuous basis, the challenge is all around the calibration—specifically, how to
improve calibration of PTT to BP accuracy, how to make calibration simple and
easy, and how to make calibration less frequently are among the bottlenecks that
could present as the biggest hurdle for PWV/PTT based technology become
reality. One potential method is the universal calibration wherein the parameters
of the model relating PTT to BP are determined simply from the subject’s age,
gender, and other such information including cardiovascular risk factors.
Another approach is to involve more of an individual’s biometrics parameters
and use machine learning to improve the predictive power of future BP.
2. BP change tracker—This application will reflect the BP changes, without giving
out the absolute BP value. Such BP changes are in mmHg unit, and such changes
could be used to guide individuals to take appropriate measures.
3. Trigger for absolute BP measurement—In this application, the PTT device
serves as a trigger point based on individualized setup. Once BP changes moni-
tored by the PTT technology reach a preset value, the trigger will alert the indi-
viduals to take a BP measurement using traditional cuff-based devices. Even
though the PWV or PTT based technology is not the main unit for BP measure-
ment, such technology provide a quite user friendly setting with continuous pro-
tection against elevated BP level.
Machine learning Nowadays artificial intelligence through machine learning has
been used in research for BP prediction and cuffless BP measurement. The general
approach is to initially extract surrogate cardiovascular indexes from physiological
signals, and then to use machine learning to train and adapt to the model, and finally
to predict BP using the trained model.
Key features could be extracted from physiological signals such as ECG, PPG,
activity, or sleep using either time domain or frequency domain. A feature selection
method was used to remove irrelevant or redundant features to avoid over fitting.
Machine learning methods such as linear regression, neural network, and Bayesian
network can be used to establish the BP prediction model.
Chiang et al. used Random Forest with Feature Selection to enhance the perfor-
mance of the BP prediction by filtering out unnecessary features from wearable
sensors signals and past BP readings [30]. Xing et al. used fast Fourier transform on
the PPG signal to extract the amplitude and frequency features to train an artificial
neural network to estimate BP [31]. Kachuee et al. extracted multiple physiological
parameters from ECG and PPG signals using multiple regression [32].
Future machine learning and math modeling—select optimal features that can
best contribute to dynamic BP changes and then combine the physiological and
mathematical modeling to predict continuous BP noninvasively and continuously.
Considering the complexity of the cardiovascular system, selection of multiple indi-
cators and an appropriate model is critical and requires full-system integration to
ensure the accuracy of indirect measurements.
72 J. Li
References
1. Pressman G, et al. A transducer for the continuous external measurement of arterial blood pres-
sure. IEEE Trans Biomed Electron. 1963;10:73–81.
2. Peñáz J. Photoelectric measurement of blood pressure, volume and flow in the finger. In:
Digest of the 10th International Conference on Medical and Biological Engineering. 1973.
3. Weltman G, et al. The continuous measurement of arterial pulse wave velocity. Med Electron
Biol Eng. 1964;2:145–54.
4. Kortekaas MC, et al. Small intra-individual variability of the preejection period justifies the use
of pulse transit time as approximation of the vascular transit. PLoS One. 2018;13(10):e0204105.
5. Peter L, et al. A review of methods for non-invasive and continuous blood pressure monitoring:
pulse transit time method is promising? IRBM. 2014;35:271–82.
6. Fung P, et al. Continuous noninvasive blood pressure measurement by pulse transit time. Conf
Proc IEEE Eng Med Biol Soc. 2004;1:738–41.
7. Nichols WW, et al. McDonald’s blood flow in arteries: theoretical, experimental and clinical
principles, 3rd ed. J Cardiopulm Rehabil Prev. 1991;11:407.
8. Proença J, et al. Is pulse transit time a good indicator of blood pressure changes during short
physical exercise in a young population? In: Proceedings of the 2010 Annual International
Conference of the IEEE Engineering in Medicine and Biology (EMBC), Buenos Aires,
Argentina. 31 Aug–4 Sept 2010. p. 598–601.
9. Geddes L, et al. Pulse transit time as an indicator of arterial blood pressure. Psychophysiology.
1981;18:71–4.
10. Wong MY, et al. An evaluation of the cuffless blood pressure estimation based on pulse transit
time technique: a half year study on normotensive subjects. Cardiovasc Eng. 2009;9:32–8.
11. Masè M, et al. Feasibility of cuff-free measurement of systolic and diastolic arterial blood
pressure. J Electrocardiol. 2011;44:201–7.
12. Baek HJ, et al. Enhancing the estimation of blood pressure using pulse arrival time and two
confounding factors. Physiol Meas. 2009;31:145–57.
13. Marcinkevics Z, et al. Relationship between arterial pressure and pulse wave velocity using
photoplethysmography during the post-exercise recovery period. Acta Univ Latv Biol.
2009;753:59–68.
14. Longo A, et al. Posture changes and subfoveal choroidal blood flow. Investig Ophthalmol Vis
Sci. 2004;45:546–51.
15. Parati G, et al. Comparison of finger and intra-arterial blood pressure monitoring at rest and
during laboratory testing. Hypertension. 1989;13:647–55.
16. Steptoe A, et al. Pulse wave velocity and blood pressure change: calibration and applications.
Psychophysiology. 1976;13:488–93.
17. Shelley KH. Photoplethysmography: beyond the calculation of arterial oxygen saturation and
heart rate. Anesth Analg. 2007;105:S31–6.
18. Maeda Y, et al. Relationship between measurement site and motion artifacts in wearable
reflected photoplethysmography. J Med Syst. 2011;35:969–76.
19. Risacher F, et al. Impedance plethysmography for the evaluation of pulse-wave velocity in
limbs. Med Biol Eng Comput. 1993;31:318–22.
20. Critchley LA. Impedance cardiography. The impact of new technology. Anaesthesia.

1998;53:677–84.
21. Geddes LA, et al. Pulse arrival time as a method of obtaining systolic and diastolic blood pres-
sure indirectly. Med Biol Eng Comput. 1981;19:671–2.
22. Inan OT, et al. Ballistocardiography and seismocardiography: a review of recent advances.
IEEE J Biomed Health Inform. 2015;19(4):1414–27.
23. Yoshizawa M, et al. A great impact of green video signals on tele-healthcare in daily life,
especially for rural or disaster areas. Trans Jpn Soc Med Biol Eng. 2013;51:M-55.
24. Chen W, et al. Continuous estimation of systolic blood pressure using the pulse arrival time
and intermittent calibration. Med Biol Eng Comput. 2000;38:569–74.
25. McCarthy BM, et al. An investigation of pulse transit time as a non-invasive blood pressure
measurement method. J Phys Conf Ser. 2011;307:012060.
26. McCarthy BM, et al. An examination of calibration intervals required for accurately tracking
blood pressure using pulse transit time algorithms. J Hum Hypertens. 2013;27:744–50.
2015;62(8):1879–901.
28. Sharma M, Barbosa K, Ho V, Griggs D, Ghirmai T, Krishnan SK, Hsiai TK, Chiao JC, Cao H.
Cuff-less and continuous blood pressure monitoring: a methodological review. Technologies.
2017;5:21.
29. Ding XR, et al. Continuous blood pressure measurement from invasive to unobtrusive:

Celebration of 200th birth anniversary of Carl Ludwig. IEEE J Biomed Health Inform.
2016;20(6):1455–65.
30. Chiang PH, et al. Personalized effect of health behavior on blood pressure: machine learn-
ing based prediction and recommendation. In: IEEE 20th International Conference on
e-Health Networking, Applications and Services (Healthcom). 2018. https://doi.org/10.1109/
HealthCom.2018.8531109.
31. Xing X, et al. Optical blood pressure estimation with photoplethysmography and FFT-based
neural networks. Biomed Opt Express. 2016;7:3007–20.
32. Kachuee M, et al. Cuff-less high-accuracy calibration-free blood pressure estimation using
pulse transit time. In: 2015 IEEE International Symposium on Circuits and Systems (ISCAS).
2015. p. 1006–9.
Chapter 7
Pulse Decomposition Analysis Techniques
Martin C. Baruch
Abstract Pulse decomposition analysis (PDA) uses a pulse contour analysis

approach to quantify hemodynamic parameters such as blood pressure and arterial
tone changes. It is based on the concept that two central reflection sites are respon-
sible for the shape of the pressure pulse envelope of the upper body.
The two reflection sites, one located at the aortic juncture of thoracic and abdom-
inal aortas, and the other at the iliac bifurcation, reflect the primary left ventricular
ejection pulse to give rise to two reflected and two re-reflected component pulses.
Within the pulse pressure envelope of each cardiac cycle these five component
pulses arrive sequentially in the arterial periphery. Quantification of the temporal
and amplitudinal behavior of the first three component pulses establishes a formal-
ism that can be used to monitor certain hemodynamic states and their changes.
The observational evidence and motivation for PDA are presented, as are pulse
modeling approaches, practical implementation considerations and physiological
confounders. Benchmark and clinical study comparisons are provided. The current
status and outlook of the CareTaker physiological monitor, which utilizes PDA as
its operational principle and has demonstrated compliance with several regulatory
standards, are described.
Keywords Noninvasive · Continuous blood pressure · Arterial reflections ·

Finger cuff · CareTaker
Pulse Decomposition Analysis
Pulse decomposition analysis (PDA) uses a pulse contour analysis approach to

quantify hemodynamic parameters such as blood pressure and arterial tone changes.
It is based on the concept that central reflection sites, as opposed to distal sites in the
arterial periphery, are primarily responsible for the shape of the pressure pulse
M. C. Baruch (*)
Caretaker Medical LLC, Charlottesville, VA, USA
e-mail: martin@caretakermedical.net

76 M. C. Baruch
envelope of the upper body. Specifically, PDA postulates that five individual com-
ponent pulses give rise to the observed pulse shape. The first of these component
pulses to arrive in the arterial periphery is the left ventricular ejection pulse which
is then followed by reflections and re-reflections of the ejection pulse from two
central arteries reflection sites. PDA further postulates that the quantification of the
temporal and amplitudinal behavior of these component pulses gives rise to a for-
malism that can be used to monitor certain hemodynamic parameters and their
changes. PDA has cleared several practical implementation stages and is the opera-
tional principle of the CareTaker physiological monitor, which has demonstrated
compliance with several regulatory standards.
Underlying Considerations
The existence and the physiological consequences of reflections in the arterial tree
are now commonly accepted [1–4]. In this physiological model the arterial pressure
pulse originates from the left ventricle and travels away from the heart through the
arterial tree and is reflected at sites where the arterial tree branches or different
diameter sections join, since these sites present an impedance mismatch to the prop-
agating arterial pressure pulse.
Clinical studies and theoretical modeling efforts have investigated various
aspects of arterial pulse reflections, such as the “second systolic peak”, yet no clear
model has been proposed specifying where exactly the reflections arise. For exam-
ple, an asymmetric T-shaped model [5] has been proposed where the pulse origi-
nates at the T junction from the heart and the ends of the T represent generalized
reflection sites of the lower body and the upper body.
Similarly it has been proposed that the principal mechanism giving rise to reflec-
tions in the arterial tree are the various artery/arteriole interfaces throughout the
body, since these sites, characterized by significant lumen changes and therefore
impedance mismatches, will give rise to reflected pressure pulses that, counter prop-
agating, will return from the arterial periphery.
Evidence of Central Reflection Sites
In contrast to these models proposing distributed reflection sites, Pulse

Decomposition Analysis (PDA) is based on the concept that distinct reflection sites
dominate the shape of the arterial pressure pulse envelope, resulting in a superposi-
tion of distinct component pulses. Focusing on the structure of the digital or the
radial arterial pressure pulse, because these are usually the clinically most relevant
monitoring sites, the component pulses that are the temporal features of the arterial
pulse envelope that follow the primary left ventricular ejection pulse are reflections
arising from reflection sites in the core arteries, specifically the junction of thoracic
7 Pulse Decomposition Analysis Techniques 77
and abdominal aortas, and the interface between abdominal aorta and the common
iliac arteries.
Why is it reasonable to assume that there are distinct reflection sites in the arte-
rial tree as opposed to the assumption that, as an alternative scenario, “the lower
body” as a whole gives rise to reflections? The answer is twofold. One is that the
features of the reflected wave are too distinct as to be the convolution of different
reflections originating from different sites with different time delays and different
amplitudes, which would tend to broaden out specific pulse features. The second
answer is that the arrival times, determined by well-known arterial pulse propaga-
tion velocities, of the specific features of the radial pulse very much narrow the
location possibilities of the reflection sites.
Figure 7.1 presents radial pulse signatures collected from different individuals of
different ages. Both traces exhibit pulse-like protrusions (black and red arrows) that
have a time duration comparable to that of the primary pulse (blue arrows). Data
that clarifies this point is presented in Fig. 7.2, which presents radial pulse data col-
lected during a Valsalva episode. One consequence of Valsalva is the shortening of
the cardiac ejection period. As a result, it is possible, in a comparatively young and
elastic arterial tree, to see the complete separation of primary pulse and reflected
pulse. Clearly the reflected pulse shows little to no broadening compared to the
primary systolic peak, supporting the hypothesis that it originated at a distinct refec-
tion site. Figure 7.3 seeks to clarify this point further. While a distinct reflection site
will give rise to a reflection bearing strong resemblance to the primary pulse, dis-
tributed and multitudinous reflection sites will give rise to a multitude of reflected
pulses, arriving at different time delays and with different amplitudes.
Fig. 7.1 Examples of high-fidelity radial arterial pulse shapes. Top: 20 y. m. athlete. Bottom: 52
y. m. catheter laboratory patient. Note the pulse-like features, indicated by black and red arrows,
following the primary ejection pulse (blue arrows), that have a similar temporal and amplitude
profile as the primary pulse
78 M. C. Baruch
Fig. 7.2 Radial pulse during onset of Valsalva maneuver. Notice the vanishing of the second
systolic pulse
Fig. 7.3 Qualitative comparison between reflected pulse (red on left resulting from distinct reflec-
tion sites (top) as compared to a reflected pulse resulting from distributed, amorphous reflection sites
The superposition of such a system of reflection sites would result in a feature-

less, broadened pulse. The presence of distinct pulse-like features in the radial sig-
natures shown therefore suggests that, past the primary systolic peak, distinct
reflection sites are responsible for the sequence of reflected pulses comprising the
“diastolic wave.”
While the presence of distinct pulse-like features in the radial pulse suggests the
existence of specific and powerful reflection sites, their time of arrival relative to the
primary pulse makes the argument significantly more concrete. Figure 7.4 presents
an example of the radial pulse of a 44 year old male as well as the time intervals
between its various component pulses. The first timing issue worth considering is to
what degree the pulse features are influenced by the geometry of the arm, that is,
could one of the pulse features observed be due to a reflection site in the arm?
Arterial pulse velocities in the radial artery are on the order of 7–8 m/s. Since the
pulse signal is collected at the wrist, the distance from that site to a site of a potential
reflection, the interface between artery and arterioles at the wrist, is on the order of
centimeters. Therefore, the reflection would return in a matter of a few milliseconds,
as indicated in Fig. 7.4 by a blue line in the immediate vicinity of the primary pulse.
Since all the reflected pulse features in the radial pulse appear at far greater time
delays, as indicated in the figure, they have to originate elsewhere in the arterial tree.
Second Systolic Peak
Diastolic Peak
"Radial Reflection"
(not visible)
115 ms
172 ms
193 ms 168 ms
121 ms
0.2 Seconds
Fig. 7.4 Distinct pulse structure in the radial arterial pulse of a 44 y. male
Since arterial pulse propagation velocities are well known, it is possible to match
time delays with potential reflection sites. Figure 7.5 presents a simplified sketch of
the components of the aorta and the connecting arteries of the legs and the left arm.
The sketch also lists typical arterial diameters as well as arterial pulse propagation
velocities at the different sites as published in the medical literature [1]. Using
approximate arterial distances and their respective velocities, the “second systolic”
peak matches readily with the site labeled “reflection site I” while the third peak
matches with “reflection site II.” Work by others supports these conclusions [2–4].
In 1985, Latham [6] performed a detailed experimental study to map out the shape
of the pressure pulse in the different sections of the aorta using a specially designed
catheter with spaced micro manometers (Fig. 7.6). His work demonstrated the exis-
tence of two major reflection sites to the distally traveling arterial pulse, one in the
region of the renal arteries, the other beyond the bifurcation of the iliac arteries.
At the location of the renal artery the diameter of the aorta, tapering continuously
away from the heart, undergoes its greatest change at the juncture between the larger
diameter thoracic aorta and the smaller diameter abdominal aorta. This discontinu-
ity presents a significant impedance mismatch to the traveling pressure pulse, result-
ing in an appreciable part of its amplitude being reflected.
Referring back to Fig. 7.2 and the Valsalva maneuver, the phenomenological
explanation is that the maneuver reduces this reflection because, due the increasing
pressure within the thoracic cavity, the diameter of the thoracic aorta decreases
while the diameter of the abdominal aorta, which is outside the thoracic pressure
cage, does not. The maneuver therefore alleviates the aortic diameter change at the
renal arteries, reducing the impedance mismatch and thereby lowering the site’s
reflection coefficient.
80
Fig. 7.5 Sketch of the aorta/arm complex arterial system and its effect on the arterial pressure pulse line shape observed at the radial/digital artery. Two reflec-
tion sites, one at the height of the renal arteries, the other one in the vicinity of the iliac bifurcation, give rise to the reflected pulses (gray) that trail the primary
left ventricular ejection (black)
M. C. Baruch
Fig. 7.6 Arrangement of the catheter sensor positions in the aorta, with examples of pressure
waveforms from patient C. Diameter and SD values refer to elastic tube model used to simulate
observed effects. Reproduced with permission
Latham also found a second reflection site beyond the bifurcation of the iliac
arteries, the contribution of which to arterial pulse reflections in the aorta was ascer-
tained using manual femoral artery occlusion maneuvers. Other contributions to the
tail end of the aortic pulse were attributed to diffuse arterial pulse reflections from
the periphery. This, however, appears to be unlikely, given the distinct peak struc-
ture with a spacing comparable to that of the “second systolic” and the “diastolic”
peak. Furthermore, the time delay of such diffuse reflections would extend up to
250 ms past the “diastolic” peak if they truly traversed the length of the legs. Indeed,
other work by J. Kriz et al. [7] supports the hypothesis that the peaks visible past the
“diastolic” peak are in fact due to re-reflections between the two reflection sites, a
reasonable proposition given the strength of the sites’ reflection coefficients (10–
15% in the case of the renal arteries reflection site and up to 30% in the case of the
iliac arteries reflection site [8]).
The work by Kriz showed that it is possible to use force plate measurements as a
noninvasive method to perform ballistocardiography, the body’s recoil due to the
momentum generated by the heart’s activity, by displaying the motion of the heart
muscle and the subsequent propagation of the pulse wave along the aorta and its
branches. With subjects lying horizontally on a bed placed on a force plate they
were able to identify the ground reaction forces arising from such center-of-mass
82 M. C. Baruch
altering events as the heart muscle contraction as well as the resulting ejection pres-
sure pulse. The resolution of the apparatus was sufficient to clearly resolve events
involving the redirection of momentum of the propagating arterial pulse, such the
pulse’s traversal of the aortic arch, its partial reflection at the renal artery site, the
iliac reflection well as the subsequent re-reflections of the reflected pulses. As an
aside, in subjects with an aortic aneurism, the site of the arterial distention was
clearly identifiable due to its effect on the neighboring “normal” reflection sites.
The basic PDA model of the radial/digital arterial pressure pulse is therefore one
of a convolution of the primary systolic peak, its single-pass reflections from the
renal arteries and iliac arteries reflection sites, as well as their double-pass
re-reflections.
Implementation
Modeling of Pulse Reflections
The existence of two distinct central pressure pulse reflection sites make it is pos-
sible to propose a simple model of the arterial paths the primary pulse and its reflec-
tions traverse and to compare its predictions with observations regarding the relative
arrival times of the different component pulses. The model’s equations predict the
time of arrival of each individual component pulse, subject to the total distance the
pulse has traveled and the pressure-dependent pulse propagation velocity in each
arterial segment. The different relevant arterial paths are denoted by xn, where x1
refers to the arm arterial path, while x2 and x3 refer to the thoracic and abdominal
aorta, respectively. The variable tn refers to the time of arrival of the nth component
pulse at the radial/digital arterial peripheral site. While in the case of the #1 pulse its
arrival time, t1, is determined only by its travel along the arm complex arteries (x1
path), the arrival times for the #2 and #3 pulses take into account their initial travel
as the primary ejection pressure pulse as well as, after impacting a reflection site,
their subsequent return as a reflected pulse. As an example, the “second systolic”
(#2) pulse traverses the thoracic aorta at systolic pressure, traverses it again as an R2
reflection after redirection at the renal arteries reflection site (indicated as R2 of
pulse pressure plus diastolic pressure) and then enters the arm arteries where it loses
another percentage of its amplitude due to the R1 reflection coefficient that incorpo-
rates artery segment transitions, such as the aortic/subclavian junction.
The pressure dependence of the pulse propagation velocity is implemented using
the Moens–Korteweg [9] equation relating pressure and velocity, ν = √ ((hEeζP)/
(2ρα)). Its definitions are as follows: ν(P) is the velocity of the xth arterial pulse path
at the pressure P indicated. E is the Young’s modulus, α is the artery’s diameter, h is
the arterial wall thickness, ρ is the fluid density, ζ is the arterial compliance and P is
the pressure. The Young’s modulus and the arterial compliance ζ are different for the
different arterial segments.
Another significant feature of the model is that R2, the renal reflection coeffi-
cient, is dependent on pressure. The motivation for this is based on the following
consideration.
x1
t1 = , (7.1)
vx1 , P11

x2 x2 x1
t2 = + + (7.2)
vx2 , P21 vx2 , P22 vx3 , P23

x2 x3 x3 x2 x1
t3 = + + + + (7.3)
vx2 , P31 vx3 , P32 vx3 , P33 vx2 , P34 vx1 , P35

P11 = Psyst - R1 PPulse (7.4)

P21 = Psyst (7.5)

P22 = Pdiast + R2 PPulse (7.6)

P23 = Pdiast + R2 (1 - R1 ) PPulse (7.7)

P31 = Psyst (7.8)

P32 = Psyst - R2 PPulse (7.9)

P33 = Pdiast + R3 (1 - R2 ) PPulse (7.10)

P34 = Pdiast + R3 (1 - R2 ) (1 - R2 ) PPulse (7.11)

P35 = Pdiast + R3 (1 - R2 ) (1 - R2 ) (1 - R1 ) PPulse (7.12)

As discussed, the renal reflection (P2 pulse) originates at the junction between tho-
racic and abdominal aorta, a junction characterized by a significant change in arte-
rial diameter. Since the thoracic aorta is the softest artery in the body, as evidenced
by the lowest pulse pressure propagation velocities (4–5 m/s) and much more exten-
sible than the abdominal aorta, increasing pressure will enlarge the diameter mis-
match, giving rise to a more pronounced renal reflection pulse amplitude while
falling pressure will produce the opposite effect, an effect observed in manipulative
experiments performed by Latham. The central insight then is that the amplitude of
the renal reflection will increase relative to the amplitude of the primary systolic (P1
pulse) peak because, while both component pulses travel the arteries of the arm
complex, and are therefore both subject to the pulse narrowing and heightening due
to the taper and wall composition changes of the peripheral arteries, only the renal
84 M. C. Baruch
reflection will have sampled the pressure-induced aortic impedance mismatch

changes. This provides the motivation for taking the ratio of the amplitudes of the
#2 and the #1 pulse, which is the PDA parameter P2P1.
These considerations are put in context given the different response characteris-
tics of central versus peripheral arteries that have been reported and discussed by
others. Specifically, the fact that central arterial elasticity is determined by BP and
not smooth muscle contraction, in contrast to peripheral arteries, provides a physi-
ological explanation for the effects that are quantified and utilized as part of the
PDA formalism [10, 11].
Preliminary tests involving a fit middle-age male subject demonstrate that this
comparatively simple model is able to adequately predict the arrival times of the
three primary component pulses during a maneuver such as Valsalva. Figure 7.7a–c,
which present predicted and measured time delay curves for the three primary
0.250
Pulse #1
Delay TIme (Seconds)
0.225
0.200
0.175
0.150
A
0.125
0.38
0 20 40 60 80 100
Delay Time (Seconds)
0.36
Pulse #2
0.34
0.32
0.30
0.28
0.26 B
0.24
0.56
Delay Time (Seconds)
0 20 40 60 80 100
0.54
Pulse #3
0.52
0.50
0.48
0.46
0.44
0.42
0.40 C
0.38
0.36
0 20 40 60 80 100
Time (Seconds)
Fig. 7.7 (a, b, c) Relative overlap of delay times of the three primary pulses measured (red) and
obtained using the model (black) with the diastolic and systolic blood pressures obtained from the
Colin-Pilot, clinical monitor of noninvasive continuous blood pressure
pulses, give a sense of the agreement between the two. The predicted delay time
values were obtained by isolating the diastolic and systolic peak to peak blood pres-
sure values obtained from a continuous radial artery tonometer (Colin Pilot) and
inserting these values into the PDA model. The timing of the individual component
pulses was obtained using the QRS complex of a simultaneously obtained ECG
signal as a starting signal. The agreement of the range of delay time values is no
surprise since correlations were used to relate the blood pressures measured with
the Colin unit to the measured pulse delay times. Encouraging is the fact that the
overall time evolution of the predicted and measured delay times agrees well.
In order to arrive at the above results, the pressure/velocity response curve for
each of the three primary pulses had to be quantified by correlating the systolic and
diastolic blood pressures measured with the tonometer with the delay times of the
three primary pulses. In addition, the BP response behavior of the renal reflection
coefficient R2 has to be quantified. The resulting fitted functions are displayed in
Fig. 7.8 for the velocity responses of the different arterial sections and in Fig. 7.9 for
the pressure response of reflection coefficient. The difference in velocity response,
and therefore time delay response, between the different pulses is significant. While
the arm complex displays an exponential response, thoracic and abdominal veloci-
ties follow more linear relationships. The pulse propagation velocity of the abdomi-
nal aorta region exceeds that of the arm arteries, comparable to results published in
the medical literature. Of course, this result holds for subjects with “elastic” arter-
ies. Results are quite different for patients with increased arterial stiffness, where
arm pulse propagation velocities can reach 15 m/s.
9
Arm complex
Adominal aorta
Thoracic aorta
8
Velocity (m/s)
100 150 200

Pressure (mmHg)
Fig. 7.8 Arterial pulse velocity profiles as a function of pressure, based on fit to experimental
results in Fig. 7.6
86 M. C. Baruch
0.55
Reflection coefficient (unitless)
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
75 100 125 150 175
Pressure (mmHg)
Fig. 7.9 Pressure response of R2, reflection coefficient of the junction of thoracic/abdominal aor-
tic sections at the height of the renal arteries
The behavior of the three pulses is summarized in Fig. 7.10. The figure, which
for graphical clarity inverts dependent and independent variables, summarizes the
arrival time response, along the abscissa, of the individual component pulses as a
function of varying arterial blood pressure and the correspondingly varying pulse
propagation velocity, along the ordinate. Specifically, as blood pressure rises, so
does the pulse propagation velocity. However, while in the diastolic regime the
velocity increase is approximately linear with a linear pressure increase, the systolic
regime is characterized by an exponential velocity increase response [12].
Commensurate with increasing blood pressure and increasing propagation velocity
is a decreasing arrival time in the arterial periphery. Individual component pulses
sample different sections of the velocity response curve depending on their pressure
amplitude. Since the response curve is nonlinear, the different component pressure
amplitudes give rise to different velocity variations between the different compo-
nent pulses, that is, not only will the pulse envelope accelerate as blood pressure
varies, but its components will do so relative to each other, changing the envelope of
the process. Specifically, while the #1 pulse samples the top of the systolic pressure
regime throughout its travel along the arterial tree to the radial pulse site, the #2 and
#3 pulses do so only on the initial traversal of sections of the aorta, with a much
greater part of their propagation time in significantly lower blood pressure ranges.
As a result the exponential pressure/velocity relationship that governs their travel as
outward bound primary pulses is masked by the linear pressure/velocity relation-
ship that governs their travel as reflected pulses. More importantly, differential
changes in travel time between the different pulses can be resolved because of the
Fig. 7.10 Effect of small pressure variations on the propagation velocity of the three primary
component pulses. While the #1 pulse samples the nonlinear section of the response curve, the #2
and #3 pulse responses are essentially linear while the traverse the arterial system as reflections
different functional forms and gains of the velocity curves that govern the propaga-
tion of the different component pulses.
While the timing considerations outlined above lend more qualitative credence
to the approach, their relevance in the context of obtaining hemodynamic informa-
tion through pulse wave analysis is somewhat limited because the external timer
start, which in the above experiments was the ECG’s QRS peak, is usually not avail-
able. That leaves only relative timing determinations between the component pulses
of a given pulse envelope which yield significantly less information because, as the
Valsalva example above made clear, the component pulses display similar delay
time evolutions, making their differential determination more difficult. In addition,
detection of particularly the renal reflection can be challenging because of its highly
dynamic amplitude response to blood pressure changes and the fact that it is the
component pulse most prone to be obscured by the smoothing arterial pulse shape
changes associated with stiffening arterial walls, a point that is discussed more in-
depth in the following section.
Modeling of Pulses
Just as is the case in the timing considerations above, a comparatively simple model
can be used to generate the peripheral arterial pressure pulse envelopes of the upper
body that are encountered clinically. Specifically, the triple overlap of a generalized
asymmetric exponential function of the following form,
88 M. C. Baruch
1
1- æ w1 ö
ç - x - xc + ÷
è 2 ø
A* 1+ e w3
(7.13)
æ w1 ö
ç - x - xc + ÷
è 2 ø
1+ e w2

can generate pulse envelopes that bear close resemblance to pulse shapes ranging
from those recorded on young athletes to shapes associated with the arterial wall
compliance and cardiac timing changes associated with more advanced age
(Figs. 7.11 and 7.12, below which the parameters to generate the simulated curves
are provided.). The individual component pulse in each case is modeled using an
asymmetric line shape that is characterized by a fast onset and a significantly slower
decay associated with peripheral resistance. Remarkably close pressure pulse enve-
lope representations can be generated by simply adjusting the amplitudes and delays
of the otherwise identical component pulses.
While this type of modeling further supports the underlying PDA hypothesis and
could potentially be useful in the short-term analysis of patient data with the goal of
characterizing arterial wall health, it is impractical for the implementation of hemo-
dynamic monitoring on a continuous, beat by beat basis. Since the procedure is a
0.6
0.4
0.2
0
0 100 200 300 400 500 600 700
-4
10
2
-2
-4
0 100 200 300 400 500 600 700
Fig. 7.11 Simulation of an arterial pressure pulse of a younger person with flexible arteries using
Eq. (7.13) (top graph). Second derivative of the envelope is presented in the lower graph. Front
rise = 0.4; backend = 1.5; backend3 = 2.3; delay1 = −2.0; ampl1 = 1.6; delay2 = −4.4; ampl2 = 0.3;
delay3 = −7.0; ampl3 = 0.55; w1 = 0.5
0.8
0.6
0.4
0.2
0
0 100 200 300 400 500 600 700
-4
10
2
-2
-4
0 100 200 300 400 500 600 700
Fig. 7.12 Simulation of an arterial pressure pulse of an older person. Note the decreased delay
times and enhanced renal reflection. Front rise = 0.4; backend = 1.5; backend3 = 2.3; delay1 = −2.0;
ampl1 = 1.6; delay2 = −3.4; ampl2 = 1.1; delay3 = −4.9; ampl3 = 0.5; w1 = 0.5
multiparameter fit, the likelihood for an optimization algorithm to arrive at a non-

sensical but optimized line shape is very high, and the computational load will
become very significant as heart rates increase.
The approach can, however, be used to guide a real-time pulse analysis approach
where the aim is to examine sections of the pulse envelope or its derivatives and to
track their evolution as hemodynamic changes such as that in blood pressure and
heart rate occur. Other groups have performed similar simulations, with three and
more underlying component pulses in the context of, for example, extracting infor-
mation about cardiovascular function [13].
In what follows the principal physiological confounders affecting pulse analysis
are discussed before giving an overview of the actual implementation issues.
Physiological Confounders
A real-time pulse analysis approach has to be able to accommodate the different

pulse shapes encountered across a wide patient spectrum. Long-term pulse shape
changes that are age- and disease-related and arise as cardiac function and arterial
wall structures are altered have been studied and modeled extensively, for example
via the augmentation index [14]. Less well understood are the shorter-term effects
90 M. C. Baruch
that can modify the arterial pulse envelope significantly in a timeframe of minutes
and severely compromise a previous blood pressure calibration of pulse
parameters.
Based on our research the principal short-term physiological confounders are
arterial stiffness, heart rate and left ventricular ejection time (LVET) changes, as
well as peripheral resistance changes. These confounders will be discussed next.
Arterial Stiffness Considerations
An observational fact that distinguishes arterial pressure pulses associated with stiff
arteries from those associated with flexible arteries is the degree of features, or, in
the context of PDA, the degree to which the component pulses are resolvable in the
pressure pulse envelope. While pressure pulses recorded on flexible arteries have
visually readily assessed distinct features, those recorded from stiffer arteries show
fewer or more rounded features, or none at all. See Figs. 7.13, 7.14 and 7.15.
While the above examples are due to long-term changes in the arterial wall struc-
ture, equally significant pulse shape changes due to, for example, vasodilation can
be observed on much shorter time frames. Figures 7.16 and 7.17 display examples
of vasodilation in different context which, in both cases, significantly modify the
feature profile of the original pulse.
Fig. 7.13 Young athlete:

Location of P1 is indicated
by the vertical line at about
100. The location of P2 is
indicated by the two short
vertical lines bracketing
200. For this individual, P1
and P2 are clearly resolved
Fig. 7.14 53 y. m. cath lab

patient: P1 is indicated by
the vertical line at 130,
while P2 is indicated by
the two short vertical lines
bracketing 200. P2,
following closely behind
P1, has merged with P1 as
an additive shoulder
Fig. 7.15 67 y m.
pancreaticoduodenectomy
surgery patient. P1 and P2
have essentially merged.
Even the incisura in front
of P3, indicated by the
vertical line at 310, is
essentially
indistinguishable
4 4 P1
10 10
6 P1 P2 4
P2
P3 3
4 P3
2
Value
Value
2
1
0
0
-2 -1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 0.2 0.4 0.6 0.8 1 1.2
Time (s) Time (s)
Fig. 7.16 Digital pulse of a 58 y. m. prior to (left) and 30 s after ingestion of red wine. Note redis-
tribution of amplitudes as well as the more pronounced inversion
4
10 P1 P2 10000 P1
6
P2
P3 P3
4 5000
Value
Value
2
0
0
-2 -5000
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (s)
Time (s)
Fig. 7.17 Digital pulse of a 34 y. f. Cesarean section patient prior to (left) and after (right) admin-
istration of spinal anesthesia. Aside from the redistribution of component pulse amplitudes the
three primary individual component pulses are now resolved (note one inversion on the left, two on
the right) as the entire pulse envelope has lengthened with a decrease of arterial pulse propagation
velocities (note the time shift to the right of P3), resolving the second systolic peak P2 (renal
reflection) that was essentially indistinguishable before
Part of the PDA framework is an arterial stiffness parameter (AS) which quanti-
fies the spectral content of the arterial pressure pulse that is due to the component
pulses. The featuredness spectral content in turn is related to arterial stiffness as it is
the mechanical filtering of the arterial wall that determines the extent to which the
structure of the component pulses is resolved noninvasively. As an aside, because
the featuredness spectral content of the pulse envelope is driven significantly by the
resolution of the region of overlap of P2 and P3, respectively the renal and the iliac
reflection pulses, the AS factor incorporates the pulse region of a parameter intro-
92 M. C. Baruch
35
30
a
25
20
15
10
5
0
–5
d
–10 P3
–15
P2
–20
P1
–25
–30
0 100 200 300 400 500 600
Fig. 7.18 Calculation of AS factor in the second derivative and the d/a ratio introduced by
Takazawa
duced by Takazawa that he was able to link to arterial stiffness on the evidence of
vasodilator/pressor studies as well as demographic characteristics as part of an epi-
demiological study [15]. Similarly, preliminary validation tests indicate that the AS
parameter tracks expected trends after the introduction of vasoactive agents as well
as age-related population trends.
The AS parameter is calculated using the numerical integral over the shaded area
of the second derivative of the arterial pressure pulse profile (Fig. 7.18). The second
derivative analysis approach provides better resolution in the identification of the
component pulses, that is, the left ventricular ejection pulse (P1), the renal reflection
pulse (P2) and the iliac reflection pulse (P3), in the pulse envelope. Also indicated
in the figure are the constituents of the d/a ratio that was introduced by Takazawa,
who labeled all of the inversions of the second derivative and then performed cor-
relation studies between different ratios of these inversions and clinically relevant
parameters. Both the AS integral and the d/a ratio resolve the region of the pulse
envelope most affected by arterial stiffness changes, the overlap region of the P2
and P3. Since these reflection pulses are significantly weaker in pressure amplitude
than the primary ejection pulse, the mechanical filtering of the arterial wall will
have the largest effect here, obscuring the overlap as the wall stiffens and resolving
if it is in a more compliant state. The benefit of integration is robustness, because
often the identification of distinct inversions is difficult or impossible due to high
heart rates or the smoothing effect of extremely stiff arteries.
As Figs. 7.16 and 7.17 suggest, arterial stiffness changes are significant in the
context of pulse analysis for the purpose of tracking blood pressure. In the case of
PDA the differential blood pressure response of central and peripheral arteries,
which the ratio P2P1 represents, will be affected as the changed mechanical filtering
of the pulse effects the amplitude and temporal distribution of the component pulses.
Specifically, an increase in arterial stiffness will tend to increase the ratio while a
decrease will have the opposite effect, with the possibility of overestimating a paral-
lel trend or masking an opposite trend in blood pressure.
Heart Rate Considerations
Heart rate changes will affect primarily the timing aspects of pulse analysis, as the
width of the component pulses changes with LVET, correspondingly shifting asso-
ciated fiduciary markers. This is particularly a problem as heart rates approach and
exceed approximately 120 bpm because, in addition to the changes in LVET, the
average stress state of the arterial wall increases since the time duration between
cardiac cycles is too short for the wall to completely relax, causing pulse propaga-
tion velocities to rise [16].
The overall effect is that the features being tracked on the pulse envelope will
shift temporally relative to each other, with time intervals narrowing for rising heart
rates and extending in the opposite case. In the case of stationary search windows,
the need for shifting them in response to changed heart rates clearly has to be taken
into account. In the case where discrete fiduciary points are tracked, such as for
example certain inversions in the derivatives of the pulse envelope, the problem is
more complex. In this case inversions points that were clearly resolvable at lower
heart rates will merge with other features or lose profile amplitude, impeding
detection.
Peripheral Resistance Considerations
A discussion of the effect of peripheral resistance on the pressure pulse envelope

requires a brief clarification since the effect is sometimes invoked as being the cause
of arterial pressure reflections [17]. More current views propose that it is the periph-
eral resistance that sustains pressure once the left ventricular ejection as subsided
[18]. Consequently, it will determine the rate at which stored pressure “bleeds” into
the arterial periphery, determining the decay time on the falling side of arterial pres-
sure pulse.
The above Fig. 7.19 presents an example of the evolution of the arterial pulse of
a 35 male subject prior to, during, and shortly after a 2 min 100 W workout. While
the changes due to heart rate/LVET/cardiac output in the front section of the pulse
are obvious, the collapse of the reflected pulses on the falling side of the pulse enve-
lope are also clearly evident as the corresponding reflection sites diminish to accom-
modate the increased blood flow associated with the increased peripheral muscle
oxygen requirements. The interesting aspect, considering peripheral resistance, is
the final slope at the tail end of the pulse envelope. While at rest the slope is very
94 M. C. Baruch
resting 15 seconds past

end of workout 5 min past
0.25 Seconds/div
Fig. 7.19 Evolution of the arterial pressure pulse of a 35 m from rest (left), toward the end of a
2 min 100 W workout, 15 s into recovery, and 5 min into recovery. Peripheral resistance changes
in the tail end of the pulse are visible. Vertical red lines indicate, from left to right, the temporal
positions of P1, P2, P3, respectively left ventricular ejection, renal reflection, iliac reflection
shallow, suggesting increased peripheral resistance, the slopes at the end of the
workout and just after are much steeper, suggesting a steep pressure drop at the end
of diastole. Five minutes later, while clearly not fully recovered as indicated by the
still elevated heart rate, the slope is again dropping. Similar effects were observed
in experiments specifically designed to alter peripheral resistance pharmacologi-
cally with the simultaneous introduction of multiple agents [19]. From the point of
view of pulse analysis, extracting blood pressure during such hemodynamic changes
is one of the most challenging tasks.
Final Considerations on Implementation
This section has provided a brief overview of the challenges any blood pressure
pulse analysis approach faces. They will affect the detectability of most fiduciary
points on the pulse profile, which will vary or vanish entirely under certain circum-
stances. This applies to detection on the actual pressure pulse profile as well as any
of its derivatives, which, while sometimes able to enhance detection of curvature
inversions, can also obscure it because they are a subject to both amplitude as well
as slope changes. As a result, for example, an initially low-amplitude inversion in
the pulse envelope occurring over a short time frame will be amplified in its deriva-
tives, facilitating detection, while being diminished as the amplitude grows but the
slope change of the inversion decreases. As a result, it can in this case be easier to
detect the fiduciary point in the original pulse profile. An example is P3, the iliac
reflection, in the left and right panels of Fig. 7.17 (teal vertical line in rear section of
pulse in both cases). While in the right panel the peak of P3 is easily identified, in
the left panel a derivative has to be used in the detection. Clearly, different detection
approaches have to be used depending on circumstances. Problems arise connecting
the detection of the same fiduciary point detected in different differentiation states
because the use of derivatives, and the associated need for low-pass filtering or
smoothing, introduces time shifts in the same detected peak depending on which
differentiation state was used. These time shifts will exhibit as amplitude noise if
the threshold condition for choosing one differentiation state versus another is
crossed repeatedly.
In the case of the detection of P2 there are a number of specific difficulties. Due
to its high dynamic amplitude its temporal behavior is also dynamic, that is, it accel-
erates and decelerates relative to the P3 and P1 component pulses, as indicated in
Fig. 7.10. At the same time P1 undergoes the most significant velocity changes,
modulating the detectability of P2 significantly as it is more or less embedded in the
tail decay of P1 (see component pulse simulations in Fig. 7.11). In our experience
the most robust approaches involve using detection windows, involving calculating
the average amplitude of a window extending, for example, 40 to 140 ms past P1, to
track the P2 evolution, as opposed to attempting to detect the P2 time location
directly.
Another important consideration concerns the details of the response curve of
P2/P1. While in mid-range pressures (90–140 mmHg) the response curve is linear,
in the hypotensive and hypertensive range nonlinear corrections have to be imple-
mented based on clinical data.
Compensation for the confounder effects described above is another important
topic. In the case of AS the challenge is at least twofold; on the one hand the AS
score has to be stable if it is to be used in any compensatory calculation. Since it
involves an integration over a derivative, care has to be taken that the line shapes are
stable, that is, noise-contaminated line shapes are either rejected and/or adequately
filtered. On the other hand is the actual compensation calculation. Dividing by an
AS factor has proven most effective in mitigating the effects due to changing arterial
stiffness. Compensation for heart rate changes can be implemented by narrowing
and shifting search windows as component pulses narrow. Care has to be taken that
heart rate inputs are filtered and adjustments are very gradual to avoid introducing
additional noise.
Benchmarks and Clinical Evidence
Completed and ongoing clinical studies [20–22], both published and internal, have
sought to validate the PDA model and demonstrate accuracy. These efforts are
ongoing to further enhance and refine the approach.
96 M. C. Baruch
165 CT Systole
160
CT Systole
155
150
145
140
0 500 1000
Time (Seconds)
Fig. 7.20 Blood pressure results for a 25 y. m. to the pressor. The immersion period is indicated
by the red vertical lines. The subject’s digital pulse shape prior to and during application of the
pressor is shown in the adjacent figures
Ice Stimulus Experiments
As part of these experiments, subjects were monitored using the CareTaker™ (CT),
which is the hardware platform on which the PDA formalism has been implemented
before, during and after a 1-min ice–water immersion episode, an intervention that
brings about a temporary increase in blood pressure on the order of 15–25 mmHg in
the majority of subjects.
Figure 7.20 presents an example of the systole response during and after immer-
sion while Figs. 7.21 and 7.22 give examples of the pulse shape change prior to and
during application of the pressor. Figure 7.23 displays the blood pressure response
as well as the associated pulse shapes of a different subject. While the direction and
magnitude of the responses in both cases agree with expectations, there is a differ-
ence in the time response to the stimulus between the two subjects that has been
observed in other subjects also. The origin of the time delay in some subjects is
unclear. Given that peripheral and central blood pressure can track differently at
least temporarily—the differential treatment of central vs. peripheral hypertension
(CAFE [23] study) is an example—it is conceivable that the ice immersion stimulus
applied in the arterial periphery could elicit a delayed or, at least initially, modulated
response in the core arteries and therefore in the CT’s response characteristics.
Studies involving the simultaneous monitoring of both central and peripheral pres-
sure will be required to further investigate this observation.
Valsalva Experiments
A different set of experiments involved the investigation of the CareTaker™

response to the Valsalva maneuver,
Fig. 7.21 Pulse shape prior to application of pressor
Fig. 7.22 Pulse shape during peak of response
There are four main phases in the Valsalva maneuver [24]. In phase I, there is a
transient rise in BP due to increased intrathoracic and intra-abdominal pressure
causing mechanical compression of the aorta. In the early part of phase II, reduced
preload and reduced stroke volume lead to a fall in cardiac output. Total peripheral
resistance then increases, reversing the fall in BP to the point where, in some sub-
jects, mean arterial pressure (MAP) can be at resting MAP level or above at the
beginning of phase III. Phase III lasts a few seconds during which time BP falls due
to a sudden decrease in intra-thoracic pressure. As part of phase IV, venous return
and cardiac output return to normal while peripheral resistance remains high,
resulting in an overshoot of BP. Figure 7.24 displays results for a 57 y. m. perform-
ing a Valsalva by maintaining a pressure in excess of 40 mmHg for 20 s on the distal
side of an orifice. All four phases are resolved with expected relative amplitudes.
98 M. C. Baruch
180
Blood Pressure (mmHg)
1 Second
160 1 Second
140
1 Second 1 Second
CareTaker Systole
120
CareTaker Diastole
Omron Systole
100
1 Second Omron Diastole
80
16:28 16:30 16:32 16:34 16:36

Time (hh:mm:ss)
Fig. 7.23 Blood pressure results for a 37 y. m. with very strong response to the pressor. The
immersion period is indicated by the shaded rectangle. The subject’s digital pulse shape changed
significantly as a result of the immersion episode and the associated temporary blood pressure
increase
P2P1 parameter (unitless)
0.35
0.30
0.25
0.20
100 125 150 175 200 225
Time (Seconds)
Fig. 7.24 The four phases recorded in the P2P1 PDA parameter during a Valsalva maneuver
Clinical Comparisons
Central Arterial Line Comparisons
In these experiments, performed at the Catheterization Laboratory at the University

of Virginia Medical Center, the aortic blood pressures of 63 patients undergoing
cardiac catheterization were monitored using central line catheters while the
Fig. 7.25 Overlap of central systolic pressure (red) obtained from catheter signal and P2P1 ratio
obtained from PDA analysis of noninvasively obtained arterial signal (black) for patient 38
CareTaker™ system collected pulse line shapes at the proximal phalange of the pol-
lex and an automatic cuff determined brachial blood pressure. While the patient
rested in a supine position, the catheter was inserted into the femoral artery and
advanced toward the heart through the aorta. As part of the study the catheter was
positioned in the aorta at the height of the renal arteries for 90 s under fluoroscopy
while the catheter signal was recorded. The CareTaker™ system recorded data
throughout the preparation period as well as the 90 s overlap window. Both data
streams were time synchronized by matching the recording computer’s time as
closely as possible to the laboratory’s central time and matching the beat-to-beat
inter-beat interval variability, whose randomness provides a unique time stamped
signature. PDA parameters were then extracted, beat by beat, from the noninva-
sively collected CareTaker™ data and converted to systolic and diastolic blood
pressures for comparison with systolic and diastolic blood pressures obtained
directly from the catheter data tracings. Figure 7.25 displays an example of the
overlap of the PDA pulse parameter P2P1 and the systolic blood pressure recorded
by the central catheter, while Figs. 7.26 and 7.27 display the overall correlation
comparisons of the study.
Peripheral Arterial Line Comparisons
As part of a study comparing blood pressures measured with the CareTaker™ and
peripheral arterial line at Cooper Hospital, data from 24 adult patients requiring
hemodynamic monitoring during major open abdominal surgery were analyzed.
Patients were not excluded due to other medical conditions. Measurements were
100 M. C. Baruch
Fig. 7.26 Overall correlation of systolic blood pressures obtained through conversion of PDA
parameters from noninvasively obtained arterial pulse signal, and central systole
Fig. 7.27 Overall correlation of diastolic blood pressures obtained through conversion parameters
from noninvasively obtained arterial pulse signal, and central diastole
obtained during general anesthesia in these patients starting with induction. The
induction point was chosen because the blood pressure fluctuations and variability
typically found during this period provided an opportunity to compare tracking
accuracy under baseline and induced controlled dynamic conditions.
Figure 7.28 presents an example of an overlap lasting almost 3 h. A total of 3870
comparative data points was obtained from the A-line and CT device for the 30 min
time window comparison. For the data set collected during the entire procedure,
58,701 comparative data points were obtained, spanning approximately 114.5 h.
Across the 24 subjects, the percentage mean of excluded data was 2.8% (SD: 4.0,
range: 0–12.7%) while the median was 1.0%. The correlation between the a-line
and the CT device for MAP, systolic, and diastolic were 0.92, 0.86, and 0.91, respec-
tively (p < 0.0001 for all). The Bland–Altman comparison yielded a bias (as
measured by overall mean difference) of −0.57, −2.52, and 1.01 mmHg for systolic,
diastolic, and mean arterial pressures, respectively with a standard deviation of
7.34, 6.47, and 5.33 mmHg for systolic, diastolic, and mean arterial pressures,
respectively (p < 0.001 for all comparisons). The corresponding results for data col-
lected during the entire procedure (58,701 data points) including the 30-min study
for MAP, systolic, and diastolic were 0.87, 0.89, and 0.82, respectively (p < 0.0001
for all the comparisons). Corresponding Bland–Altman analyses for MAP, systole
and diastole yielded standard deviations of, respectively, 9.73, 13.13 and
10.23 mmHg (p<0.0001 for all values).
Fig. 7.28 The overlap of systolic and diastolic blood pressures recorded with a catheter (black,
blue) and the CareTaker™ (red, pink)
102 M. C. Baruch
Currently a number of studies are testing the performance of the CareTaker™

system in other operating room, dialysis, ER, ambulance, and other clinical
settings.
Hardware Implementation and Future Developments
The hardware platform on which the PDA formalism has been developed and
refined is the CareTaker™. The platform’s original intent was a physiological signs-
of-life assessment tool for military applications, with the original research spon-
sored by the US Army, the US Navy, DARPA, and the National Institutes of Health.
Over the course of three laboratory prototype generation developments the refine-
ment of front-end electronics and sensors, with the resulting enhanced resolution of
arterial pulse features, has expanded the possibilities of physiological modeling and
commensurately the range of measurable hemodynamic parameters.
Currently in its fourth generation, the CareTaker™ physiological monitor has
demonstrated compliance with the ANSI/AAMI/ISO 81060-2:2013 standard and
received FDA (K151499, K163255, K181196) and CE clearances for the noninva-
sive and continuous monitoring of blood pressure, heart rate and respiration rate.
In its current implementation the CareTaker™ is a physiological sensing system
that communicates wirelessly via Bluetooth. The device uses a low pressure [35–
45 mmHg], pump-inflated, cuff usually surrounding the center phalange of the third
digit that pneumatically couples arterial pulsations via air pressure to a custom-
designed piezo-electric pressure sensor. This sensor converts the pressure pulsations
into a voltage signal that is then digitized at 500 Hz, sufficient to adequately overs-
ample the maximally 25 Hz spectral range that is available due to the mechanical
filtering constraints of the arterial wall. Pulse detection, pulse envelope identifica-
tion and verification, fiduciary point identification, derivation of pulse parameters
and their conversion into hemodynamic parameters are accomplished in firmware
running currently on an ARM7 processor platform. The form factor is a wrist-
mounted unit providing minimal user interaction via a small vital sign display and a
single, multifunction, button. Hemodynamic data, raw pulse data, as well as exten-
sive device status-related information are wirelessly conveyed to a remote data dis-
play, such as the Android-based tablet shown in Fig. 7.29. The tablet display
provides significantly more in-depth physiological information, both current and
historical, and opportunities for user interaction and control. Communication
modalities such as Wi-Fi and cellular and protocols such as HL7 provide system
interface options based on customer requirements. Because of the opportunities
PDA provides, principally the tracking of central pressure effects which eliminates
the requirement for tracking hydrostatic head and, because it is a pulse analysis
approach, the only occasional use of the finger cuff-inflating pump, the device is
light (7 oz.), features a small footprint, and can operate for about 24 h on a single
battery charge. CareTaker™ 5, the next-generation device, is in the early stages of
Fig. 7.29 CareTaker™ Wireless Continuous Blood Pressure and Heart Rate Monitor with Finger
Cuff Technology. Copyright 2018. Used with written permission
development as of this writing and, commensurate with the remarkable technologi-

cal advances of its components, will likely be half its predecessor’s size while per-
mitting stand-alone operation as it will feature the communication and user interface
capabilities presently reserved for the remote data display.
Equally important will be the expansion of hemodynamic parameters that the
system will be able to provide. While LVET and arterial stiffness are currently
implemented as supporting parameters in the PDA formalism, they will be
incorporated into the set of FDA-cleared parameters once further clinical validation
is complete. Other parameters now in development include cardiac output and hem-
orrhage detection. The former has been studied extensively in the context of pulse
analysis and some of the insights gained from the PDA physiological model may
provide opportunities for refinement. The latter represents a largely elusive goal
long sought in both military and civilian markets, especially for the detection of
occult bleeding. Preliminary CareTaker™ clinical studies involving lower-body-
negative-chambers for the simulation of central hemorrhage [25] and blood donation
104 M. C. Baruch
studies suggest that the arterial pulse undergoes distinct changes in response.
Extracting them reliably in the presence of other masking hemodynamic changes
will be required to validate the approach.
Another area of interest with promising early results is the prediction of severe
hypotension in the subgroup of women undergoing C-section, specifically the pre-
procedure identification of those individuals most likely to develop severe hypoten-
sion once spinal anesthesia is administered [26].
Pulse decomposition analysis has many possibilities as a physiological model.
Other studies are planned to further validate and confirm use of the technology.
Much interesting work lies ahead.
References
1. O’Rourke MF, Yaginuma T. Wave reflections and the arterial pulse. Arch Intern Med.
1984;144(2):366–71.
2. Quick CM, Berger DS, Noordergraaf A. Constructive and destructive addition of forward and
reflected arterial pulse waves. Am J Physiol Heart Circ Physiol. 2001;280(4):H1519–27.
3. Söderström S, Sellgren J, Pontén J. Aortic and radial pulse contour: different effects of nitro-
glycerin and prostacyclin. Anesth Analg. 1999;89(3):566–72.
4. Olufsen MS, Peskin CS, Kim WY, Pedersen EM, Nadim A, Larsen J. Numerical simulation
and experimental validation of blood flow in arteries with structured-tree outflow conditions.
Ann Biomed Eng. 2000;28(11):1281–99.
5. McDonald DA. Blood flow in arteries. 4th ed. London: Arnold; 1998. p. 177.
6. Latham RD, et al. Regional wave travel and reflections along the human aorta: a study with six
simultaneous micromanometric pressures. Circulation. 1985;72:1257–69.
7. Kriz J, et al. Force plate measurement of human hemodynamics. http://arxiv.org/abs/
physics/0507135.
8. Greenwald SE, Carter AC, Berry CL. Effect of age on the in vitro reflection coefficient of the
aortoiliac bifurcation in humans. Circulation. 1990;82(1):114–23.
9. Korteweg DJ. Über die Fortpflanzungsgesschwindigkeit des Schalles in elastischen Rohren.
Ann Phys Chem. 1878;5:520–37.
10. Hallock P, Benson IC. Studies on the elastic properties of human isolated aorta. J Clin Investig.
1937;16:595–602.
2015;62(8):1879–901.
12. Anliker M, Histand MB, Ogden E. Dispersion and attenuation of small artificial pressure
waves in the canine aorta. Circ Res. 1968;23(4):539–51.
13. Couceiro R, Carvalho P, Paiva RP, Henriques J, Quintal I, Antunes M, Muehlsteff J, Eickholt
C, Brinkmeyer C, Kelm M, Meyer C. Assessment of cardiovascular function from multi-
Gaussian fitting of a finger Photoplethysmogram. Physiol Meas. 2015;36(9):1801–25.
14. O’Rourke MF, Mancia G. Arterial stiffness. J Hypertens. 1999;17(1):1–4.
15. Takazawa K, Tanaka N, Fujita M, Matsuoka O, Saiki T, Aikawa M, Tamura S, Ibukiyama
C. Assessment of vasoactive agents and vascular aging by the second derivative of photople-
thysmogram waveform. Hypertension. 1998;32(2):365–70.
16. Lantelme P, Mestre C, Lievre M, Gressard A, Milon H. Heart rate: an important confounder of
pulse wave velocity assessment. Hypertension. 2002;39(6):1083–7.
17. O’Rourke MF, Kelley RP, Avolio AP. The arterial pulse. Philadelphia: Lea & Febiger; 1992.
18. Esper SA, Pinsky MR. Arterial waveform analysis. Best Pract Res Clin Anaesthesiol.

2014;28(4):363–80.
19. Chia CW, Saul JP, Lee CC, Mark RG. Monitoring the changes in peripheral vascular resistance
using the shape of the radial blood pressure pulse. Comput Cardiol. 1992;19:567–70.
20. Irwin Gratz DO, Edward Deal DO, Francis Spitz MD, Baruch MC, Allen E, Seaman JE,
Pukenas E, Jean S. Continuous non-invasive finger cuff CareTaker® comparable to invasive
intra-arterial pressure in patients undergoing major intra-abdominal surgery. BMC Anesthesiol.
2017;17:48.
21. Baruch MC, Kalantari K, Gerdt DW, Adkins CM. Validation of the pulse decomposition analy-
sis algorithm using central arterial blood pressure. Biomed Eng Online. 2014;13:96.
22. Phillips AA, Burr J, Cote AT, Foulds HJ, Charlesworth S, Bredin SS, Warburton

DE. Comparing the Finapres and CareTaker systems for measuring pulse transit time
before and after exercise. Int J Sports Med. 2012;33(2):130–6.
23. Williams B, et al. Differential impact of blood pressure-lowering drugs on central aortic
pressure and clinical outcomes: principal results of the Conduit Artery Function Evaluation
(CAFE) study. Circulation. 2006;113(9):1213–25.
24. Low PA. “Laboratory evaluation of autonomic function.” Clinical autonomic disorders. 2nd
ed. Philadelphia: Lippincott-Raven; 1997. p. 186–7.
25. Baruch MC, Warburton DE, Bredin SS, Cote A, Gerdt DW, Adkins CM. Pulse decomposition
analysis of the digital arterial pulse during hemorrhage simulation. Nonlinear Biomed Phys.
2011;5(1):1.
26. Gratz I, et al. A predictive model for the development of hypotension following spinal anes-
thesia for elective cesarean section patients based on arterial stiffness (AS) calculated by a
continuous blood pressure device (CareTaker). American Society of Anesthesiologists Annual
Meeting 2018, Abstract Number: A3033.
Chapter 8
Pulse Wave Analysis Techniques
Martin Proença, Philippe Renevey, Fabian Braun, Guillaume Bonnier,

Ricard Delgado-Gonzalo, Alia Lemkaddem, Christophe Verjus,
Damien Ferrario, and Mathieu Lemay
Abstract Pulse wave analysis, or PWA, is a technique based on the morphological

analysis of blood pressure waveforms, whose shape reflects crucial information on
the properties of the arterial wall and blood pressure itself. Although the first histori-
cal developments of PWA date back to the nineteenth century, the technique started
to prosper with the development of applanation tonometry in the 1960s. More
recently, the approach has been increasingly applied to photoplethysmographic sig-
nals, due to the appeal in deriving blood pressure-related information from signals
routinely measured in clinical settings. In this chapter, after an introduction on the
historical and physiological background of PWA, a review of the waveform features
most commonly encountered in the literature is given, followed by an overview of
PWA-based clinical studies and an outlook on the clinical potential of the
technique.
Keywords Pulse wave analysis (PWA) · Blood pressure (BP) · Applanation

tonometry · Photoplethysmography (PPG) · Waveform analysis · Arterial stiffness ·
Augmentation index · Clinical studies · Transfer function
Introduction
Pulse wave analysis, or PWA, is a technique consisting in the morphological analy-

sis of the pressure pulse waveform. With the purpose of deducing indicators of
cardiovascular health, the technique can provide indirect means of assessing blood
pressure. In this chapter, we briefly review the historical developments of PWA,
from the design of the first sphygmographs to the main current cuffless modalities
used for the measurement of the pulse wave. We then overview the main
M. Proença (*) · P. Renevey · F. Braun · G. Bonnier · R. Delgado-Gonzalo

A. Lemkaddem · C. Verjus · D. Ferrario · M. Lemay
Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et
de Microtechnique), Neuchâtel, Switzerland
e-mail: martin.proenca@csem.ch

108 M. Proença et al.
physiological aspects required to grasp PWA techniques, that is, to comprehend the
various cardiovascular factors that play a role in shaping the pressure waveform. An
abundance of morphological features of the pressure waveform have been proposed
in the literature over the years; we provide an overview of the ones most frequently
encountered. We then see how these various PWA features have been linked in sev-
eral different clinical studies to various indicators of cardiovascular health and dis-
eases. Finally, we provide an outlook about the short- to mid-term future of PWA
techniques and their role in the assessment and monitoring of blood pressure in vari-
ous contexts and use cases.
Short History Overview
The history of PWA goes back to the middle of the nineteenth century, when the first
sphygmograph was invented by Vierodt [1]. The working principle of his device
consisted in a mechanism gradually applying a pressure onto the radial artery using
weights and levers, to the point of interrupting the blood flow, thereby allowing a
(rather crude) estimation of systolic arterial pressure. Marey [2] significantly
improved on Vierodt’s rather bulky sphygmograph by making it portable and more
accurate, and by incorporating a mechanism allowing the recording of the pressure
waveform on smoked paper (Fig. 8.1). Several further improvements have been
brought over time by researchers such as Mahomed, who significantly contributed
to the advancement of sphygmography, with his characterization of the radial hyper-
tensive pulse [1, 3].
A turning point occurred at the end of the nineteenth century when Riva-Rocci
invented the sphygmomanometer, a device made of an air cuff intended to be inflated
around the upper arm, thereby greatly facilitating the occlusion of the artery and
allowing an easier and more accurate measurement of systolic pressure [5]. The
auscultatory method for blood pressure measurement came into existence in 1905,
Fig. 8.1 Sphygmograph recording a pulse waveform on smoked paper. Image source and credit:
Wellcome Collection [4]
8 Pulse Wave Analysis Techniques 109
when Korotkov combined the use of Riva-Rocci’s sphygmomanometer and a

stethoscope, thus allowing for the first time the measurement of diastolic pressure
[6]. These advances drastically slowed down the development of studies dedicated
to the recording of the pressure waveforms. Indeed, with the possibility to easily
measure the diastolic and systolic components of blood pressure, it was considered
at the time that measuring pressure waveforms would not bring more information
regarding the cardiovascular health of the patient [7]. Towards the beginning of the
1960s, the use of arterial catheters in clinics started increasing significantly [8].
With this more frequent access to pressure waveforms, their importance became
more apparent. It quickly became clear that a full picture of a patient’s cardiovascu-
lar health was hardly explainable by a simple pressure measurement, particularly
regarding the clinical management of hypertensive patients. Indeed, other cardio-
vascular parameters, such as central arterial stiffness or its correlate, aortic pulse
wave velocity, are known to be better independent predictors of cardiovascular
events and all-cause mortality [9, 10]. In addition, Riva-Rocci’s sphygmomanome-
ter solely gives access to brachial blood pressure, less relevant a predictor of cardio-
vascular outcome than central pressure [11, 12]. All these considerations contributed
to the progressive revival and flourishing of PWA techniques.
Applanation Tonometry
The first major advances of PWA are directly linked to the development of a cuffless
modality known as applanation tonometry. Introduced in 1963, applanation tonom-
etry is a pressure waveform measurement technique dedicated to be used in super-
ficial arteries, such as the radial, carotid, brachial or femoral arteries [13]. By
applying a pressure sensor lightly onto the artery, such that the artery is slightly
flattened between the sensor and the underlying structures, the tangential forces in
the arterial wall are eliminated: a pressure waveform faithfully reproducing the
waveform of a catheter can be measured (Fig. 8.2) [14, 15]. The positioning of the
sensor is extremely sensitive and a trained operator is needed to perform the
Fig. 8.2 Metrological concept of applanation tonometry: a superficial artery is flattened between
the sensor and the underlying structures, thereby eliminating the tangential forces in the arterial
wall and allowing the measurement of intra-arterial pressure
measurement [16]. Nevertheless, although a perfect positioning of the sensor onto

the artery allows the absolute measurement of the pulse pressure, the tonometric
measurement does not provide an absolute measurement of blood pressure: the
measured waveform requires an initialization via a cuff sphygmomanometer in
order to be adequately translated to pressure units. This is typically done by using
the diastolic and mean pressures, as these are virtually the same in the aorta and the
more peripheral arteries such as the brachial or radial arteries [17]. An overview of
PWA studies and clinical evidence based on applanation tonometry will be detailed
later in section “Clinical Evidence”.
Among the main commercial systems implementing the technique of applana-
tion tonometry, the well-known SphygmoCor® (AtCor Medical Pty Ltd., Australia)
uses the pressure waveform measured at the radial artery and a generalized transfer
function to transform the former into an aortic pressure waveform. Alternatively, the
carotid pressure waveform can be used as a direct surrogate of the aortic waveform
due to the similarity between both morphologies [14, 18], as in the case of the
PulsePen® (DiaTecne s.r.l., Italy). Complior Analyse® (Alam Medical, France) uses
a similar approach using a piezo-electric sensor held via a system against the carotid
artery. The Millar SPT-301 Pulse Wave Tonometer (Millar Instruments, Inc., USA)
uses a high-fidelity micromanometer and can be used for recording pressure wave-
forms from the radial, carotid or femoral arteries. While most radial tonometers are
handheld devices, tonometers embedded in wrist supports or bracelets have also
been proposed, such as the BPro (HealthSTATS International Pte Ltd., Singapore)
or the OMRON HEM-9000AI (OMRON Healthcare, Japan).
Photoplethysmography
Applanation tonometry is well-suited for single individual measurements as snap-

shot assessments of cardiovascular health. However, its dependency on the operator
and its sensitivity to sensor positioning make it impractical for the continuous moni-
toring or unsupervised assessment of blood pressure. In that context, the clinical
potential of a cuffless modality known as photoplethysmography (PPG), a non-
occlusive optical technology already routinely used clinically in pulse oximeters for
the continuous monitoring of blood oxygen saturation and heart activity, appears as
an appealing alternative. Photoplethysmographs measure pulsating blood volume
changes by illuminating a pulsatile tissue with a light source such as an LED, and
measuring the light intensity received by a photoreceptor after it has gone through
the tissue. PPG measurements can be done in transmission mode (through tissue,
e.g. at the fingertip or the ear) or in reflectance mode (e.g. at the wrist, the chest, or
any other location with adequate superficial vascularization) (Fig. 8.3). It is mainly
the pulsatility of the small arteriolar bed that generates variations of received light
intensity in the PPG signal, whose morphology resembles that of pressure wave-
forms. Although the relation between both waveforms is complex [19], it has been
shown that the PPG signal at the fingertip contains the same physiological
Fig. 8.3 Transmission versus reflectance PPG modes: the intensity of light received at the photo-
diode is modulated by the changes in pulsating blood volume occurring in the illuminated tissue.
Adapted from [23]
determinants as the radial pressure waveform [20, 21], and undergoes similar mor-
phological changes in vascular disease [22]. An overview of PWA studies and clini-
cal evidence based on photoplethysmography is given in section “Clinical Evidence”.
The amplitude of the PPG signal is closely related to tissue perfusion, and thus it
is more the morphological information of the waveform, rather than its absolute
amplitude, that is of interest for the monitoring of blood pressure. Therefore, as for
applanation tonometry, a blood pressure measurement via PPG necessarily requires
an initialization procedure in order to translate the pulse wave analysis features into
blood pressure values. The continually increasing use of PPG technology is likely
to lead to the emergence of several PWA-based solutions on the market in the near
future. However, to this day most PPG-based commercial systems for BP monitor-
ing do not use PWA per se but still rely on pulse arrival time (see Chap. 5) or pulse
transit time (see Chap. 6) techniques despite their limitations.
Brief Physiology Description
PWA is the study of the morphology of the pulse pressure waveform in order to
obtain information about the state of ventricular ejection and the (elastic and geo-
metric) properties of the arterial tree. It is therefore crucial to first address the basics
regarding the origins of the blood pressure pulse and the genesis of its waveform.
As blood is being ejected from the heart, a blood pressure wave is generated and
travels along the walls of the arterial tree. This pressure wave is partly reflected at
each arterial bifurcation or discontinuity due to the mismatch of hydraulic imped-
ance. These reflected waves travel back to the heart. The resulting aortic pressure
waveform is thus a superposition of the forward wave travelling from the heart
towards the periphery and a global reflected wave travelling back to the heart [1,
24]. With age, the arteries stiffen which implies an increase in the velocity at which
the pressure wave travels along the arterial tree. Thus, the return of the reflected
wave occurs earlier in older subjects, as also illustrated in Fig. 8.4. While in younger
subjects the time of reflection occurs in late systole or early diastole (Fig. 8.4a), in
older subjects this occurs earlier in systole (Fig. 8.4b) and thus leads to an increased
systolic pressure.
One way of quantifying this augmented portion of pressure is via the augmenta-
tion index (AIx) which provides a surrogate measure for wave reflection:
SBP − Pr −1, if Pr occurs afterSBP

AIx = 100% ⋅ ⋅ x with x = {
SBP − DBP 1, if Pr occurs beforeSBP
Therefore, AIx is typically negative in younger subjects (Pr after SBP) and positive
in older subjects (Pr before SBP). Based on AIx, and thus the arrival time of the
backward wave, pressure waveforms can be classified into four different types as
suggested by Murgo et al. [1, 25] and shown in Fig. 8.5. While the inflection point
Pr is clearly visible in younger subjects (Type C and B), it is less apparent in older
subjects (Type A) and may even completely disappear with age (Type D).
Distinguishing different types of waves according to this classification can help to
better extract characteristic features from the pressure waveform. So far, we have
only mentioned a few of such features (SBP, Pr, Tr, LVET, etc.) but will address a
variety of common PWA features in the next section.
The pressure waveforms presented so far were showing central aortic pressure.
Yet, unless an invasive catheter is used, this particular pressure cannot be easily
Fig. 8.4 Two illustrative examples of blood pressure (BP) pulse waveforms in (a) a younger and
(b) an older subject. The final pressure waveform (black solid line) is a superposition of a forward
wave (grey dashed line) and a reflected backward wave (grey dotted line). In older people with
higher arterial stiffness the reflected wave returns earlier in the cardiac cycle (low Tr) when com-
pared to younger people (higher Tr). This results in an increased systolic pressure (SBP) in older
subjects. The systolic phase ends with the closure of the aortic valve at the end of the left ventricu-
lar ejection time (LVET, characterized by the local inflexion called dicrotic notch or incisura). With
DBP as diastolic blood pressure and Pr as the pressure at time of reflection, i.e. at time Tr
assessed. This is why in practice, most pressure measurements are performed in the
periphery (brachial, radial, digital, carotid or femoral arteries). However, peripheral
pressure is less good an indicator for cardiovascular risk than central pressure [11,
12], because the pressure truly “seen” as afterload by the heart and defining its
workload is central aortic pressure. Even though diastolic and mean pressures
remain virtually identical in all aforementioned arteries [17], this is not the case for
systolic pressure which can vary significantly between central and peripheral [1, 24]
as also shown in Fig. 8.6. To overcome this problem of not being able to access
central systolic pressure and PWA indicators from peripheral waveforms, several
solutions have been proposed, such as transfer functions (TF), that is, mathematical
Fig. 8.5 Classification of different pressure waveform morphologies into four types (A to D)
according to Murgo and Nichols [1, 25]. Adapted from [7]
Fig. 8.6 Amplification of the pressure waveform. As the blood pressure pulse travels away from
the heart, systolic blood pressure is amplified more and more in the periphery while diastolic and
mean pressure remain at similar levels. Adapted from [11]
Fig. 8.7 Estimation of the radial pressure waveform from an aortic pressure waveform using a
generalized transfer function (GTF), and conversely. Pc(f) and Pp(f): aortic and radial pressure
waves as a complex Fourier series, respectively; H(f): frequency response of the aortic-to-radial
GTF. Adapted from [27]
functions typically used to convert a peripheral pressure into a central pressure

waveform. Due to the pulsatile nature of pressure waves, Fourier analysis can be
applied to pressure waveforms to obtain the magnitude and phase of each harmonic
component of the cardiac frequency [24, 82], allowing to represent a pressure wave-
form as a complex Fourier series P(f). Then, a dedicated generalized TF, or GTF,
trained on a certain population and characterized by its frequency response H(f),
allows to mimic the changes in amplitude and phase undergone by the various car-
diac harmonics of the central pulse as it travels towards the periphery. Thus, the
GTF converts the central pressure Pc into a peripheral pressure Pp, i.e. Pp(f) = H(f)Pc(f)
(see Fig. 8.7). Conversely, when estimating the central waveform from a peripheral
waveform, as is typically the case in practice, the central waveform is estimated
using the inverse frequency response of the GTF, namely, H−1(f). A large variety of
TFs have been proposed in the literature for different peripheral locations and popu-
lations, ranging from subject-specific TFs to generalized models [26, 27]. Once this
transformation has been performed the central pressure waveform obtained can be
analyzed for a variety of PWA features, as described in the following section.
PWA Features
A myriad of morphological features of the pulse waveform has been proposed in the
literature over time. These features can provide quantitative information concerning
general cardiovascular health, and central blood pressure in particular. Hereafter, we
will first introduce the features frequently encountered in PWA, namely, those
dedicated to the analysis of actual pressure waveforms, such as those that would be
measured via an arterial line or a tonometer. Several of these features are directly
linked to the amplitude of the wave, and are therefore expressed in actual pressure
units (mmHg). This aspect, in addition to the known morphological differences
between the pressure waveform and the photoplethysmogram [19, 20], have led to
the development a plurality of amplitude-related features specifically for the analy-
sis of PPG waveforms, for which the absolute amplitude values do not have a direct
interpretation. Therefore, after introducing frequently encountered features of PWA
in section “Frequently Encountered Features”, we describe various PPG-specific
morphological features that have been proposed in the literature in section
“Photoplethysmography-Specific Features”. Lastly, we also briefly overview an
important aspect of PWA when used in the context of blood pressure measurement
or monitoring: the translation of PWA features into estimates of blood pressure via
an initialization procedure.
Frequently Encountered Features
In Table 8.1 and Fig. 8.8, the most frequently encountered parameters of central
PWA are represented and described. We have already mentioned earlier one the
most widely used of these features, the augmentation index (AIx). It is closely
related to Tr, the timing of the onset of the backward reflected wave. Both indexes
are known to correlate reasonably well with aortic pulse wave velocity [1, 28] and
have been extensively used in clinical studies as indicators of central arterial stiff-
ness (as later detailed in section “Clinical Evidence”). It is important to note that
several factors affect AIx and Tr, including the subject’s height and age, and heart
rate (HR). In order to account for this dependency on the heart rate, the AIx@75—
namely, the AIx corrected for a heart rate of 75 bpm—was proposed as [83]:
AIx@ 75 = 0.39 ⋅ ( HR − 75 ) + AIx.
Thus, for instance, we find that an AIx of 23% at a heart rate of 63 bpm produces
the same AIx@75 that an AIx of 14% at a heart rate of 86 bpm produces, as
0.39 ⋅ ( 63 − 75 ) + 23 ≅ 0.39 ⋅ ( 86 − 75 ) + 14 ≅ 18.3%.
One example of use cases where these indicators (AIx, AIx@75 or Tr) have been
extensively used is in pregnancy-related studies, for the assessment of hypertensive
complications such as preeclampsia [29]. The timing of onset of preeclampsia and
the severity of the disorder were linked to higher values of AIx and AIx@75, and
shorter values of Tr [30]. Another essential feature of PWA is the sub-endocardial
viability ratio (SEVR), an index of myocardial oxygen supply and demand, or car-
diac ischemic risk, which is increased with aortic stiffening. The SEVR estimates
the ratio of myocardial perfusion relative to cardiac workload and can be
Table 8.1 Frequently encountered parameters in central pulse wave analysis

Parameter Acronym/definition Description
Systolic blood SBP The maximal blood pressure value
pressure (mmHg) over one heartbeat
Diastolic blood DBP The minimal blood pressure value
pressure (mmHg) over one heartbeat
Pulse pressure PP = SBP – DBP The maximal peak-to-peak
(mmHg) amplitude of the pressure wave over
one heartbeat
Mean arterial pressure MAP The average blood pressure value
(mmHg) over one heartbeat
Mean systolic blood MSBP The average blood pressure value
pressure (mmHg) during systole
Mean diastolic blood MDBP The average blood pressure value
pressure (mmHg) during diastole
End-systolic pressure ESP The blood pressure value at end of
(mmHg) systole, when the aortic valve closes
Left ventricular LVET The duration of systole
ejection time (s)
Diastolic time (s) DT The duration of diastole
Time to systolic peak Tp Time at which systolic blood
pressure reaches its peak
Travel time of the Tr Time delay between the forward and
reflected wave (s) backward reflected waves
Blood pressure at time Pr Blood pressure value at the onset of
of reflection (mmHg) the backward reflected wave
Augmented pressure AP = (SBP − Pr)·x The increase in blood pressure
(mmHg) induced by the early arrival of the
reflected wave, with x = − 1 if Pr
occurs after SBP, and x = 1
otherwise
Augmentation index The increase in blood pressure
(%) AP induced by the early arrival of the
AIx = 100 ⋅
PP reflected wave, relative to the
peak-to-peak amplitude of the
pressure wave
Augmentation index AIx@75 = 0.39 ⋅ (HR − 75) + AIx Augmentation index corrected for its
corrected for a heart dependency on heart rate (HR)
rate of 75 bpm (%)
Form factor (−) FF = (MAP − DBP)/PP Index used to evaluate in a single
number the morphology of a
pressure wave (FF increases with
age)
Systolic pressure-time SPTI = MSBP ⋅ LVET Area under the systolic part of the
index (mmHg·s) pressure wave, representing an index
of myocardial oxygen demand
(continued)
Parameter Acronym/definition Description
Diastolic pressure- DPTI = MDBP ⋅ DT Area under the diastolic part of the
time index (mmHg·s) pressure wave, representing an index
of coronary blood supply
Sub-endocardial Index of sub-endocardial oxygen
viability ratio (−) DPTI supply and demand
SEVR =
SPTI
Fig. 8.8 Pressure waveform with the main features of interest. (Left) Typical waveform for a
young subject, where the inflection point Pr induced by the onset of the reflected wave occurs after
the maximum of the wave. (Right) Typical waveform for an older subject, where the inflection
point Pr occurs before the maximum of the wave, indicating stiffer arteries
approximated from the ratio of the diastolic and systolic areas under the central
pressure waveform as:
DPTI
SEVR = ,
SPTI
where DPTI and SPTI are defined in Table 8.1, with the remainder of the main clas-
sical features of PWA. These are quite self-explanatory and will therefore not be
further detailed here.
Photoplethysmography-Specific Features
Although the morphology of the PPG signal looks similar to the arterial pressure
pulse, the waveform contains some important differences [19, 20, 31]. In this con-
text, a plurality of PPG-specific features has been proposed in the literature.
Systolic amplitude, that is, the peak-to-peak amplitude of the PPG wave, was
suggested as an estimate of blood pressure in conjunction with the pulse arrival time
by [32], although its value is mainly determined by the pulsatile changes in blood
volume and by peripheral changes in arterial tone. The reflection index RI = y/x (see
Fig. 8.9a) was suggested as a correlate of brachial-ankle pulse wave velocity [33].
A correlate of total vascular resistance was suggested by Wang [34] as the ratio of
Fig. 8.9 (a) Typical PPG waveform measured at the fingertip. (b) Second derivative of the PPG
waveform (acceleration plethysmogram) with the key wave-points a, b, c, d and e
the area under the curve from the foot to the dicrotic notch to the leftover area after
the dicrotic notch. Finally, Millasseau [35] used the time delay ΔT between the
systolic and diastolic peaks (see Fig. 8.9a) to propose an index of the stiffness of the
large arteries as SI = h/ΔT, where h is the height of the subject. They found SI to
correlate with carotid-to-femoral pulse wave velocity, age and mean arterial blood
pressure.
The second derivative of the photoplethysmogram, sometimes referred to as
acceleration plethysmogram (APG), has been extensively used in the pulse wave
analysis of PPG waveforms. As defined by Takazawa et al. [36, 37], the waveform
of the APG includes four systolic waves and one diastolic wave (see Fig. 8.9b),
namely, the a-wave (initial positive wave), the b-wave (early negative wave), the
c-wave (re-increasing wave), the d-wave (re-decreasing wave) and the e-wave (dia-
stolic wave). The position of the e-wave is aligned with the dicrotic notch. The c-
and d-waves are sometimes absent of the APG. APG-based pulse wave analysis is
usually performed in terms of the amplitude ratios of each wave with respect to the
a-wave. These ratios (b/a, c/a, d/a and e/a) have been shown to correlate with a
plurality of cardiovascular health indicators, as will be later detailed in section
“Clinical Evidence”.
From PWA Features to Blood Pressure
We have seen that many classical features of PWA—such as those that would be
calculated on a tonometric pressure waveform—are already expressed in actual
pressure units (mmHg). However, the prerequisite behind this is that the tonometric
waveform has been properly initialized prior to the measurement. This is standardly
done using a traditional sphygmomanometer at the brachial artery. As the diastolic

and mean pressures are virtually the same in the aorta and the more peripheral arter-
ies such as the brachial or radial arteries [17], the systolic value can be found by
exploiting the definition of mean arterial pressure: the tonometer-derived estimated
aortic pulse, with an offset equal to diastolic pressure, is scaled between its diastolic
and systolic values such that its average value equals the targeted mean arterial
pressure.
Nevertheless, we have seen that many other features—particularly PPG-specific
features—are unitless or expressed in other units than pressure. In the context of
blood pressure monitoring, these features need to be translated to pressure units: a
procedure commonly referred to initialization or calibration. Numerous initializa-
tion strategies have been implemented in the past, including population-wise initial-
izations, and patient-dependent initializations (based either on one-shot or
intermittent measurements with a reference, on hydrostatic pressure changes, or on
numerical models). In the following, we quickly skim over consideration of tonom-
eters and generic features. For a more detailed description of initialization processes
and their use as calibration methods, see Chap. 10.
Translating one or a combination of features to actual values of blood pressure in
mmHg has been approached with diverse statistical tools over time. The most used
methods are (multi)linear regressions. These tools are well understood, simple and
therefore the method of choice for medical environments when dealing with patient-
dependent initialization procedures. Non-linear methods have also been used to
tackle the problem in order to deal with the non-linearities that arise when estimat-
ing central blood pressure from peripheral measurements [38], or due to the physi-
ological non-linearities linking the various features with blood pressure [1, 39]. In
recent years, machine-learning approaches have become popular thanks to the
increasing availability of databases with proper references. These techniques are
capable of handling high levels of non-linearities, but can be difficult to interpret in
the medical community.
The process of translating the waveform features to blood pressure is usually
user-dependent and changes with age [39]. This means that a process of initializa-
tion is always needed for every new subject. Great effort is currently being put in
order to obtain global initialization functions that would work for large segments of
the population. In the design of these initialization functions, the benefit of ease of
use comes at the cost of decreased individual accuracy within a cohort.
Clinical Evidence
Multiple clinical studies have been performed to assess the accuracy of PWA tech-
niques in terms of central blood pressure estimation or cardiovascular health assess-
ment, and the correlation between PWA features and clinical outcome related to
cardiovascular diseases. In the following, we provide an overview of some of the main
clinical studies performed both via applanation tonometry (Table 8.2) and
Table 8.2 Examples of PWA clinical studies based on applanation tonometry. Most of the
included studies are focused on the accuracy of the estimation of central blood pressure, although
a wide range of studies in the literature are rather centred on the assessment cardiovascular risk
factors, indices of arterial stiffness and prediction of clinical outcome from general PWA features,
and not particularly on blood pressure itself. BP blood pressure, SBP systolic blood pressure, DBP
diastolic blood pressure, PP pulse pressure, MAP mean arterial pressure, ESP end-systolic
pressure, (G)TF = (generalized) transfer function, NPMA N-point moving average, SBP2 late
systolic shoulder on peripheral pressure waveforms, SBP2-MVR multivariate regression mainly
based on SBP2, AP augmented pressure, AIx augmentation index, SPTI systolic pressure-time
index, DPTI diastolic pressure-time index, SEVR sub-endocardial viability ratio, LVET left
ventricular ejection time, Tp time to systolic peak, Tr travel time of the reflected wave, Pr blood
pressure at time of reflection, RMSE root mean square error, N/A not applicable (no initialization
performed)
Aortic BP
Study Population Measurement Initialization estimate Comparator
Pauca [17] 62 Radial N/A SphygmoCor Aortic
catheter GTF catheter
Catheter-derived radial waveforms transformed to aortic waveforms via the SphygmoCor GTF
were compared to catheter-derived aortic waveforms, yielding to small differences. SBP:
0.0 ± 4.4 mmHg; DBP, 0.6 ± 1.7 mmHg; MAP: −0.7 ± 4.2 mmHg; PP: −0.5 ± 2.0 mmHg. The
errors remained similar after nitroglycerine infusion
Pauca [53] 21 Radial N/A SBP2 Aortic
catheter catheter
Catheter-derived radial waveforms were used to estimate aortic SBP via the SBP2 method,
yielding to excellent agreement (0.6 ± 1.5 mmHg). The authors concluded that the late systolic
shoulder of the radial waveform, sometimes seen in older hypertensive patients, does indeed
represent aortic SBP
Soderstrom [54] 12 Radial N/A SphygmoCor Aortic
were compared to catheter-derived aortic waveforms. SBP was underestimated by 6–8 mmHg
(SD: 2–3 mmHg) and DBP overestimated by 4 (SD: 2). Despite these differences, the
synthesized waveforms tracked the measured waveforms at rest, following midazolam,
sublingual nitroglycerine and during Valsalva manoeuvres
Davies [55] 28 Radial N/A SphygmoCor Aortic
were compared to catheter-derived aortic waveforms. SBP was significantly underestimated
(−7.2 ± 10.1 mmHg) and DBP was significantly overestimated (12.2 ± 7.1 mmHg). SBP derived
from an oscillometric cuff was not significantly different from aortic (3.4 ± 10.4 mmHg), and
DBP was significantly overestimated (11.7 ± 7.2 mmHg). The authors concluded that the
SphygmoCor GTF is no better at estimating aortic BP than a standard oscillometric cuff
Sharman [56] 30 Radial Aortic SphygmoCor Aortic
tonometer catheter GTF catheter
Servocontrolled tonometer-derived radial waveforms transformed to aortic waveforms via the
SphygmoCor GTF were compared to catheter-derived aortic waveforms, at rest and during
ergometer exercise. Aortic SBP, DBP, PP and ESP were all strongly correlated (r > 0.98 at rest,
r > 0.97 during exercise), with excellent agreement between aortic and predicted aortic. SEVR
and LVET also strongly agreed both at rest and during exercise
Chen [57] 20 Radial Aortic Custom GTF Aortic
tonometer catheter catheter
(continued)
Aortic BP
Radial waveforms were derived from a wristwatch-like automated tonometer and initialized with
an aortic catheter were used. A GTF was trained using the patients’ radial and aortic catheter
waveforms. The agreement between the tonometer-derived predicted aortic BP and the aortic
catheter-derived BP was assessed at steady state and during hemodynamic transients (Valsalva
manoeuvre, abdominal compression, nitroglycerin, and vena cava obstruction). SBP:
0.0 ± 3.7 mmHg, PP: 0.2 ± 3.8; AIx: −7 ± 9% (−30 ± 45% accuracy)
Hope [58] 78 Radial Aortic Custom GTF Aortic
Tonometer-derived radial waveforms initialized with an aortic catheter were compared to the
catheter’s BP. SBP: 2.9 ± 7.3 mmHg, r = 0.94; DBP: 8.0 ± 2.5, r = 0.97. Several other PWA
parameters were compared: While significant correlations were found for Tp, LVET, SPTI,
DPTI, SEVR and Pr, only a weak correlation was found for AP and no correlation for AIx and
Tr, suggesting limitations in terms of clinical utility
Hope [59] 93 Radial Aortic Custom GTF Aortic
Tonometer-derived radial waveforms initialized with aortic catheter were used to test the
generalizability of radial-to-aortic transfer functions. The authors found that including more than
20 individual TFs for the GTF did not improve generalizability. Agreement between predicted
aortic and catheter-derived aortic BP was evaluated. SBP: 2 ± 11 mmHg, r = 0.83; DBP: −6 ± 6,
r = 0.79; PP: 8 ± 12 mmHg, r = 0.77. Good correlations (r > 0.65, p < 0.001) were found
between tonometer-derived SPTI, DPTI, SEVR, Tp, and their catheter-derived counterparts. The
correlation for LVET was weaker, while no correlation was found for Tr nor AIx
Rajani [60] 14 Radial Oscillom. SphygmoCor Aortic
tonometer cuff/aortic GTF catheter
catheter
Tonometer-derived radial waveforms initialized once with an aortic catheter and once via an
oscillometric device were compared to those of the aortic catheter in patients with aortic
stenosis. SBP was not significantly difference between oscillometric brachial and catheter aortic,
and between catheter-initialized tonometer-derived predicted aortic and catheter aortic.
Cuff-initialized tonometer-derived predicted aortic was significantly different (−8 ± 7 mmHg).
SphygmoCor-derived LVET differed from echocardiography-derived LVET. After correcting the
predicted aortic waveform with the LVET of echocardiography, SEVR was found to be accurate
Hickson [61] 1978 + 38 Radial Oscillom. SphygmoCor Aortic
tonometer cuff/aortic GTF and catheter
catheter SBP2
Tonometer-derived radial waveforms initialized via an oscillometric device were measured in
1978 patients. The aortic SBP estimate based on the late systolic shoulder of the radial
waveform (SBP2) was compared to the aortic SBP obtained via SphygmoCor’s GTF. The
agreement between both methods was good (1 ± 4 mmHg, r = 0.99), although larger errors
occurred at lower pressures. In 38 patients, tonometer-derived radial waveforms initialized using
the aortic catheter BP were used to derive both aforementioned estimates of aortic SBP. The
agreement between SBP2 and the catheter was good (2 ± 6 mmHg, r = 0.92), and between
SpyhgmoCor’s GTF and the catheter as well (−1 ± 9 mmHg, r = 0.88). This study showed that
SBP2 approximates central BP in a large cohort of patients, but may be inaccurate at low SBP
values
Hope [62] 42 Radial Oscillom. Custom GTF Aortic
tonometer cuff/aortic catheter
catheter
(continued)
Aortic BP
Tonometer-derived radial waveforms initialized once with an aortic catheter and once via an
oscillometric device were compared in terms of BP and PWA parameters with the aortic
catheter. With the invasive initialization: SBP: −1 ± 8 mmHg, r = 0.94; DBP: 9 ± 3, r = 0.96;
PP: −10 ± 9 mmHg, r = 0.92. With the cuff-based initialization: SBP: 7 ± 12 mmHg, r = 0.86;
DBP: 3 ± 6, r = 0.76; PP: 4 ± 12 mmHg, r = 0.82. With both initializations good correlations
(r > 0.65, p < 0.001) were found between tonometer-derived SPTI, DPTI, SEVR, AIx, LVET,
and their catheter-derived counterparts. The correlation for Tp was weaker, while no correlation
was found for Tr. The authors found that using the untransformed radial waveform instead of the
GTF-derived waveform produced smaller BP estimation errors when initializing with an
oscillometric cuff
Smulyan [63] 50 Radial Oscillometric SphygmoCor Aortic
tonometer cuff GTF catheter
Tonometer-derived radial waveforms initialized via an oscillometric device and compared to
aortic catheter BP yielded good correlations but differences with large scatter. SBP:
1.5 ± 11.1 mmHg, r = 0.89; DBP: −10.4 ± 12.7 mmHg, r = 0.59; PP: 11.5 ± 13.6, r = 0.80
Cloud [86] 30 Radial Oscillometric SphygmoCor Aortic
Tonometer-derived radial waveforms initialized via an oscillometric device and compared to
aortic catheter BP yielded differences with large bias and scatter. SBP: 13.3 ± 15.1 mmHg;
DBP: −11.5 ± 9.8 mmHg; MAP: −1.8 ± 9.5 mmHg. It was found that cuff-derived brachial BP
values provided an overall better estimate of aortic BP than tonometer-derived central BP values
from the SphygmoCor system
Zuo [64] 45 Radial Oscillometric SphygmoCor Aortic
Tonometer-derived radial waveforms initialized via an oscillometric device and compared to an
aortic catheter BP yielded differences with large bias and scatter. SBP: 4.2 ± 16.7 mmHg,
r = 0.84; DBP: −15.9 ± 12.5 mmHg, r = 0.59; PP: 20.1 ± 14.3 mmHg, r = 0.82. Differences
between the oscillometric cuff were in the same order of magnitude. The authors concluded that
the inaccurate measurement of the cuff for the initialization was the major limiting factor for the
use of SphygmoCor’s GTF in clinical settings
Ding [65] 33 Radial Oscillometric SphygmoCor Aortic
tonometer cuff GTF & catheter
SBP2-MVR
Tonometer-derived radial waveforms initialized via an Omron oscillometric device were
measured with the SphygmoCor device and the Omron HEM9000-AI device. Aortic BP was
estimated by applying SphygmoCor’s GTF to SphygmoCor’s signal, while it was estimated
from Omron’s signal using a multivariate regression equation involving SBP2 (Cheng 2010).
Both estimates where compared to catheter-derived aortic SBP. SphygmoCor: −15 ± 9 mmHg,
r = 0.91; Omron HEM9000-AI: −2 ± 10 mmHg, r = 0.90. Comparison of Omron’s oscillometric
cuff BP with invasive brachial BP showed that the cuff used for the initialization underestimated
brachial SBP by −19 ± 14 mmHg and was responsible for SphygmoCor’s underestimation of
SBP
Takazawa [66] 18 Radial Oscillometric SBP2 Aortic
tonometer cuff catheter
(continued)
Aortic BP
Radial waveforms derived form a wristwatch-like tonometer (HEM9000-AI, Omron) were
initialized via an oscillometric device. Aortic SBP estimation based on the SBP2 was compared
to catheter-derived aortic SBP in a protocol involving the injection of nicorandil. The
vasodilator-induced negative change of aortic SBP was estimated well (−0.6 ± 6.4 mmHg,
r = 0.91), and was estimated slightly less well when using radial BP directly (4.8 ± 8.8 mmHg,
r = 0.81). The authors also reported correlations between aortic and radial AIx, and concluded
that the SBP2 may enable more accurate estimation of aortic SBP than radial
Williams [67] 217 + 383 + 20 Radial Oscillometric NPMA SphygmoCor
tonometer cuff GTF and
aortic cath.
Tonometer-derived radial waveforms initialized via an oscillometric device were used to train
the NPMA on 217 patients. The training purpose was to determine the factor k such that the
moving average, which consists of N = Fs/k points where Fs is the sampling frequency in Hertz,
minimizes the estimation errors on aortic SBP when compared to SphygmoCor’s GTF. They
found k = 4. On a first validation set of 383 patients, the aortic SBP estimated via the NPMA
was compared to the SBP derived via SphymoCor’s GTF, yielding excellent agreement:
0.3 ± 1.0, r = 0.997. On a second validation set of 20 patients, a radial tonometer attached via a
wrist strap (A-Pulse, Healthstats) and initialized via an oscillometric cuff was used. Aortic SBP
obtained via the NPMA was compared to invasive SBP, yielding here also excellent agreement:
−0.4 ± 2.5, r = 0.99
Garcia-Ortiz [68] 104 Radial Oscillometric A-Pulse GTF SphygmoCor
tonometer cuff GTF
Radial waveforms derived form a wristwatch-like tonometer (B-Pro, Healthstats) initialized via
an oscillometric device estimated aortic BP via the dedicated A-Pulse software. BP and PWA
parameters from SphygmoCor’s GTF were used as reference. Good agreement (1.47 ± 5.15,
r = 0.949) was found on aortic SBP, but radial AIx showed differences (5.85 ± 21.09%,
r = 0.436) probably related to the late systolic shoulder of the radial wave. The authors also
reported the correlations between aortic SBP, PP and the SBP2 with indicators of arterial
stiffness, such as age, carotid-to-femoral PWV and the ankle-brachial index. All variables
correlated with age. All correlated with PWV, except B-Pro’s radial AIx, and with aortic AIx,
except B-Pro’s aortic PP
Westerhof [69] 50 Brachial N/A Custom GTF Aortic
catheter catheter
Catheter-derived brachial waveforms were transformed to catheter-derived aortic waveforms via
individual TFs and via a custom GTF. The RMSE on the estimation of the aortic waveform was
not significantly different when using the individual TFs or the GTF (4.0 ± 2.0 vs.
4.4 ± 2.0 mmHg). Aortic SBP, DBP, PP and AIx were also not significantly different between
both approaches. As details of the wave shape were reproduced slightly better with the
individual TFs but not significantly, the authors concluded that individualization of pressure
transfer is not effective in resting patients
Shih [70] 40 Brachial N/A Custom GTF Aortic
catheter catheter
Catheter-derived brachial waveforms were transformed to catheter-derived aortic waveforms via a
custom GTF. An oscillometric cuff also measured brachial BP. The purpose of the study was to
quantify separately the errors due to the GTF and those due to initialization via an oscillometric
cuff. The authors found that the apparent errors of the GTF came almost exclusively from cuff
errors, and that nearly 100% of input errors were transferred to the GTF output
(continued)
Aortic BP
Cheng [71] 100 Brachial Oscillometric SBP2-MVR Aortic
pulse volume cuff catheter
pleth.
Catheter-derived brachial waveforms and pulse volume plethysmography brachial waveforms
from an oscillometric BP monitor (initialized with the latter) were used to estimate aortic SBP
via the SBP2 method. Both for the invasive and non-invasive brachial waves, multivariate linear
regression involving SBP2, ESP, SPTI and DPTI, and where SBP2 was the major contributor,
were trained on 50 patients. Both multivariate models were then validated against catheter-
derived aortic SBP, yielding good agreement both for the invasive (0.5 ± 4.5 mmHg, r = 0.98)
and non-invasive (−0.1 ± 7.6 mmHg, r = 0.93) approaches. Using only SBP2 as an estimate of
aortic SBP yielded larger differences, both for the invasive (3.0 ± 5.2 mmHg, r = 0.96) and the
non-invasive (0.9 ± 9.4 mmHg, r = 0.89) approach
Van Bortel [72] 19 Carotid Aortic Direct Aortic
PP obtained from initialized tonometer waveforms at the common carotid artery yielded
estimation errors of 1.8 ± 5.2 mmHg (r = 0.92) when compared with catheter-derived aortic PP,
and −10.2 ± 14.3 mmHg (r = 0.39) when not initialized
Karamanoglu [73] 13 Carotid Aortic Generalized Aortic
tonometer catheter modif. catheter
Windkessel
Carotid waveforms predicted from aortic catheter waveforms transformed using a generalized
modified Windkessel model yielded significantly smaller differences than untransformed aortic
waveforms (RMSE of 3.4 ± 1.3 mmHg vs. 4.4 ± 1.6 mmHg)
Salvi [18] 10 Carotid Oscillometric Direct Aortic and
tonometer cuff carotid
catheters
Tonometer-derived carotid waveforms initialized via an oscillometric device were compared to
catheter-derived carotid and aortic waveforms. Statistical analysis showed no significant
difference between the first 6 harmonics of the tonometer waveforms and their invasive
counterparts. Aortic DBP was similar between both techniques, while aortic SBP and PP were
slightly underestimated by the tonometer (−2.7 ± 2.8% and − 5.7 ± 6.1%, respectively)
Kelly [74] 18 Carotid Oscillometric Direct Aortic
tonometer cuff catheter
Tonometer-derived carotid waveforms initialized via an oscillometric device and compared to
aortic catheter BP and AIx were not significantly different. SBP: 126 ± 28 vs. 127 ± 28 mmHg,
r = 0.96; DBP: 71 ± 10 vs. 66 ± 7 mmHg, r = 0.60; AIx: 23.9 ± 9.3% vs. 30.7 ± 11.9%,
r = 0.87
photoplethysmography (Table 8.3). The included tonometry studies are mainly focused

on the accuracy of estimating central blood pressure from peripheral tonometric mea-
surements, although a wide range of tonometry-based studies in the literature are rather
focused on the assessment cardiovascular risk and prediction of clinical outcome from
general PWA features, and not particularly on the accuracy of central blood pressure
estimation. On the other hand, most PPG-related PWA studies in the literature are not
directly focused on the estimation of blood pressure per se, but rather on evaluating
cardiovascular risk factors and indices of arterial stiffness.
Table 8.3 Examples of PWA clinical studies based on photoplethysmography. Most studies in the
literature are not directly focused on the estimation of blood pressure per se, but rather on evaluating
cardiovascular risk factors and indices of arterial stiffness via various PPG-based PWA techniques.
APG acceleration plethysmogram, BP blood pressure, SBP systolic blood pressure, DBP diastolic
blood pressure, PP pulse pressure, MAP mean arterial pressure, HR heart rate, LVET left ventricular
ejection time, AIx augmentation index, AGI vascular aging index, CVD cardiovascular disease,
GTF generalized transfer function, BMI body mass index, PWV pulse wave velocity, RMSE root
mean square error
Study Population PPG signal of interest Comparator
Takazawa [37] 39 + 600 APG indices Aortic catheter
(BP, arterial
stiffness indices)
Second derivative analysis in catheterized patients during baseline, angiotensin injection and
administration of sublingual nitroglycerin showed that -d/a was correlated (r = 0.80) with aortic
AIx and may be useful for evaluating the effects of vasoactive agents. In a second study in 600
subjects (50 men and women per decade from the third to the eighth decade), b/a was found to
increase (r = 0.75), and c/a (r = −0.67), d/a (r = −0.72) and e/a (r = −0.25) to decrease with age.
The so-called vascular aging index (AGI) was thus defined as (b-c-d-e)/a. AGI increased with
age (r = 0.80), and may be useful for evaluating vascular aging and for screening for
arteriosclerotic disease
Imanaga [47] 82 APG indices Cuff,
ultrasonography
(arterial
distensibility)
Second derivative analysis in 82 patients found that the b/a ratio of the APG was correlated to
carotid distensibility (r = 0.892), and that it may be a useful index of atherosclerosis and altered
arterial distensibility
Bortolotto [42] 524 APG indices and Cuff, arterial
PPG-derived AIx stiffness indices,
comparative
study
In 524 essential hypertensives (140 with atherosclerotic alterations (AA)), second derivative
analysis showed that patients with AA had significantly higher PPG-derived AIx (32.2 ± 16 vs.
25.2 ± 9%) and AGI (−0.093 ± 0.03 vs. −0.271 ± 0.018). AIx was independently influenced by
age and SBP, whereas the AGI was by age and AA. Building a logistic regression model for the
presence of AA showed that AIx was not a significant independent determinant, but may however
be useful for evaluating vascular aging in hypertensives
Hashimoto [44] 294 APG indices Oscillometric
cuff, arterial
stiffness indices
In hypertensive patients, univariate analysis showed that d/a was independently correlated with
age, MAP, and HR. In addition, b/a was independently associated with age, SBP, HR, body-mass
index (BMI), and gender. AGI was independently correlated with SBP, HR, BMI, and age
(continued)
Hashimoto [45] 848 APG indices Cuff, arterial
stiffness indices,
comparative
study
The APG ratios b/a, d/a and the AGI were calculated in 544 normotensives and 304
hypertensives. Univariate and multivariate analysis was performed to determine the determinants
of APG indices. The AGI and d/a showed a positive and b/a a negative independent correlation
with age and BP. In contrast, AGI and d/a showed a negative and b/a a positive correlation with
HR. The multivariate-adjusted d/a and AGI were significantly higher and the multivariate-
adjusted b/a significantly lower in hypertensives vs. normotensives, suggesting that APG indices
may be useful for detecting vascular aging accelerated by hypertension
Simek [48] 144 APG indices Cuff, arterial
stiffness indices,
comparative
study
Multivariate regression analysis was performed on APG indices in a group of 120 healthy
subjects and 24 with essential hypertension. The ratios b/a and c/a correlated only with age,
while d/a independently correlated with age, HR, MAP, body height and gender, and e/a with
age and MAP. After adjustment for HR, BMI and MAP, independent discriminative power
between hypertensives and healthy controls was preserved only for b/a and c/a
Otsuka [75] 211 APG indices Cuff, coronary
risk factors,
Framingham risk
score
The study was performed in 211 subjects with no apparent atherosclerotic disorders. APG
indices significantly correlated with several coronary risk factors and with the Framingham risk
score in both genders (b/a: r = 0.43 and r = 0.54, and d/a: r = −0.38 and r = −0.58, for male and
female, respectively). The b/a ratio discriminated high-risk subjects for coronary heart disease
with a sensitivity and specificity of 0.85 and 0.58 in males, and 0.83 and 0.72 in females
Otsuka [76] 973 APG indices Oscillometric
cuff, CVD risk
factors
In middle-aged Japanese men, age, hypertension, dyslipidaemia, impaired fasting glucose/
diabetes mellitus and lack of regular exercise were found to be independent determinants of an
increased b/a ratio, while age, hypertension and alcohol intake 6 or 7 days per week were
independent determinants of a decreased d/a ratio. The APG indices thus reflect arterial
properties that are affected by several CVD risk factors
Kawada [77] 1055 APG indices CVD risk factors
The APG indices were calculated in 1055 male workers. Both b/a and d/a were inversely but
weakly correlated with BP. However, factor analysis revealed that APG indices were categorized
with age and differed in characteristics from metabolic syndrome components, and that d/a is a
risk factor for the development of metabolic syndrome
Millasseau [20] 60 Custom GTF Radial catheter
(continued)
The authors trained a GTF to transform fingertip PPG waveforms into radial pressure waveforms
in 60 patients (20 with hypertension). All waveforms (reference radial waveforms and PPG
waveforms) were first normalized to the same PP by means of an oscillometric cuff for a
meaningful comparison. Via the GTF the radial waveforms could be predicted with an average
RMSE of 4.4 ± 2.0 mmHg. The GTF was not influenced by effects of hypertension or sublingual
nitroglycerin. These results suggest that the PPG waveform contains the same physiological
determinants than the radial waveform
Millasseau [35] 87 Stiffness index Oscillometric
cuff, carotid-to-
femoral PWV
A PPG-based large artery stiffness index (SI) was proposed as the ratio of body height—
Hypothesized to be proportional to the arterial path travelled by the pressure wave from the
aortic valve the major sites of reflection in the lower body—To the time delay between the
systolic and diastolic peaks of the PPG waveform. Glyceryl trinitrate (GTN) was used to
examine the sensitivity (or lack thereof) of SI to changes in tone of the small arteries, as GTN
markedly influences the contour of the PPG waveform but has little effect on large arteries. By
univariate analysis, SI was found to correlate with age (r = 0.67), sphygmomanometer-derived
SBP (r = 0.32), DBP (r = 0.48) and MAP (r = 0.45), as well as carotid-to-femoral PWV
(r = 0.65). The administration of GTN produced similar changes in SI and PWV, suggesting that
SI provides a simple measure of large artery stiffness
Gunarathne [87] 247 Stiffness index CDV risk factors,
comparative
study
In a cohort of 247 individuals, 60 had established risk factors of cardiovascular diseases (CVD).
On univariate analysis, SI was associated with standard risk assessments (r = 0.56) and increased
in ordinal fashion from low risk, to medium, to high and to very high. Sensitivity-specificity
analysis showed SI to be the best discriminator between low to medium and high-risk categories
when compared to other indicators such as total cholesterol, plasma glucose, SBP and waist-to-
hip ratio, suggesting that SI may aid in the stratification of CVD risk
Chan [78] 13 Custom PPG analysis Doppler aortic
flow velocity
In a protocol involving head-up tilting, LVET was estimated from a combination of three
estimates of LVET based on the first, second and third derivatives of the PPG waveform. A
population-wise correlation of 0.897 was found with LVET derived by Doppler aortic flow
velocity, and differences of 14 ± 14 ms. The strong agreement between the PPG-based LVET and
the Doppler-derived LVET on an intra-subject basis (r = 0.945 ± 0.043, −14 ± 10 ms) would
support the application of PPG to detect directional changes in LVET of an individual: a useful
feature for the early identification of progressive hypovolemia
Tusman [79] 15 Custom PPG analysis Radial catheter
In 15 cardiac surgery patients, episodes of hypotension and hypertension, defined as SAP
<90 mmHg and SAP >140 mmHg respectively, were identified manually in the arterial line
signal. The authors proposed a classification system based on the position of the dicrotic notch
and the amplitude of the PPG pulse, to differentiate between vasoconstriction and vasodilation. A
classification accuracy for hypotension and hypertension episodes of 98% was reported
(continued)
Solà [52] 35 Custom PPG analysis Oscillometric
cuff
In a protocol involving static leg extension exercises in 35 healthy subjects, PPG waveforms
acquired via a commercial smartphone camera at the fingertip were analyzed using a custom
PWA scheme. DBP was increased by 30 ± 6 mmHg on average during the protocol. After a first
initialization, an average difference of 4.0 ± 8.2 mmHg was achieved when compared to
oscillometric cuff reading, thereby suggesting the possibility of monitoring BP at very large
scales via smartphones
Coutrot [80] 61 Custom PPG analysis Radial catheter
In 61 patients undergoing general anaesthesia, the changes in perfusion index and relative
dicrotic notch height were used to detect intraoperative hypotensive events (defined as a MAP
reduction of >20%). Combining both features led to a sensitivity and specificity of 84%. The
perfusion index and the relative dicrotic notch height showed reasonable correlation (r = −0.62
and r = 0.73, respectively) with MAP variations
Ghamri [50] 40 Custom PPG analysis Radial catheter
In 40 patients undergoing general anaesthesia, PPG-based acute changes in SBP, DBP, and
MAP, defined as changes of >20% in less than 1 min, were strongly correlated (r = 0.94) and
accurately tracked (concordance rate of 100%) when compared with invasively assessed
changes. The amplitude of the changes was also accurately estimated, as assessed via Bland–
Altman analysis, with low biases (1.5%) and standard deviations (11%) for all three pressures,
and >90% of percentage errors falling below the threshold of 20%. This novel approach, which
may be used to trigger oscillometric measurements, may help to decrease the duration and
clinical consequences of hypotensive events in patients with no arterial line monitoring
Applanation Tonometry
Since the 1970s, several studies analyzed pulse waveforms obtained by applanation
tonometry and showed similarity with pressure pulse recorded intra-arterially by
catheterization [14, 15]. Since then, multiple techniques have been developed to
initialize and process the peripheral tonometry waveforms to extract different fea-
tures related to central blood pressure. Different reviews [27, 40, 41, 84] investi-
gated—through the results of multiple clinical studies (see Table 8.2)—the
correlation between PWA features with central blood pressure and clinical outcome.
The impact of the device and technique used, as well as the initialization methodol-
ogy on the accuracy of the results, were reported.
Miyashita [27] aimed at providing basic knowledge and information on different
techniques used to estimate central blood pressure using applanation tonometry. In
this context, he focused his analysis on four different techniques, namely, the gen-
eralized transfer functions (GTF) used in the SphygmoCor® device (AtCor Medical
Pty Ltd., Australia), the late or second systolic shoulder of the peripheral pressure
wave (SBP2) used in the OMRON HEM-9000AI device (OMRON Healthcare,
Japan), the N-point moving average (NPMA) used in the BPro® device with the
A-PULSE CASP® software (HealthSTATS International Pte Ltd., Singapore), and
the simple substitution method used with any carotid tonometric device, when the
carotid waveform is used as a direct surrogate of aortic waveform. Although carotid
tonometry has been shown to provide waveforms similar to central pressure wave-
form and only requires initialization without specific transformation, it is not con-
sidered as a standard method for clinics [1]. Furthermore, the possible activation of
baroreceptors in the carotid artery may affect the measurements. Concerning the
radial tonometry techniques, Miyashita’s review concluded that the NPMA tech-
nique accuracy has shown not to be superior to the GTF, because individualized
optimization is practically impossible with this technique, as it is with a GTF. GTF
and SBP2 techniques appeared even in terms of accuracy but only if central and
peripheral pressure initializations are unified. Finally, the review revealed the impact
of reproducibility of the measurement and suggested to use automated arterial
tonometry devices for more accurate results.
Cheng et al. [40] who performed a meta-analysis on 22 studies (857 subjects,
1167 measurements), also aimed at assessing the accuracy of the GTF and SBP2
methods to estimate central blood pressure, although they mainly focused on the
impact of initialization. In particular, they compared invasive and cuff-based initial-
izations, and concluded that current tonometry-based central blood pressure estima-
tion methods are acceptable when initialized with invasive aortic pressure, but
showed evident errors when initialized with brachial sphygmomanometer.
Narayan et al. [84] analyzed 164 studies using the different techniques and ini-
tialization methods mentioned before (GTF, SBP2, NPMA), but also categorized
some of these studies by disease states (hypertension, coronary disease, renal
impairment, diabetes) and analyzed their impact on central blood pressure accuracy
estimation. Their conclusions validated the previous results on the better accuracy
of invasive initialization, and did not show significant differences in central blood
pressure estimation between the different groups of patients.
In parallel with the previous studies that assessed the accuracy of central blood
pressure from a methodological point of view, Cheng et al. [41] evaluated the prog-
nostic potential of PWA based on two independent studies on large populations: the
Kinmen study including 1272 individuals with right carotid artery pulse waveforms
acquired with a tonometer, and the Cardiovascular Disease Risk Factors Two-
Township Study including 2221 individuals with central aortic pressure waveform
acquired with the SphygmoCor® device. Their findings—based on the correlation
between biomarkers derived from the pulse waveform and cardiovascular mortal-
ity—suggested that PWA could predict long-term cardiovascular risk.
Photoplethysmography
The use of PPG waveforms to estimate indicators of central blood pressure or arte-
rial stiffness has been developed using multiple approaches (see Table 8.3). One of
the major techniques considered to estimate BP-related features based on the flow-
pressure relationship is the second derivative of the photoplethysmogram, or accel-
eration plethysmogram (APG, as previously introduced in section
“Photoplethysmography-Specific Features”). Millasseau et al. [21] reviewed dif-
ferent studies that related the APG features to blood pressure, arterial properties
and aging. These are based on the pioneer work developed by investigators in Japan
which distinguishes five different waves in the APG signal and analyses the ratios
of their amplitudes (see Fig. 8.9b). All ratios (b/a, c/a, d/a, and e/a) were found to
correlate with age [37, 42, 46] when applied to large population groups, leading to
the definition of a vascular aging index as (b-c-d-e)/a, hypothesized to be useful for
screening arteriosclerosis. An alternative definition of this index was proposed by
Baek [43] as (b-e)/a when the c- and d-waves are missing. The negative ratio -d/a
showed to correlate with the aortic augmentation index and was found to be a
potentially useful way of evaluating acute effects of vasoactive agents [36, 37]. The
ratio d/a however showed only a mild correlation with blood pressure, and the cor-
relation was even less significant with the b/a ratio [42, 44–46, 81]. The b/a ratio
was however found to be a useful index of atherosclerosis and altered arterial dis-
tensibility [47]. It was further found that b/a and c/a discriminate independently
between subjects with essential hypertension and healthy controls [48]. Based on
previous studies, Millasseau et al. [21] concluded that PPG waveform analysis pro-
vides a rapid and simple means of assessing vascular tone and arterial stiffness, but
warned on the importance of employing appropriate signal conditioning since
inappropriate filtering can distort the signal, especially when using second deriva-
tive techniques. They also insisted on the fact that the measurement has to be per-
formed under standardized conditions (e.g. subjects have to be at rest when
measured). Based on a similar mathematical approach than the APG five-wave
method, a recent technique [49] also showed potential for blood pressure monitor-
ing. The recent studies performed with this new technique assessed its accuracy in
different contexts—including during induction of general anaesthesia—and
showed promising results for continuous blood pressure monitoring and detection
of acute blood pressure changes [50–52].
The Future of the Technology
Two different cuffless technologies aiming at the analysis of the pulse waveform
have been presented in this chapter. The first is based on the local measurement of
peripheral arterial pressure (applanation tonometry) and the second is based on the
measurement of the variation of the local blood volumes in the tissues using optical
measurements (photoplethysmography). The objectives of these two techniques are

the estimation of central values based on peripheral measurements by means of
PWA techniques. This section presents the directions where we think the research
on PWA is heading towards and discusses the associated challenges that will be
faced. As any prediction of the future, the discussed developments are certainly
approximate. However, the presented material represents a likely scenario that
opens a window into the future of this technology.
The main challenges in applanation tonometry are the inaccuracies involved
with initializations based on oscillometric cuffs [40, 84], and the dependency of
the observed signals to the operator [85]. While the problem of initialization
remains difficult to solve, new systems and methods need to be developed to
minimize the dependency on the operator. Such systems would make possible
and facilitate the comparison of results across different studies resulting from
different measurement campaigns. Although tonometer embedded in wristwatch-
like supports exist for measurements at the radial artery, they remain intrinsically
limited by the difficulty in reliably measuring pressure waveforms given the
small size of the location of interest. In contrast, the positioning of PPG probes at
the fingertip is easy, independent on the operator, and reproducible. However, the
analysis of pulse waveform based on PPG measurements is more challenging due
to the facts that it is a more indirect observation of the central hemodynamic
parameters and that the underlying mechanisms generating the observed signal
are only partially understood. There are a certain number of challenges that have
to be taken care off in order to obtain relevant and repeatable results. The first
challenge concerns the sensors. Optical measurements provide information about
the variations in absorption of the blood volumes that are present in the tissues
along the optical light path. In order to obtain usable signals, the system has to
provide an ad hoc contact between the sensor and the skin. Adequate optical sig-
nal quality requires that a pressure is applied by the sensor onto the skin surface.
The blood volumes that are in the vicinity of the sensor are affected by the applied
pressure (especially the venous component) and thus affects the observed signal
[19]. The challenge is therefore to design sensors such that these signal modifica-
tions are reproducible, or to develop methods that are able to adapt to the signal
variability induced by pressure changes. The second challenge is related to the
fact that the measurement site is located at the periphery of the arterial tree and
will therefore be affected by changes such as vasomotion and the control of the
autonomic nervous system. New methods have to be developed to consider their
impacts on the sensor signals in order to permit the estimation of central hemo-
dynamic features. Finally, the last identified challenge is the inter-subject vari-
ability: the observed PPG signals are impacted by the morphology and physiology
of the subject. The challenge consists in the identification the impact of these
inter-subject differences on the observed signal and to develop methods aiming in
either adapting the analysis methods to the specificities of the subject or to dis-
card them in order to recover hidden central hemodynamic characteristics such as
central blood pressure.
The second direction for the development of PWA concerns the development of
unobtrusive systems allowing the extraction of relevant information about the car-
diovascular system. The estimated physiological parameters will surely exhibit a
degraded accuracy when compared to those of reference systems. Nonetheless,
they will offer the tremendous advantage of permitting a long-term acquisition of
hemodynamic features in an unobtrusive manner. A certain number of challenges
remain open considering the application of this technology for long-term non-
invasive measurements. These challenges can be clustered into three main catego-
ries: (1) the measurement system, (2) the exogenous components in the acquired
signals and (3) the validation of long-term measurements. The first challenge, the
measurement system, consists in the development of a system that is able to oper-
ate in an “out-of-the-lab” environment. This means that such system has to be
unobtrusive, easy to use, and has to obtain measurements with a sufficient quality
to apply PWA algorithms. The development of such system will require an impor-
tant amount of work in the design of ergonomic solutions. The second challenge
consists in the determination, the identification and the separation of endogenous
components (that are related to the function of the cardiovascular system) from
exogenous components. The endogenous components are related directly to the
central hemodynamic features (e.g. diastolic and systolic blood pressure, augmen-
tation index) that PWA aims to identify. The exogenous components are of two
types. The first is related to the non-central measurement location, whose signals
will be subject to variations due to physiological and physical factors such as the
control of the autonomic nervous system, the influence of gravity, or vasomotion.
These variations will affect the local measurement and lead to erroneous estima-
tion of the central values if their effects are not corrected for. The second type of
exogenous components is due to other factors, such as motion, sensor pressure,
that can affect the estimation of central values. Finally, the third challenge that has
to be faced is the problem of the validation of the measurement in “out-of-the-lab”
conditions. The development of a standalone system for long term measurements
opens the ability to continuously measure hemodynamic parameters, but no refer-
ence system exists for the validation of such a system (the current reference for
long-term measurement is the oscillometric cuff sphygmomanometer, limited by
its accuracy, its incompatibility with continuous monitoring and its obtrusiveness,
particularly at night).
Conclusion
This chapter presents an overview of the techniques aiming at extracting relevant

haemodynamic parameters from the analysis of pulse waves. Early developments
were based on distally obtained pressure measurements (applanation tonometry)
generally obtained at the radial location. These measurement techniques are avail-
able and routinely used in clinical settings. However, the field of research in this
domain remains open and some challenges remain to be solved.
Recently two factors have opened new opportunities for research and develop-
ment of new PWA solutions and products. The first factor is the recent development
of PPG-based wearable systems that enable the acquisition of physiological signals
in ambulatory conditions such as the daily life. The second factor is the recent trans-
position of the concepts highlighted in tonometry to optical measurements. The
fusion of these two domains opens great opportunities for the measurement and the
understanding of haemodynamic variations in the early detection and follow-up of
pathological conditions.
References
1. Nichols W, O’Rourke M, Vlachopoulos C. McDonald’s blood flow in arteries: theoretical,

practical and clinical principles. Boca Raton, FL: CRC; 2011.
2. Marey EJ. Physiologie médicale de la circulation du sang basée sur l'étude graphique.
A. Delahaye; 1863.
3. Mahomed FA. The physiology and clinical use of the sphygmograph. Med Times Gazette.
1872;1:62.
4. Wellcome Collection (image credit). 2018. https://wellcomecollection.org/works/x92jruvr.
Accessed Oct 2018.
5. Riva-Rocci S. Un nuovo sfigmomanometro. Gazzetta medica di Torino. 1896;47:981–1001.
6. Shevchenko YL, Tsitlik JE. 90th anniversary of the development by Nikolai S. Korotkoff of
the auscultatory method of measuring blood pressure. Circulation. 1996;94(2):116–8.
7. Salvi P. Pulse waves. How vascular hemodynamics affects blood pressure. Berlin: Springer;
2017.
8. Tiru B, Bloomstone JA, McGee WT. Radial artery cannulation: a review article. J Anesth Clin
Res. 2012;3(5):1000209.
9. Mitchell GF. Arterial stiffness and wave reflection: biomarkers of cardiovascular risk. Artery
Res. 2009;3(2):56–64.
10. Vlachopoulos C, Aznaouridis K, Stefanadis C. Prediction of cardiovascular events and all-
cause mortality with arterial stiffness: a systematic review and meta-analysis. J Am Coll
Cardiol. 2010;55(13):1318–27.
11. McEniery CM, Cockcroft JR, Roman MJ, et al. Central blood pressure: current evidence and
clinical importance. Eur Heart J. 2014;35(26):1719–25.
12. Roman MJ, Devereux RB, Kizer JR, et al. Central pressure more strongly relates to vascu-
lar disease and outcome than does brachial pressure: the Strong Heart Study. Hypertension.
2007;50(1):197–203.
13. Drzewiecki GM, Melbin J, Noordergraaf A. Arterial tonometry: review and analysis.

J Biomech. 1983;16(2):141–52.
14. Chen CH, Ting CT, Nussbacher A, et al. Validation of carotid artery tonometry as a means of
estimating augmentation index of ascending aortic pressure. Hypertension. 1996;27(2):168–75.
15. Kelly R. Noninvasive carotid pressure wave registration as an indicator of ascending aortic
pressure. J Vasc Med Biol. 1989;1:241–7.
16. Matthys K, Verdonck P. Development and modelling of arterial applanation tonometry: a
review. Technol Health Care. 2002;10(1):65–76.
17. Pauca AL, O’Rourke MF, Kon ND. Prospective evaluation of a method for estimating ascend-
ing aortic pressure from the radial artery pressure waveform. Hypertension. 2001;38(4):932–7.
18. Salvi P, Lio G, Labat C, et al. Validation of a new non-invasive portable tonometer for deter-
mining arterial pressure wave and pulse wave velocity: the PulsePen device. J Hypertens.
2004;22(12):2285–93.
19. Reisner A, Shaltis PA, McCombie D, Asada HH. Utility of the photoplethysmogram in circula-
tory monitoring. Anesthesiology. 2008;108(5):950–8.
20. Millasseau SC, Guigui FG, Kelly RP, et al. Noninvasive assessment of the digital volume
pulse: comparison with the peripheral pressure pulse. Hypertension. 2000;36(6):952–6.
21. Millasseau SC, Ritter JM, Takazawa K, Chowienczyk PJ. Contour analysis of the photople-
thysmographic pulse measured at the finger. J Hypertens. 2006;24(8):1449–56.
22. Allen J. Photoplethysmography and its application in clinical physiological measurement.
Physiol Meas. 2007;28(3):R1.
23. Sazonov E, Neuman MR, editors. Wearable sensors: fundamentals, implementation and appli-
cations. Amsterdam: Elsevier; 2014.
24. Westerhof N, Stergiopulos N, Noble MI. Snapshots of hemodynamics: an aid for clinical
research and graduate education. Cham: Springer Science & Business Media; 2010.
25. Murgo JP, Westerhof N, Giolma JP, ltobelli SA. Aortic input impedance in normal man: rela-
tionship to pressure wave forms. Circulation. 1980;62(1):105–16.
26. Karamanoglu MMAR, O’Rourke MF, Avolio AP, Kelly RP. An analysis of the relation-
ship between central aortic and peripheral upper limb pressure waves in man. Eur Heart
J. 1993;14(2):160–7.
27. Miyashita H. Clinical assessment of central blood pressure. Curr Hypertens Rev.

2012;8(2):80–90.
28. London G, Guerin A, Pannier B, Marchais S, Benetos A, Safar M. Increased systolic pressure
in chronic uremia. Role of arterial wave reflections. Hypertension. 1992;20(1):10–9.
29. Hausvater A, Giannone T, Sandoval YHG, et al. The association between preeclampsia and
arterial stiffness. J Hypertens. 2012;30(1):17–33.
30. Khalil A, Jauniaux E, Harrington K. Antihypertensive therapy and central hemodynamics in
women with hypertensive disorders in pregnancy. Obstet Gynecol. 2009;113(3):646–54.
31. Elgendi M. On the analysis of fingertip photoplethysmogram signals. Curr Cardiol Rev.
2012;8(1):14–25.
32. Chua CP, Heneghan C. Continuous blood pressure monitoring using ECG and finger photople-
thysmogram. In: Engineering in Medicine and Biology Society, 2006. EMBS’06. 28th Annual
International Conference of the IEEE; 2006. p. 5117–5120.
33. Padilla JM, Berjano EJ, Saiz J, et al. Assessment of relationships between blood pressure,
pulse wave velocity and digital volume pulse. In: Computers in cardiology, 2006. Lyon: IEEE;
2006. p. 893–6.
34. Wang L, Pickwell-MacPherson E, Liang YP, Zhang YT. Noninvasive cardiac output estima-
tion using a novel photoplethysmogram index. In: Proceedings of 31st IEEE EMBS, 2009.
p. 1746–1749.
35. Millasseau SC, Kelly RP, Ritter JM, Chowienczyk PJ. Determination of age-related increases
in large artery stiffness by digital pulse contour analysis. Clin Sci. 2002;103(4):371–7.
36. Takazawa K. Clinical usefulness of the second derivative of a plethysmogram (acceleration
plethysmogram). J Cardiol. 1993;23:207–17.
37. Takazawa KTN, Fujita M, Matsuoka O, et al. Assessment of vasocative agents and vas-
cular aging by the second derivative of photoplethysmogram waveform. Hypertension.
1998;32:365–70.
38. Mukkamala R, Hahn JO. Toward ubiquitous blood pressure monitoring via pulse transit time:
predictions on maximum calibration period and acceptable error limits. IEEE Trans Biomed
Eng. 2018;65(6):1410–20.
39. Mukkamala R, Hahn JO, Inan OT, et al. Toward ubiquitous blood pressure monitoring via
pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62(8):1879–901.
40. Cheng H-M, Lang D, Tufanaru C, Pearson A. Measurement accuracy of non-invasively
obtained central blood pressure by applanation tonometry: a systematic review and meta-
analysis. Int J Cardiol. 2013;167(5):1867–76.
41. Cheng HM, Chuang SY, Wang JJ, et al. Prognostic significance of mechanical biomarkers
derived from pulse wave analysis for predicting long-term cardiovascular mortality in two
population-based cohorts. Int J Cardiol. 2016;215:388–95.
42. Bortolotto LA, Blacher J, Kondo T, et al. Assessment of vascular aging and atherosclerosis in
hypertensive subjects: second derivative of photoplethysmogram versus pulse wave velocity.
Am J Hypertens. 2000;13:165–71.
43. Baek HJ, Kim JS, Kim YS, et al.. Second derivative of photoplethysmography for estimating
vascular aging. In: Information technology applications in biomedicine, 2007. ITAB 2007. 6th
International Special Topic Conference on IEEE; 2007. p. 70–72.
44. Hashimoto J, Chonan K, Aoki Y, et al. Pulse wave velocity and the second derivative of the
finger photoplethysmogram in treated hypertensive patients: their relationship and associating
factors. J Hypertens. 2002;20(12):2415–22.
45. Hashimoto J, Watabe D, Kimura A, et al. Determinants of the second derivative of the fin-
ger photoplethysmogram and brachial-ankle pulse-wave velocity: the Ohasama study. Am
J Hypertens. 2005;18(4):477–85.
46. Takada H, Washino K, Harrell JS, Iwata H. Acceleration plethysmography to evaluate aging
effect in cardiovascular system. Using new criteria of four wave patterns. Med Prog Technol.
1996;21(4):205–10.
47. Imanaga I, Hara H, Koyanagi S, Tanaka K. Correlation between wave components of the sec-
ond derivative of plethysmogram and arterial distensibility. Jpn Heart J. 1998;39:775–84.
48. Simek J, Wichterle D, Melenovsky V, et al. Second derivative of the finger arterial pressure
waveform: an insight into dynamics of the peripheral arterial pressure pulse. Physiol Res.
2005;54(5):505.
49. Proença M, Solà J, Lemay M, Verjus C. Method, apparatus and computer program for deter-
mining a blood pressure value. WO 2016 138965 A1, 9th of September 2016; 2016.
50. Ghamri Y, Proença M, Hofmann G, et al. Using pulse oximetry waveform analysis to track
changes in blood pressure during anesthesia induction. 2019. Manuscript in preparation.
51. Solà J, Proença M, Braun F, et al. Continuous non-invasive monitoring of blood pressure in
the operating room: a cuffless optical technology at the fingertip. Curr Direct Biomed Eng.
2016;2(1):267–71.
52. Solà J, Proença M, Schoettker P, et al. Blood pressure monitoring using a smartphone cam-
era: performance of the OBPM technology. In: IEEE Conference on Biomedical and Health
Informatics; 2018.
53. Pauca AL, Kon ND, O’Rourke MF. The second peak of the radial artery pressure wave
represents aortic systolic pressure in hypertensive and elderly patients. Br J Anaesth.
2004;92(5):651–7.
54. Soderstrom S, Nyberg G, O'Rourke MF, et al. Can a clinically useful aortic pressure wave be
derived from a radial pressure wave? Br J Anaesth. 2002;88:481–8.
55. Davies JI, Band MM, Pringle S, et al. Peripheral blood pressure measurement is as good
as applanation tonometry at predicting ascending aortic blood pressure. J Hypertens.
2003;21(3):571–6.
56. Sharman JE, Lim R, Qasem AM, et al. Validation of a generalized transfer function to nonin-
vasively derive central blood pressure during exercise. Hypertension. 2006;47(6):1203–8.
57. Chen CH, Nevo E, Fetics B, et al. Estimation of central aortic pressure waveform by math-
ematical transformation of radial tonometry pressure: validation of generalized transfer func-
tion. Circulation. 1997;95(7):1827–36.
58. Hope SA, Tay DB, Meredith IT, Cameron JD. Use of arterial transfer functions for the deriva-
tion of aortic waveform characteristics. J Hypertens. 2003;21(7):1299–305.
59. Hope SA, Meredith IT, Tay D, Cameron JD. Generalizability’ of a radial-aortic transfer function
for the derivation of central aortic waveform parameters. J Hypertens. 2007;25(9):1812–20.
60. Rajani R, Chowienczyk P, Redwood S, et al. The noninvasive estimation of central aortic blood
pressure in patients with aortic stenosis. J Hypertens. 2008;26(12):2381–8.
61. Hickson SS, Butlin M, Mir FA, et al. The accuracy of central SBP determined from the second
systolic peak of the peripheral pressure waveform. J Hypertens. 2009;27(9):1784–8.
62. Hope SA, Meredith IT, Cameron JD. Effect of non-invasive calibration of radial wave-
forms on error in transfer-function-derived central aortic waveform characteristics. Clin Sci.
2004;107(2):205–11.
63. Smulyan H, Siddiqui DS, Carlson RJ, et al. Clinical utility of aortic pulses and pressures cal-
culated from applanated radial-artery pulses. Hypertension. 2003;42:150–5.
64. Zuo JL, Li Y, Yan ZJ, et al. Validation of the central blood pressure estimation by the
SphygmoCor system in Chinese. Blood Press Monit. 2010;15(5):268–74.
65. Ding FH, Fan WX, Zhang RY, et al. Validation of the noninvasive assessment of central blood
pressure by the SphygmoCor and Omron devices against the invasive catheter measurement.
Am J Hypertens. 2011;24(12):1306–11.
66. Takazawa K, Kobayashi H, Shindo N, et al. Relationship between radial and central arte-
rial pulse wave and evaluation of central aortic pressure using the radial arterial pulse wave.
Hypertens Res. 2007;30(3):219.
67. Williams B, Lacy PS, Yan P, et al. Development and validation of a novel method to derive
central aortic systolic pressure from the radial pressure waveform using an N-point moving
average method. J Am Coll Cardiol. 2011;57(8):951–61.
68. Garcia-Ortiz L, Recio-Rodríguez JI, Canales-Reina JJ, et al. Comparison of two measuring
instruments, B-pro and SphygmoCor system as reference, to evaluate central systolic blood
pressure and radial augmentation index. Hypertens Res. 2012;35(6):617.
69. Westerhof BE, Guelen I, Stok WJ, et al. Individualization of transfer function in estimation
of central aortic pressure from the peripheral pulse is not required in patients at rest. J Appl
Physiol. 2008;105(6):1858–63.
70. Shih YT, Cheng HM, Sung SH, et al. Quantification of the calibration error in the transfer
function-derived central aortic blood pressures. Am J Hypertens. 2011;24(12):1312–7.
71. Cheng HM, Wang KL, Chen YH, et al. Estimation of central systolic blood pressure using an
oscillometric blood pressure monitor. Hypertens Res. 2010;33(6):592.
72. Van Bortel LM, Balkestein EJ, van der Heijden-Spek JJ, et al. Non-invasive assessment of local
arterial pulse pressure: comparison of applanation tonometry and echo-tracking. J Hypertens.
2001;19:1037–44.
73. Karamanoglu MMAR, Feneley MP. Derivation of the ascending aortic-carotid pressure trans-
fer function with an arterial model. Am J Phys Heart Circ Phys. 1996;271(6):H2399–404.
74. Kelly R, Fitchett D. Noninvasive determination of aortic input impedance and external left
ventricular power output: a validation and repeatability study of a new technique. J Am Coll
Cardiol. 1992;20(4):952–63.
75. Otsuka T, Kawada T, Katsumata M, Ibuki C. Utility of second derivative of the finger photople-
thysmogram for the estimation of the risk of coronary heart disease in the general population.
Circ J. 2006;70(3):304–10.
76. Otsuka T, Kawada T, Katsumata M, et al. Independent determinants of second derivative of the
finger photoplethysmogram among various cardiovascular risk factors in middle-aged men.
Hypertens Res. 2007;30(12):1211.
77. Kawada T, Otsuka T. Factor structure of indices of the second derivative of the finger
Photoplethysmogram with metabolic components and other cardiovascular risk indicators.
Diabetes Metab J. 2013;37(1):40.
78. Chan GS, Middleton PM, Celler BG, et al. Automatic detection of left ventricular ejection time
from a finger photoplethysmographic pulse oximetry waveform: comparison with Doppler
aortic measurement. Physiol Meas. 2007;28(4):439.
79. Tusman G, Acosta CM, Pulletz S, et al. Photoplethysmographic characterization of vascular
tone mediated changes in arterial pressure: an observational study. J Clin Monit Comput. 2018.
80. Coutrot M, Joachim J, Dépret F, et al. Noninvasive continuous detection of arterial hypoten-
sion during induction of anaesthesia using a photoplethysmographic signal: proof of concept.
Br J Anaesth. 2019;122(5):605–12.
81. Avolio A. The finger volume pulse and assessment of arterial properties. J Hypertens.

2002;20(12):2341–43.
82. Avolio AP, Butlin M, Walsh A. Arterial blood pressure measurement and pulse wave analy-
sis—their role in enhancing cardiovascular assessment. Physiol Meas. 2010;31(1):R1.
83. Wilkinson IB, MacCallum H, Flint L, et al. The influence of heart rate on augmentation index
and central arterial pressure in humans. J Physiol. 2000;525(1):263–70.
84. Narayan O, Casan J, Szarski M, et al. Estimation of central aortic blood pressure: a systematic
meta-analysis of available techniques. J Hypertens. 2014;32(9):1727–40.
85. Hirata K, Kawakami M, O'Rourke MF. Pulse wave analysis and pulse wave velocity. Circ
J. 2006;70(10):1231–9.
86. Cloud GC, Rajkumar C, Kooner J, et al. Estimation of central aortic pressure by SphygmoCor®
requires intra-arterial peripheral pressures. Clin Sci. 2003;105(2):219–225.
87. Gunarathne A, Patel JV, Hughes EA, Lip GY. Measurement of stiffness index by digital vol-
ume pulse analysis technique: clinical utility in cardiovascular disease risk stratification. Am
J Hypertens. 2008;21(8):866–72.
Chapter 9
Machine Learning Techniques
Xiaorong Ding
Abstract Driven by the exponential growth in the computational power and the
increasing size of the collected data sets, there has been growing interest in using
data-driven approaches based on machine learning techniques to resolve the prob-
lems and overcome the challenges that facing the area of cuffless blood pressure
measurement. Compared with the theory-driven analytical approaches, the machine
learning method is very promising with its ability to learn the function of the com-
plex system if the model is trained well, and to address the latent affecting factors
that cannot be considered in the analytical model. This chapter first addresses the
motivation of employing data-driven method, then provides a brief introduction of
machine learning method for cuffless blood pressure estimation. It then presents
some of the state-of-the-art examples and applications of such technology and
finally discusses the outlook of its future development.
Keywords Machine learning · Cuffless blood pressure · Data-driven method ·

Analytical model · Pulse transit time · Blood pressure indicator · Learning model ·
Deep learning
Introduction
Ever since 1980s, various research efforts have been devoted to the development of
cuffless techniques for blood pressure (BP) measurement, which is to obtain the
continuous BP reading indirectly and continuously without the traditional inflatable
cuff. The classical way is to calibrate the indicators that can reveal the BP changes
to the BP values, to get the indirect BP estimation with the measurement of the
indicators. Those indicators are usually the features that are available from non-
invasive cardiovascular signal recordings. The cardiac signals can be vital signs
such as the electrocardiogram (ECG) and the photoplethysmogram (PPG). Pulse
X. Ding (*)
Institute of Biomedical Engineering, Department of Engineering Science,
University of Oxford, Oxford, UK
e-mail: xiaorong.ding@eng.ox.ac.uk

140 X. Ding
transit time (PTT), for example, is one of the most common technical indicators and
has been widely used for the estimation of the cuffless BP. To get the indirect BP
from the indicator(s), there is a need to map the indicator(s) to the BP via calibration
algorithm that can model the relationship between the indicator(s) and BP.
In classical calibration, the PTT is projected to the BP values through explicit
theoretical expressions based on the underlying physiological mechanism (e.g. the
principle of pulse wave velocity propagation (PWV)). The readers may refer to
Chap. 7 for more details of the PWV techniques. When there are more BP indicators
being involved, and the interpretations of the relationship between these indicators
and BP are not always straightforward, BP is regressed on these indicators for a set
number of known sample subjects, and the regression models are used to estimate
the BP values for an unknown subject from its measured indicators. However, there
are too many variables relating to the BP changes, and the relationship between
each variable and BP are too complicated to be expressed via simple physiologi-
cally based mathematic model.
As data sets get bigger and computers become more powerful, the data-driven
approaches based on machine learning techniques have been attempted to model the
relationship between BP indicator variables and BP. In this chapter, we will first
address the motivation of employing data-driven method, following by the brief
introduction of machine learning for cuffless BP estimation. Finally, we will illus-
trate some of the state-of-the-art examples and applications of such technology and
present the outlook of its future development.
Motivation of Using Data-Driven Method
Model-Based Method
Most PTT-based cuffless BP measurement approaches are based on the principle of

PWV recording through the Moens–Korteweg (M-K) equation [1]:
Eh
PWV = (9.1)
ρD

which correlates PWV with the modulus of elasticity of the artery E, the thickness
of the arterial wall h, the diameter of the artery D and the blood density ρ. Since the
changes of ρ are negligible, ρ can be assumed to be constant. Then, E is the major
factor that PWV relies on, among the other three variables (i.e., E, h, and D). Further
according to Hughes et al. [2], the elastic modulus E is exponentially correlated
with the mean distending pressure P as given by Eq. (9.2):
9 Machine Learning Techniques 141
E = E 0 eγ P (9.2)

where E0 is the zero-pressure modulus, and γ a constant typically between 0.016 and
0.018. Both E0 and γ are site and individual dependent. The combination of Eqs.
(9.1) and (9.2) thus gives the relationship between PWV and arterial BP as given by
Eq. (9.3):
hE0 eγ P
PWV = . (9.3)
ρD

As PWV is reciprocally related with PTT as given by Eq. (9.4):
L
PWV = (9.4)
PTT
where L represents the transit distance of the pressure waveform traveling between
two arterial sites, the arterial pressure P can be derived accordingly with Eq. (9.4)
substituted in Eq. (9.3), as given by Eq. (9.5).
1 ρ L2 D 
P=  −2 ln PTT + ln  (9.5)
γ hE0 

Therefore, with initial calibration of PTT to BP, beat-to-beat BP can be theoretically
derived from beat-to-beat PTT that can measured from each cycle of the cardiac
pulse signals.
The Relationship Between PTT and BP
Since Weltman’s seminal work on PTT in 1964 [3], a considerable number of studies
have focused on the application of PTT in psychophysiological research in 1970s, as
PTT is an indirect continuous measure of BP variation. At that period, the non-inva-
sive continuous recording of arterial BP was not yet available due to many unsolved
physiological and technical problems [4]. During 1970s–1980s, there have been active
research efforts on the relationship between PTT and BP, aiming to investigate whether
PTT is feasible to evaluate BP changes [3–22]. Since 2000, the study on PTT for BP
estimation has attracted more and more attention [23]. The fundamental of the appli-
cations of PTT for BP measurement is its relationship with BP. The methods and
results of the studies about PTT-BP relationship are briefly summarized in Table 9.1.
Those studies have investigated the relationship between PTT and BP with
experimental stressors or clinical interventions to elicit changes in BP. At the early
stage of PTT study in 1980s, some investigators pointed out to use arterial PWV as
a continuous measure of BP changes, and they carried out experiment to assess the
relationship between PTT and mean BP (MBP). Gribbin et al. [6] measured
142 X. Ding
Table 9.1 Summary of the studies about the relationship between PTT and BP
Correlation coefficient
Reference N Condition BP measurement SBP DBP MBP
[6] 26 Externally applied pressures Intra-arterial – – 0.92–
0.99
[8] 14 Rest, cold pressor, pornographic Non-invasive −0.85 −0.30 –
7 movie, unsignalled shock Inter-arterial −0.76 −0.44 –
[11] 94 Unsignalled shock, Stroop Sphygmomanometer −0.49 −0.03 –
color–word interference test,
isometric handgrip
[12] 4 Rest, paced respiration, and Intra-arterial −0.62 −0.32 −0.48
mental arithmetic
[17] 44 Rest Sphygmomanometer −0.95 0.09 –
[18] 14 Lower body negative pressure Intra-arterial 0.80 – –
[19] 18 Bicycle test Sphygmomanometer >0.80 – –
[20] 20 Drug administration (glyceryl Intra-arterial −0.62 −0.14 −0.28
trinitrate, angiotensin II,
norepinephrine, salbutamol)
[22] 41 Exercise Sphygmomanometer −0.92 −0.38 –
brachial-to-radial PWV in humans and correlated it with MBP changes. The PWV
was found to change linearly with mean pressure. Similar study by Steptoe et al. [7]
reported that the changes in PTT was dependent on arterial pressure changes which
were caused by manoeuvres including mental arithmetic and isometric exercise.
The PTT that calculated from ECG and peripheral pulse, was linearly correlated
with mean arterial pressure, with correlation coefficients varying between −0.91
and −0.98 for five subjects. After that, there have been studies examining the cor-
relation of PTT with systolic BP (SBP) and diastolic BP (DBP). As can be seen
from Table 9.1, PTT had stronger correlation with SBP than DBP for almost all the
studies [8, 9, 11–13, 17].
In addition, although the correlations between PTT and BP depend upon, how
PTT and BP were measured, the method of recording and BP changes induction
method, and whether the subjects displayed differences in BP reactivity, the results
of those studies reveal that: although most of the studies indicate the changes in PTT
can reliably track the variations in BP, which support the usage of PTT as an indirect
measure of BP change, majority of the studies show that the correlation coefficients
between PTT and BP vary from individual to individual, which means that in any
one subject an absolute change in arterial pressure is not exclusively due to change
in PTT.
Modelling Methods of PTT-Based BP Estimation
Based on the theoretical relationship between PTT and BP as mentioned earlier, as

well as their experimental or empirical relationship, a wide variety of research stud-
ies have been conducted to model PTT with BP to achieve BP estimation. Table 9.2
summarizes the studies about PTT-based cuffless BP measurement in the recent two
decades. The findings of Table 9.2 indicate that the PTT-BP models can be roughly
classified into three types: (1) physiological models derived based on M-K equation
or B-H equation, (2) linear or nonlinear regression models, which account for more
than 70% of these studies, and (3) empirical model with parameters identified from
a population.
The studies of the relationship between BP-signatures and BP, as well as the
models that are used to represent their relationships, demonstrate that the associa-
tion of the BP-signature and BP is far more complicated than the simple liner or
nonlinear regression model. Taking PTT for example, for different individuals it
either has positive or negative, strong or weak correlation with BP. Furthermore,
most of the analytical models work on the premise of some assumptions. For exam-
ple, the M-K equation assumes that the ratio of the wall thickness to the vessel
radius and blood density keep constant, and the artery wall is presumed to be isotro-
pic and there are volumetric changes between the artery wall and the pulse pressure.
Obviously, these assumptions do not conform to the real situation. Furthermore,
intrinsic or extrinsic factors like age, temperature, mental stress, and different
behaviour pattern would also affect the BP changes. Theoretically, all the factors
that may impact the variations of BP should be considered into the analytical model
to achieve accurate BP estimate. But in practice, it is impossible to include all these
elements with engineering implementation, such as the wearable systems. All these
disadvantages of the analytical model drive the application of the data-driven learn-
ing method to build the function for the complicated relationship between the indi-
cators and BP. The method will accordingly be elaborated in the following section.
Data-Driven Approaches Based on Machine Learning
In a nutshell, machine learning is a subset of artificial intelligence in the field of

computer science. It often uses statistical techniques to give computers the ability to
“learn” with data, without being explicitly programmed [34]. Given the exponential
growth in the computational power and the increasing size of the collected data sets,
there has been growing interest in using big data analysis techniques to resolve the
problems and overcome the challenges that facing the area of cuffless BP measure-
ment [35]. The data-driven based machine learning method is thus very promising
with its ability to learn the function of the complex system if the model is trained
well, and to address the latent affecting factors that cannot be considered in the
analytical model.
Table 9.2 Summary of the studies about PTT based cuffless BP measurement
144
First author (year, 2.1.4. Calibration (method/ 2.1.6. Accuracy (mmHg)

country) 2.1.3. Method (model) interval)/reference method 2.1.5. Subjects SBP DBP
Young (1995, US) SBP = a1/PTT + b1 Oscillometric BP (5 min)/ 35 patients −0.37 −0.01
[24] DBP = a2/PTT + b2 invasive radial BP (−29.0–28.2) (−14.9–14.8)
Chen (2000, Intermittent BP from 20 patients RMSE:
Japan) [25] 2 invasive BP (5 min)/ 3.70 ± 1.85
SBP = SBP 0 − ( PTT − PTT0 ) invasive radial arterial BP
γ PTT0
Fung (2004, Cuff sphygmomanometry 22 patients −0.08 ± 11.32

Canada) [26] a (10 min)/cuff
BP = +b
PTT 2 sphygmomanometry
Poon (2005, Cuff BP (initial 85 (39 0.6 ± 9.8 0.9 ± 5.6

China) [27] 2 calibration)/average of hypertensive)
2 PTT0 1  PTT0 
DBP = MBP0 + ln − PP0 ⋅   auscultatory and
γ PTT 3  PTT  oscillometric BP
2
 PTT0 
SBP = DBP + PP0 ⋅  
 PTT 
Muehlsteff (2006, SBP = a∗lnPTT + b Cuff BP (physical test)/ 18 healthy RMSE: 7.5
Germany) [19] SBP = a∗L/PTT + b cuff BP RMSE: 6.9
SBP = a∗(L/PTT)2 + b RMSE: 7.3
Wong (2009, SBP = a1∗PTT + b1 Oscillometric BP (exercise, 41 healthy (14 for 0.0 ± 5.3 0.0 ± 2.9
China) [22] DBP = a2∗PTT + b2 half year)/oscillometric BP half year) Half year: Half year:
1.4 ± 10.2 2.1 ± 7.3
Cattivelli (2009, SBP = a1∗PTT + b1∗HR + c1 MIMIC database (invasive 25 ICU patients −0.41 ± 7.77 −0.07 ± 4.96
US) [28] DBP = a2∗PTT + b2∗HR + c2 BP, 1 h) / invasive BP
X. Ding
Gesche (2011, Sphygmomanometry (13 63 SD: 10.10
Germany) [29] ( p3×PWV ) subjects, exercise)/ 95% CI:
BPPTT = P1 × PWV × e + P2 ×
sphygmomanometry −19.8~19.8
PWV P4 − ( BPPTT ,cal − BPcal )
Chen (2012, kij

Intra-arterial BP (9 35 healthy 1.49 ± 6.51 2.16 ± 6.23
Singapore) [30] − subjects)/intra-arterial BP
PWVs
SBP = bij e
kij
−
PWVd
DBP = bij e
9 Machine Learning Techniques
Ding (2016, Finapres BP 27 healthy −0.37 ± 5.21 −0.08 ± 4.06

China) [31] 2
PIR 0  PTT0 
SBP = DBP0 ⋅ + PP0 ⋅  
PIR  PTT 
PIR 0
DBP = DBP0 ⋅
PIR
Huynh (2018) [32] Oscillometric BP 15 healthy RMSE: RMSE:
2 8.47 ± 0.91 5.02 ± 0.73
 D 
SBP = DBP0 + B ⋅   ln (1 + K ( Z max 0 − Z max ) )
 PTT 
2
 D 
DBP = DBP0 + B ⋅   ln (1 + K ( Z max 0 − Z min ) )
 PTT 
Liu (2018) [33] MBP = HR∗(k1∗DRPPG TD + b1) Finapres BP 20 healthy SD: 2.85 SD: 1.75
PP = MBP∗(k2∗t/HP + b2)
SBP = MBP + 2/3PP; DBP = MBP-1/3PP
145
146 X. Ding
Compared with the theory-driven analytical approaches, there is no need for the
machine learning method to understand the physiological process underlying the
prediction and it is without priori assumption. For the data-driven approaches for
the cuffless BP estimation, the data can signal that are informative of the BP changes,
for example, the ECG, PPG, or any other biomedical signals that can be collected in
a non-invasive and unobtrusive manner. The machine learning method or in particu-
lar the deep learning is able to extract meaningful features from these signals, and
model the relationship between the learned features and BP with the objective to
minimize the difference between the predicated value and the reference.
The main feature and value of using the machine learning method for cuffless BP
estimation lies in its learning ability. To achieve accurate estimation of cuffless BP
with analytical model, it is often necessary to identify correlations between multiple
feature inputs and external factors that are rapidly producing millions of data points.
The analytical model is then built on past data and relying on the expert knowledge
to establish a relationship between the variables. While the machine learning method
can learn constantly from the data. It starts with the outcome variables, that is, BP,
and then automatically looks for predictor variables and their interactions. When the
target output is clear but the important input variables are unknown to make the
decision, the machine learning becomes valuable by giving it the goals and then it
learns from the data which factors are important in achieving that goal. In compari-
son to the static analytical model, the machine learning algorithms can constantly
improve over time as more data is collected and assimilated.
heory-Driven Analytical Model Versus the Data-Driven

T
Learning Model
There are three major differences between the theory-driven analytical model and
the data-driven learning model, in terms of generalizability, interpretability and
reliability.
First, the theory-driven method is generalizable, while the learning model is very
hard to generalize beyond the original problem. The learning model is often achieved
on a training dataset with the objective to solve a very specific problem. Usually for
a new problem, new information and datasets that relate to that problem is required.
Second, the analytical model is interpretable, but the machine learning model is
usually inexplicable. One prominent feature of the learning method is that it is a
black box method. The learning is driven by a well-defined objective function, with-
out any prior assumptions and the understanding of the process. The analytical
model, on the contrast, usually works under some hypothesis and requires the expert
knowledge, which makes it understandable.
Third, the analytical is light weight, while the learning model is computationally
demanding.
The major differences between these two types of methods are listed in Table 9.3.
Table 9.3 Data-driven learning method versus model-based method

Theory-driven analytical model Data-driven learning model
1 Generalizable Hard to generalize beyond the original problem
2 Interpretable No understanding of the process
3 Hypothesis No need for priori assumptions
4 Need expert opinion Learn from data
5 Light weight Computationally demanding
rief Introduction of Machine Learning for Cuffless BP

B
Prediction
Short Historical Overview
With the development of the machine learning algorithm, and its successful exploi-
tation various areas such as speech recognition, natural language processing and
computer vision, there are more and more efforts towards the field of cuffless BP
estimation or prediction.
The attempt of machine learning method for the arterial BP estimation started
since 2000s. One of the earliest studies as far as we are concerned is the one in pub-
lished 2005 by Kim and the colleagues [36]. In that study, three features such as the
PTT, weight and arm length of each subject, were used as the inputs to a two hidden-
layer artificial neural network for the estimation of SBP. It has achieved an estima-
tion error of 4.53 ± 2.68 mmHg, which outperformed the method of multiple. There
are very few studies that have used machine learning for the cuffless BP estimation
until 2011 Monte-Moreno has proposed to use machine learning algorithm to esti-
mate BP from a PPG waveform [37]. It is driven by the idea that there is functional
relationship between the morphology of the PPG waveform and the BP. With the
energy profile of the PPG signal being extracted as the inputs, several machine
learning techniques have been tested, including ridge linear regression, a multilayer
perceptron neural network, support vector machines and random forests (Fig. 9.1).
The best performance was by means of the Random Forest Tree method, which has
resulted a coefficient of determination between the reference and the prediction of
0.91 and 0.89 for SBP and DBP, respectively. The performance also was complying
with the Grade B protocol of the BHS.
Most of the machine learning based studies use features extracted from PPG
signals [38–41], since the PPG signal is the blood volume change that caused mainly
by the BP. Almost all the studies have been using machine learning to model the
relationship between feature variables and the target variable—BP. Regarding the
input features, the hand-designed features from the PPG signal are the most com-
mon type of inputs, including the time-domain profiles such as the amplitude, inter-
val, intensity, and area, and the frequency-domain features like the power, peak
frequency, and maximum amplitude. There are a few studies using the whole wave-
148 X. Ding
Fig. 9.1 Diagram of the system of machine learning based cuffless BP estimation [37]
form or the whole waveform-based features [37, 41], with most of these studies
using both ECG and PPG signals [42–45].
In the early 2010s, the simple regression methods have been commonly used,
such as the univariate or multivariate linear regression [44, 46] Later, other algo-
rithms such as support vector regression [39, 47] and random forest [37, 42, 48]
have been employed to develop the BP model. With the development of the machine
learning algorithm, and its successful application in various areas such as speech
recognition, natural language processing and computer vision, researchers began to
develop models based on the advanced machine learning method such as deep
learning method. For example, multiple neural perceptron and neural network have
been employed to develop the nonlinear relationship between the predictors and BP
[38, 49]. In the recent years, the long-short-term-memory (LSTM) architecture of a
recurrent neural network (RNN) has been studied and achieved better performance
than classical machine learning method, due to its advantage of accommodating the
multiscale temporal dependency between the sequential raw signal values and the
corresponding BP values [50–52].
However, there are few studies involving the use of the real sense of “deep learn-
ing” method for estimating cuffless BP. Deep learning often involves the representa-
tion learning of the features from raw data rather than using handcrafted features.
Although one study mentioned using deep learning [53], it does not contain the
representative learning.
How Does Machine Learning Work for Cuffless BP Estimation?
While the analytical model-based method maps BP indicators to BP via a generative

mathematical model, with some assumptions or hypothesis, the machine learning
based regression method is able to predict the BP values by a complicated model
that can be learnt from the data with specific learning algorithm. Simply speaking,
the machine learning algorithm works to learn a target function (f) that best maps
input variables (X) to an output variable (Y): Y = f(X). The general learning task is
usually to make predictions in the future (Y) given new examples of input variables
(X). This is also called predictive modelling or predictive analytics and our goal is
to make the most accurate predictions possible.
For the application of cuffless BP estimation, the handcrafted features, such as
PTT and heart rate, or other features that are extracted from cardiac signals such as
ECG and PPG signal are used as the input variable for the learning of the relation-
ship between those input variables and the output—BP. More advanced methods,
like deep learning, can also be employed to extract representative features from the
raw signals. For example, autoencoder is one representative architecture for the
abstract feature learning. Figure 9.2 illustrates the machine learning method for the
cuffless BP measurement.
Classical Machine Learning Method for Cuffless BP Estimation
The indirect estimation of BP from the features or signals is a regression problem

by nature. In statistics, the problem of regression is that of learning a function that
allows to estimate a certain quantity of interest, the dependent variable, from serval
observed variables, known as covariates, features or independent variables. For the
case of cuffless BP estimation, we are interested in estimating BP based on its indi-
cators (e.g. the cardiac output, peripheral resistance). The function that models the
Fig. 9.2 Diagram to show the machine learning method for the cuffless BP measurement, where
the boxes with red orange indicate the components that can learn from data
150 X. Ding
relationship between the response and the predictors is learnt from training data and
can then be used to predict the response for new collected data.
Traditional calibration of the PTT-BP model uses only one or two calibration
points to get the regression coefficients, which would not be accurate due the reason
we explained earlier. With a big training dataset, it is assumed that the coefficients
of the regression model would be more accurate. Commonly used classical machine
learning methods include linear regression, support vector machine regression, ran-
dom forest, and adaptive regression, which will be elaborated as below.
Linear and Polynomial Regression
Linear regression is by far the most simple and popular example of a regression
algorithm. The univariate linear regression is the simplest technique used to model
the relationship between a single input independent variable (feature variable) X and
an output dependent variable BP using a liner model [54]:
BP = β 0 + β1 X + ε (9.6)

For the application of cuffless BP measurement, there are studies in the 2000s taken
PTT as the sole feature variable and modelled the relationship between the PTT and
BP with a linear model [55–60]. Usually two calibration values were measured to
calibrate the model, rather than collecting a group of data to do the calibration.
Since the relationship between PTT and BP is far more complicated than the linear
model, the estimations with single linear model is not accurate.
The general case is the multiple variable linear regression where a model is cre-
ated for the relationship between multiple independent inputs variables X1, X2, …,
Xn = X and an output dependent variable BP:
BP = β 0 + β T X + ε . (9.7)

The model by nature is linear such that the output is a linear combination of the
input variables. With the awareness that only PTT cannot track BP well, there are
other indicators that can feature the BP changes, and there are studies using multiple
features, such as PTT, heart rate (HR), and other BP indicators to estimate BP [61,
62].
The advantages of the linear regression are that it is very efficient and useful
when the relationship to be modelled is not extremely complex and when there is
not enough data, and it is very simple to understand which variable is the most valu-
able for the estimate or prediction of the output. However, it would not be effective
at modelling highly complex non-linear relationship.
Regression Trees and Random Forests
Linear regression provides a global model of the variable or process to predict the
output, where only one predictive formula is holding over the entire data-space.
When the target output has more than one features which have complicated and
nonlinear interaction with each other, to develop a sole global model can be very
challenging. Even though it can be achieved successfully, it would be very diffi-
cult to understand. The alternative approach for nonlinear regression is to subdi-
vide, or partition the space into small regions. In the divided small regions, the
interactions are more manageable. Each of the small region is subdivided again
until finally the space that can be fitted with simple models. The global model thus
consists of two parts: the iterative partition, and a simple model for each element
of the partition [54].
Regression trees use the tree to represent the iterative partition. Each of the ter-
minal nodes (leaves) of the tree represents an element of the partition and has
attached to it a simple model which works for that element only. A point x belongs
to a leaf if x falls in the corresponding cell of the partition, and to figure out which
cell we are in, we start at the root node of the tree, and ask a sequence of questions
about the features. For classic regression trees, the model in each cell is just a con-
stant estimate of Y. That is, suppose the points (xi, yi), (xj, yj), …, (xn, yn) are the
samples belonging to the leaf node. Then our model for the leaf node is the sample
mean of the dependent variable in all the leaf nodes:
1 n
y= ∑ yi
n i =1
Random forest is a collection of decision trees, with the input vector running
through multiple decision trees. Ever since its introduction by Breiman in 2001
[63], it has become a very popular learning technique. The biggest advantage of the
regression trees is that it is great at learning complex, high non-linear relationship.
They usually can achieve high performance, better than simple linear and polyno-
mial regression and often comparable with neural networks [37, 42]. In addition, it
is very easy to interpret, with variables and variable values corresponding to nodes
can be pinpointed, thus to understand the importance of the feature variables and the
process. However, due the nature of training decision trees, they are very prone to
overfitting, but this can be overcome by using proper tree pruning and larger random
forest ensembles.
Support Vector Regression
Support vector regression (SVR) is a nonparametric technique, as it relies on the

kernel function. In ε-SVR, the set of training data includes both the predictor vari-
ables and observed response values. The goal is to find a function f(x) that deviates
from yn by a value no more than ε for each training point x, and to be as flat as pos-
sible (Fig. 9.3). The idea of SVR is closely related to that of support vector machine
152 X. Ding
Fig. 9.3 A schematic of support vector regression using ε-insensitive loss function [64]
(SVM) for classification. In SVM, a separating hyperplane is expected such that all
points are at a certain distance from this plane. If there are points that are too close
to the separating hyperplane, a penalty occurs. Similarly, in SVR, a function is
desirable such that all points are within a certain distance from this function. Again,
if points are outside this distance—the “ϵ-tube”, there will be a penalty.
Deep Learning Method for Cuffless BP Prediction
Deep learning is a subfield of machine learning. It uses deep networks with many
intermediate layers of artificial “neurons” between the input and the output, and,
like the visual cortex, these artificial neurons learn a hierarchy of progressively
more complex feature detectors. By learning feature detectors that are optimized for
a specific task such as classification, deep learning can substantially outperform
systems that rely on features supplied by domain experts or that are designed by
hand [34]. It is about using multiple levels of representation and abstraction that
help to make sense of data such as images, sound, and text.
Over the past few years, deep learning has been a popular technique for most
artificial intelligence type problems, overshadowing classical machine learning.
The underlying reason for this is that deep learning has repeatedly demonstrated its
superior performance on a wide variety of tasks including speech, natural language,
and playing games. The properties and advantages of deep learning including (1)
scales effectively with data: deep networks scale much better with more data than
classical ML algorithms, as shown in Fig. 9.4. That is why usually to improve the
accuracy with a deep network it is just better to use more data; (2) deep learning
excels at modelling extremely complicated relationships between inputs and out-
puts, as it can represent more complex features and to “learn” increasingly complex
models for predictions.
Neural networks which consists of more than three layers of neuros (including
the input and output layer) are called as the deep neural networks (DNN). And train-
ing them is called deep learning. Neural network is a machine learning algorithm
inspired from the working of human brain which enable a system to learn from
some observational data. A simple neural network consists of an input layer, a single
hidden layer and an output layer. Deep learning is a machine learning technique that
performs learning in more than two hidden layers (Fig. 9.5). It is a DNN consisting
multiple layers of nonlinear processing units (hidden layers). It performs feature
extraction and transformation. Each successive layer of DNN uses the output from
the previous layer as input. The deep learning method will perform better when
there is more data, more complicated models, and more computation.
Fig. 9.4 The performance of deep learning versus classic machine learning in terms of data size
Fig. 9.5 A deep neural network architecture for imagine recognition [65]
154 X. Ding
Currently, there is few studies in cuffless BP estimation with the real sense of
deep learning method. That is, to use the data-driven feature learning detectors to
extract abstract representations of the raw input and mapping the learned features to
the output. Instead, most of the current studies have still used the handcrafted fea-
tures as the input, and attempted the deep neural networks, such as deep neural
network, long short-term memory (LSTM) to model the relationship between the
features and the output.
Summary of State-of-the-Art, Application Examples
Review of Emerging Studies
In the recent years, most of the studies have been employed the hand-designed fea-
tures to indicate the BP changes, and attempted various machine learning algo-
rithms, including deep learning architecture to predict BP. One representative study
was conducted by Su et al., in which a deep recurrent neural network was proposed
to predict long-term BP of multiple components [50]. Figure 9.6 shows the over-
view of the proposal model.
The whole network was trained with backpropagation through time to miniatur-
ize the mean squared error (MSE) of the difference between BP prediction and the
ground truth of total N training samples:
Fig. 9.6 The Deep recurrent neural network that consists of the bottom layer bidirectional LSTM
(green dashed box) and the LSTM layer with residual connections (orange dashed box), where X1,
X2, …, Xt represent the extracted features, and y1, y2, …, yt the predicted BP time series [50]
1 N T
( )
 { x1:T , y1:T }N = ∑∑(zt − yt )2 + λθ 2
N i =1 t =1
(9.8)

where yt = [SBP, DBP, MBP] represents the ground truth, zt the corresponding pre-
diction, θ2 and λ are the L2 regulation of model parameters and the corresponding
coefficient, respectively. Validation of the proposal model on 84 health subjects with
7 features that are extracted from ECG and PPG signal showed that the 4-layer deep
RNN has achieved the best performance for SBP and DBP with RMSE of 3.73 and
2.43 mmHg, which is supervisor to the methods that are using classic machine learn-
ing algorithms, such as decision tree and SVR. Further, it can obtain accurate estima-
tion for multiple days BP, with one representative estimation illustrated in Fig. 9.7.
Wang et al. used artificial neural network with 22 extracted features and one hid-
den layer to estimate SBP and DBP, which achieved the MAE of 4.02 ± 2.79 mmHg
and 2.27 ± 1.82 mmHg for SBP and DBP, respectively [66]. Table 9.4 summarizes
the studies in the recent 3 years that are using machine learning method for the
development of BP models. It can be observed that the deep learning model with a
bigger training dataset generally can obtain a better accuracy than the classical
machine learning method [50, 52, 53, 67].
Discussions
With the advancement of machine learning techniques, more and more researchers
have attempted the machine learning method for cuffless BP measurement. However,
given the limited amount of data that are used for training of the model, the perfor-
mance of the model has not achieved to the level as expected. Taking the example
of the deep RNN, it is designed mostly to learn features from raw signals, but for the
task of BP estimation, the handcrafted features are still being used. In this case, the
use of the deep neural network would not take effect for modelling the relationship
between the input and the output.
The factors determining how well the machine learning algorithm will perform
depends on its ability to make the training error small, and to make the gap between
training and test error small. These two factors correspond to the two central chal-
lenges in machine learning: underfitting and overfitting. Underfitting occurs when
the model is not able to obtain a sufficiently low error values on the training set.
Overfitting occurs when the gap between the training error and the test error is too
large. This happens when the sample size for training is too small, which means the
training dataset does not adequately represent the whole population. As the result,
the trained model would only be accurate for the training set but not for the test
dataset. It is expected to achieve the best accuracy for the task of BP estimation with
the most appropriate machine learning method and with a big enough data set that
covers the information for the target population.
156
Fig. 9.7 Comparison of the Deep RNN prediction and the reference value of one representative subject for multi-day continuous BP estimation: (a–d) repre-
sent the results of first day, second day, fourth day and sixth month after the initial measurement, respectively [50]
X. Ding
Table 9.4 Summary of the recent studies using machine learning based method for cuffless BP
estimation
Machine
learning Training and
Reference Input features algorithm test Performance
Xing et al. Spectrum amplitude Artificial neural 69 subjects SBP: 0.06 ± 7.08 mmHg
[68] and phase of PPG network with DBP: 0.01 ± 4.66 mmHg
waveform one hidden
layer
Sun et al. PPG and ECG Multiple linear 19 subjects SBP: 0.43 ± 13.52 mmHg
[44] signals, PAT and 18 regression reference:
PPG features from Volume-
PPG and ECG signals clamp method
Leave-on-
subject-out
cross
validation
Jain et al. 32 parameters Sparse Training: 99 SBP: MAD: 4.43 mmHg
[43] extracted from ECG regression (to subjects (SD: 4.90 mmHg)
and PPG trim the Test: 10 DBP: MAD: 2.46 mmHg
redundant subjects (SD: 3.31 mmHg)
features) Reference:
OMRON
HBP1300
Duan et al. 11 out 56 features Support vector 57 subjects SBP: 4.77 ± 7.68 mmHg
[39] from PPG signal machine DBP: 3.67 ± 5.69 mmHg
regression MBP: 3.85 ± 5.87 mmHg
He et al. 18 features from ECG Random Forest One-hour SBP: 8.29 ± 5.84 mmHg
[42] and PPG signals continuous
BP: 1246
pairs DBP
1260 pairs
SBP
Shobitha 18 features extracted Relevance 26 subjects SBP: Kappa score = 0.99
et al. [40] from PPG signal vector machine DBP: Kappa score = 0.99
Miao et al. 14 features extracted Multiple linear 73 subjects SBP: −0.00 ± 3.10 mmHg
[47] from ECG and PPG regression DBP:
Support vector −0.00 ± 2.20 mmHg
regression
Lin et al. 19 PPG indicators Linear 22 subjects Combination of PPG and
[46] and PTT regression PTT achieves a better
method performance than
PTT-based method
Su et al. 7 features extracted Four-layer deep 84 healthy SBP: 3.73 mmHg
[50] from ECG and PPG RNN (LSTM) subjects (RMSE)
signals DBP: 2.43 mmHg
(RMSE)
(continued)
158 X. Ding
Machine
learning Training and
Reference Input features algorithm test Performance
Ertugrul ECG and PPG signals Extreme UCI datasets SBP: 6.93 mmHg (MAE)
et al. [45] learning MBP: 8.86 mmHg (MAE)
machine DBP: 19.43 mmHg
(MAE)
Radha Activity features A sequence-to- 120 subjects SBP: 5.65 mmHg
et al. [51] Heart rate variability sequence (RMSE)
PPG morphology model:
features perceptron +
LSTM
Wang Spectral and Artificial neural 72 subjects: SBP: 4.02 ± 2.79 mmHg
et al. [67] morphological network (one 70% training DBP: 2.27 ± 1.82 mmHg
features from PPG hidden layer) 15%
signal validation
15% testing
Polinski PTT, RR interval, and Single layer 21 subjects SBP: 1.06 mmHg (MAE)
et al. [69] respiration signal recurrent neural DBP: 0.63 mmHg (MAE)
network
Ghosh ECG and PPG LSTM 50 healthy SBP: 0.02 ± 4.8 mmHg
et al. [52] subjects DBP: 1.5 ± 3.7 mmHg
Wu et al. Waveform Eight hidden 85 subjects SBP: 3.63 mmHg (MAD)
[53] information, layer deep DBP: 2.45 mmHg (MAD)
handcrafted features neural
and personal features networks
from ECG and PPG
signals
Mousavi Whole base feature Adaptive 441 subjects SBP: −0.05 ± 8.90 mmHg
et al. [41] from PPG boosting MBP: 0.07 ± 4.91 mmHg
regression DBP: 0.19 ± 4.17 mmHg
The Future of the Technology
Challenges of the machine learning method for cuffless BP estimation are that the
training needs to cover all situations of all individuals to ensure the accuracy of the
model. The general limit and challenges of using machine learning are summarized
as below:
First, the successful training of the model relies on the enough accurate labelled
data. However, for BP measurement, even the reference BP values are sometimes
not 100% accurate. Moreover, there should have enough data for the training. That
is, to include all kinds of population, and for each subject the data should cover all
situations of BP changes, to obtain the most generalizable model for the application
of different types of population.
Second, machine learning, specifically the deep learning is kind of black box
technique. Clinicians, scientists, patients, and regulators would all prefer a simple
explanation for how a neural net arrives at its prediction of a specific case. However,
when a deep neural network is trained to make predictions on a big data set, it typi-
cally uses its layers of learned, nonlinear features to model a huge number of com-
plicated but weak regularities in the data. It is usually impossible to interpret these
features as their meaning depends on complex interactions with the learned uninter-
pretable features in other layers.
Correspondingly, there is opportunity and space for future studies to make a
breakthrough: (1) to construct standard database that can be used to validate various
algorithms. Current studies are using different datasets that are collected by differ-
ent designed clinical trials. Due to different subject characteristics and different
signal acquisition systems, there is no way to compare the performance of different
methods from different groups; (2) to use DNNs and learning algorithms, including
recurrent, generative adversarial, reinforcement, representation, and transfer [35,
70], to learn abstract features from the raw signals or learn the true data distribution
from the training set rather than with handcrafted features to capture the most rele-
vant information for the accurate prediction of BP; (3) to explore new signal modal-
ities that can be acquired with wearable/unobtrusive system.
References
1. Nichols W, O’Rourke M, Vlachopoulos C. McDonald’s blood flow in arteries: theoretical,

experimental and clinical principles. Boca Raton, FL: CRC; 2011.
2. Hughes DJ, Babbs CF, Geddes LA, Bourland JD. Measurements of Young’s modulus of elas-
ticity of the canine aorta with ultrasound. Ultrason Imaging. 1979;1(4):356–67.
3. Weltman G, Sullivan G, Bredon D. The continuous measurement of arterial pulse wave veloc-
ity. Med Biol Eng Comput. 1964;2(2):145–54.
4. Obrist PA, Light KC, McCubbin JA, Hutcheson JS, Hoffer JL. Pulse transit time: relationship
to blood pressure. Behav Res Methods Instrum. 1978;10(5):623–6.
5. Williams JGL, Williams B. Arterial pulse wave velocity as a psychophysiological measure, (in
English). Psychosom Med. 1965;27(5):408–14.
6. Gribbin B, Steptoe A, Sleight P. Pulse wave velocity as a measure of blood pressure change,
(in English). Psychophysiology. 1976;13(1):86–90.
7. Steptoe A, Smulyan H, Gribbin B. Pulse-wave velocity and blood-pressure change—calibra-
tion and applications, (in English). Psychophysiology. 1976;13(5):488–93.
8. Obrist PA, Light KC, McCubbin JA, Hutcheson J, Hoffer JL. Pulse transit time: relationship to
blood pressure and myocardial performance. Psychophysiology. 1979;16(3):292–301.
9. Allen RA, Schneider JA, Davidson DM, Winchester MA, Taylor CB. The covariation of
blood pressure and pulse transit time in hypertensive patients, (in English). Psychophysiology.
1981;18(3):301–6.
10. Geddes LA, Voelz MH, Babbs CF, Bourland JD, Tacker WA. Pulse transit time as an indicator
of arterial blood pressure,(in English). Psychophysiology. 1981;18(1):71–4.
11. Newlin DB. Relationships of pulse transmission times to pre-ejection period and blood pres-
sure. Psychophysiology. 1981;18(3):316–21.
12. Lane JD, Greenstadt L, Shapiro D, Rubinstein E. Pulse transit time and blood pressure—an
intensive analysis, (in English). Psychophysiology. 1983;20(1):45–9.
13. Pollak MH, Obrist PA. Aortic radial pulse transit time and ECG Q-wave to radial pulse wave
interval as indexes of beat-by-beat blood pressure change, (in English). Psychophysiology.
1983;20(1):21–8.
160 X. Ding
14. Marie GV, Lo CR, Vanjones J, Johnston DW. The relationship between arterial blood pressure
and pulse transit time during dynamic and static exercise, (in English). Psychophysiology.
1984;21(5):521–7.
15. Sawada Y, Yamakoshi K. A correlation analysis between pulse transit time and instantaneous
blood pressure measured indirectly by the vascular unloading method, (in English). Biol
Psychol. 1985;21(1):1–9.
16. Zong W, Moody G, Mark R. Effects of vasoactive drugs on the relationship between ECG-
pulse wave delay time and arterial blood pressure in ICU patients. In: Computers in Cardiology,
IEEE; 1998. p. 673–6.
17. Nitzan M, Khanokh B, Slovik Y. The difference in pulse transit time to the toe and finger mea-
sured by photoplethysmography (in English). Physiol Meas. 2002;23(1):85–93.
18. Ahlstrom C, Johansson A, Uhlin F, Länne T, Ask P. Noninvasive investigation of blood pres-
sure changes using the pulse wave transit time: a novel approach in the monitoring of hemodi-
alysis patients. J Artif Organs. 2005;8(3):192–7.
19. Muehlsteff J, Aubert X, Schuett M. Cuffless estimation of systolic blood pressure for short
effort bicycle tests: the prominent role of the pre-ejection period. In: Conf Proc IEEE Eng Med
Biol Soc, New York, USA, vol. 1, 2006. p. 5088–92.
20. Payne RA, Symeonides CN, Webb DJ, Maxwell SRJ. Pulse transit time measured from the
ECG: an unreliable marker of beat-to-beat blood pressure, (in English). J Appl Physiol.
2006;100(1):136–41.
21. Marcinkevics Z, Greve M, Aivars JI, Erts R, Zehtabi AH. Relationship between arterial pres-
sure and pulse wave velocity using photoplethysmography during the post-exercise recovery
period. Acta Univ Latviensis Biol. 2009;753:59–68.
22. Wong MYM, Poon CCY, Zhang Y-T. An evaluation of the cuffless blood pressure estimation
based on pulse transit time technique: a half year study on normotensive subjects, (in English).
Cardiovasc Eng. 2009;9(1):32–8.
23. Zheng YL, et al. Unobtrusive sensing and wearable devices for health informatics. IEEE Trans
Biomed Eng. 2014;61(5):1538–54.
24. Young CC, Mark JB, White W, DeBree A, Vender JS, Fleming A. Clinical evaluation of contin-
uous noninvasive blood pressure monitoring: accuracy and tracking capabilities. J Clin Monit.
1995;11(4):245–52.
25. Chen W, Kobayashi T, Ichikawa S, Takeuchi Y, Togawa T. Continuous estimation of systolic
blood pressure using the pulse arrival time and intermittent calibration (in English). Med Biol
Eng Comput. 2000;38(5):569–74.
26. Fung P, Dumont G, Ries C, Mott C, Ansermino M. Continuous noninvasive blood pressure
measurement by pulse transit time. In: Conf Proc IEEE Eng Med Biol Soc, 2004, vol. 1, IEEE,
2004. p. 738–41.
27. Poon CCY, Zhang YT. Cuff-less and noninvasive measurements of arterial blood pressure by
pulse transit time. In: Conf Proc IEEE Eng Med Biol Soc, Shanghai, China. IEEE; 2005.
p. 5877–5880.
28. Cattivelli FS, Garudadri H. Noninvasive cuffless estimation of blood pressure from pulse
arrival time and heart rate with adaptive calibration. In: Sixth international workshop on wear-
able and implantable body sensor networks (BSN 2009): IEEE; 2009. p. 114–9.
29. Gesche H, Grosskurth D, Küchler G, Patzak A. Continuous blood pressure measurement
2012;112(1):309–15.
30. Chen Y, Wen C, Tao G, Bi M. Continuous and noninvasive measurement of systolic and dia-
stolic blood pressure by one mathematical model with the same model parameters and two
separate pulse wave velocities. Ann Biomed Eng. 2012;40(4):871–82.
31. Ding XR, Zhang YT, Liu J, Dai WX, Tsang HK. Continuous cuffless blood pressure estimation
using pulse transit time and photoplethysmogram intensity ratio. IEEE Trans Biomed Eng.
2016;63(5):964–72.
32. Huynh TH, Jafari R, Chung WY. Noninvasive cuffless blood pressure estimation using pulse
transit time and impedance plethysmography. IEEE Trans Biomed Eng. 2018.
33. Liu J, Yan B, Zhang Y, Ding XR, Peng S, Zhao N. Multi-wavelength photoplethysmography
enabling continuous blood pressure measurement with compact wearable electronics. IEEE
Trans Biomed Eng. 2018.
34. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436.
35. Goodfellow I, Bengio Y, Courville A, Bengio Y. Deep learning. Cambridge: MIT Press; 2016.
36. Kim JY, Cho BH, Im SM, Jeon MJ, Kim IY, Kim SI. Comparative study on artificial neural
network with multiple regressions for continuous estimation of blood pressure. In: Engineering
in Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th Annual International
Conference of the IEEE; 2005. p. 6942–5.
37. Monte-Moreno E. Non-invasive estimate of blood glucose and blood pressure from a photople-
thysmograph by means of machine learning techniques. Artif Intell Med. 2011;53(2):127–38.
38. Ruiz-Rodríguez JC, et al. Innovative continuous non-invasive cuffless blood pressure monitor-
ing based on photoplethysmography technology. Intensive Care Med. 2013;39(9):1618–25.
39. Duan K, Qian Z, Atef M, Wang G. A feature exploration methodology for learning based cuf-
fless blood pressure measurement using photoplethysmography. In: Engineering in medicine
and biology society (EMBC), 2016 IEEE 38th annual international conference of the IEEE;
2016. p. 6385–8.
40. Shobitha S, Amita P, Krupa BN, Beng GK. Cuffless blood pressure prediction from PPG
using relevance vector machine. In: Electrical, Electronics, Communication, Computer, and
Optimization Techniques (ICEECCOT), 2017 International Conference on. IEEE; 2017.
p. 75–8.
41. Mousavi SS, Firouzmand M, Charmi M, Hemmati M, Moghadam M, Ghorbani Y. Blood pres-
sure estimation from appropriate and inappropriate PPG signals using a whole-based method.
Biomed Signal Process Control. 2019;47:196–206.
42. He R, Huang Z-P, Ji L-Y, Wu J-K, Li H, Zhang Z-Q. Beat-to-beat ambulatory blood pres-
sure estimation based on random forest. In: Wearable and implantable body sensor networks
(BSN), 2016 IEEE 13th international conference on: IEEE; 2016. p. 194–8.
43. Jain M, Kumar N, Deb S, Majumdar A. A sparse regression based approach for cuff-less blood
pressure measurement. In: Acoustics, speech and signal processing (ICASSP), 2016 IEEE
international conference on: IEEE; 2016. p. 789–93.
44. Sun S, Bezemer R, Long X, Muehlsteff J, Aarts R. Systolic blood pressure estimation using
PPG and ECG during physical exercise. Physiol Meas. 2016;37(12):2154.
45. Ertuğrul ÖF, Sezgin N. A non-invasive time-frequency based approach to estimate cuffless
arterial blood pressure 2. Methods. 2018;2:18.
46. Lin W-H, Wang H, Samuel OW, Liu G, Huang Z, Li G. New photoplethysmogram indica-
tors for improving cuffless and continuous blood pressure estimation accuracy. Physiol Meas.
2018;39(2):025005.
47. Miao F, et al. A novel continuous blood pressure estimation approach based on data mining
techniques. IEEE J Biomed Health Inform. 2017;21:1730–40.
48. Bose SSN, Kandaswamy S. Sparse representation of photoplethysmogram using K-SVD for
cuffless estimation of arterial blood pressure. In: Advanced computing and communication
systems (ICACCS), 2017 4th international conference on: IEEE; 2017. p. 1–5.
49. Pan J, Zhang Y. Improved Blood Pressure Estimation Using Photoplethysmography Based on
Ensemble Method. In: Pervasive Systems, Algorithms and Networks & 2017 11th International
Conference on Frontier of Computer Science and Technology & 2017 Third International
Symposium of Creative Computing (ISPAN-FCST-ISCC), 2017 14th International Symposium
on. IEEE; 2017. p. 105–11.
50. Su P, Ding X, Zhang Y, Li Y, Zhao N. Predicting blood pressure with deep bidirectional LSTM
network, arXiv preprint. In: 2018 IEEE EMBS International Conference on Biomedical &
Health Informatics (BHI), Las Vegas, Nevada, USA, 2018. p. 323–8.
51. Radha M, et al. Wrist-worn blood pressure tracking in healthy free-living individuals using
neural networks. arXiv preprint arXiv:1805.09121, 2018.
52. Ghosh S, Banerjee A, Ray N, Wood PW, Boulanger P, Padwal R. Using accelerometric and
gyroscopic data to improve blood pressure prediction from pulse transit time using recur-
rent neural network. In: 2018 IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP). IEEE; 2018. p. 935–9.
162 X. Ding
53. Wu D, Xu L, Zhang R, Zhang H, Ren L, Zhang Y-T. Continuous cuff-less blood pressure esti-
mation based on combined information using deep learning approach. J Med Imaging Health
Inform. 2018;8(6):1290–9.
54. Regression Trees. http://www.stat.cmu.edu/~cshalizi/350-2006/lecture-10.pdf.
55. Chan K, Hung K, Zhang Y. Noninvasive and cuffless measurements of blood pressure for
telemedicine. In: Engineering in Medicine and Biology Society, 2001. Proceedings of the 23rd
Annual International Conference of the IEEE, vol. 4. IEEE; 2001. p. 3592–3.
56. Payne R, Symeonides C, Webb D, Maxwell S. Pulse transit time measured from the ECG: an
unreliable marker of beat-to-beat blood pressure. J Appl Physiol. 2006;100(1):136–41.
57. Douniama C, Sauter C, Couronne R. Blood pressure tracking capabilities of pulse transit times
in different arterial segments: a clinical evaluation. In: Computers in Cardiology, 2009. IEEE;
2009. p. 201–4.
58. Wong MY-M, Poon CC-Y, Zhang Y-T. An evaluation of the cuffless blood pressure estimation
based on pulse transit time technique: a half year study on normotensive subjects. Cardiovasc
Eng. 2009;9(1):32–8.
59. Choi Y, Zhang Q, Ko S. Noninvasive cuffless blood pressure estimation using pulse transit
time and Hilbert–Huang transform. Comput Electr Eng. 2013;39(1):103–11.
60. Heravi MY, Khalilzadeh M, Joharinia S. Continuous and cuffless blood pressure monitoring
based on ECG and SpO2 signals ByUsing Microsoft visual C sharp. J Biomed Phys Eng.
2014;4(1):27.
61. Kim JS, Kim KK, Baek HJ, Park KS. Effect of confounding factors on blood pressure estima-
tion using pulse arrival time. Physiol Meas. 2008;29(5):615.
62. Ye S-y, Kim G-R, Jung D-K, Baik S, Jeon G. Estimation of systolic and diastolic pressure
using the pulse transit time. World Acad Sci Eng Technol. 2010;67:726–31.
63. Breiman L. Random forests. Mach Learn. 2001;45(1):5–32.
64. Deka PC. Support vector machine applications in the field of hydrology: a review. Appl Soft
Comput. 2014;19:372–86.
65. Are deep neural nets “Software 2.0”?. 2018.
66. Wang G, Atef M, Lian Y. Towards a continuous non-invasive cuffless blood pressure monitor-
ing system using PPG: systems and circuits review. IEEE Circuits Syst Mag. 2018;18(3):6–26.
67. Wang L, Zhou W, Xing Y, Zhou X. A novel neural network model for blood pressure estima-
tion using photoplethesmography without electrocardiogram. J Healthc Eng. 2018;2018
68. Xing X, Sun M. Optical blood pressure estimation with photoplethysmography and FFT-based
neural networks. Biomed Opt Express. 2016;7(8):3007–20.
69. Poliñski A, Czuszyñski K, Kocejko T. Blood pressure estimation based on blood flow, ECG
and respiratory signals using recurrent neural networks. In: 2018 11th International Conference
on Human System Interaction (HSI): IEEE; 2018. p. 86–92.
70. Yu K-H, Beam AL, Kohane IS. Artificial intelligence in healthcare. Nat Biomed Eng.

2018;2(10):719.
Chapter 10
Initialization of Pulse Transit Time-Based
Ramakrishna Mukkamala and Jin-Oh Hahn
Abstract Pulse transit time (PTT)-based monitors can potentially permit cuffless
blood pressure (BP) measurements, but initialization of the monitors so that the
measured time delay may be mathematically converted to absolute BP is difficult.
The goal of this chapter is to facilitate the achievement of reliable and practical
initialization of PTT-based BP monitors. We present parametric models to relate
PTT to BP; explain three different methods to determine the model parameters for
a patient via cuff BP measurements that range from more accurate to more conve-
nient; describe models that incorporate readily available covariates in addition to
PTT and parameter determination methods for these enhanced models; discuss the
required time period for reinitialization of the monitors with cuff BP measurements;
and make recommendations and provide our outlook on such monitors.
Keywords Aging · Arterial stiffness · Blood pressure monitoring · Calibration ·

Cuff-less blood pressure · Hypertension · Initialization · Mobile health · Pulse
transit time · Pulse wave velocity
Introduction
Blood pressure (BP) monitors based on pulse transit time (PTT, the time delay for
the pressure wave to travel between two arterial sites) are being widely pursued [1].
One reason is that PTT is inversely correlated with BP in a person due to fundamen-
tal properties of pulse wave propagation in nonlinear arteries. Hence, PTT may offer
a physics-based principle for BP measurement. The other reason is that PTT can be
simply measured as the relative timing between proximal and distal arterial
R. Mukkamala (*)
Department of Electrical and Computer Engineering, Michigan State University,
East Lansing, MI, USA
e-mail: rama@egr.msu.edu
J.-O. Hahn
Department of Mechanical Engineering, University of Maryland, College Park, MD, USA
e-mail: jhahn12@umd.edu

164 R. Mukkamala and J.-O. Hahn
waveforms. Hence, PTT does not involve variable force application for its measure-
ment and may thereby permit passive, cuff-less BP monitoring.
However, initialization of PTT-based BP monitors so that the PTT measurements
in units of ms may be mapped to BP measurements in units of mmHg is a major
challenge. To explain, BP (P) and PTT (τ) over a measured length (l) may be
approximately related via a simple model as follows:
l
P = K1 + K 2 . (10.1)
t
The slope K1 (>0) and intercept K2 (<0) vary from person to person, change over
time within a person (e.g., K1 decreases with aging and disease), and are unknown
for a given person. The problem is to determine the two (or more) “calibration
curve” parameters for a person so that PTT measurements from that person can be
converted to absolute BP. This problem is particularly difficult, because there is
more than one unknown parameter.
Initialization of PTT-based BP monitors thus amounts to construction of a cali-
bration curve. The initialization procedure generally involves defining a parametric
model to relate PTT to BP and determining the model parameters via simultaneous
measurements of cuff BP and proximal and distal arterial waveforms for detecting
PTT. BP may then be obtained from a person without using a cuff by applying PTT
measurements to the resulting calibration curve seeded into the monitor. The initial-
ization procedure should be repeated periodically for the person to account for
changes in the calibration curve due to cardiovascular aging. Additional covariates
of BP extracted from the measured waveforms may also be incorporated to improve
the accuracy in measuring both systolic and diastolic BP.
The goal of this chapter is to facilitate the achievement of reliable and practical
initialization of PTT-based BP monitors. Figure 10.1 illustrates the different ways to
initialize the monitors and provides an overview of the chapter. We begin by describ-
ing and comparing various parametric models for relating PTT to BP. We then
explain three methods to determine the model parameters that range from more
accurate, person-specific initialization to more convenient, population-based initial-
ization. We next present models that incorporate covariates such as waveform
amplitudes in addition to PTT and parameter determination methods for these
enhanced models. We thereafter discuss the required time period for reinitialization
based on theoretical predictions and experimental studies. We conclude by making
recommendations on initializing PTT-based BP monitors and providing our outlook
on such monitors.
Parametric Models to Relate PTT to BP
Theoretical and empirical models have been proposed to relate PTT to BP. Some of
these models may afford a more practical calibration curve form than others.
Fig. 10.1 Initialization of pulse transit time (PTT)-based blood pressure (BP) monitors involves defining a parametric model to relate PTT (τ) over a measured
length (l) and possibly other covariates such as waveform amplitudes to BP (P) and determining the models parameters (e.g., Kj) by employing one of three
10 Initialization of Pulse Transit Time-Based Blood Pressure Monitors
methods (person-specific, hybrid, population-based) to cuff BP measurements (initialization phase). BP may then be measured by invoking the so-obtained
calibration curve to measurements of PTT and possibly other covariates (cuff-less monitoring phase). The initialization phase should be repeated periodically
over time in a person to account for the effect of aging and disease progression on the calibration curve
165
Theoretical Models
The inverse relationship between PTT and BP is well understood [1] and based on
two physiologic mechanisms. The first mechanism is that PTT decreases with
increasing arterial elasticity (i.e., arterial stiffening) due to fluid dynamic principles.
The second mechanism is that arterial elasticity increases with BP due to nonlinear
material properties of the arterial wall. The theoretical models account for both of
these mechanisms.
The first mechanism may be quantified by assuming that an arterial vessel can be
represented as a cylindrical tube wherein the tube wall is composed of homoge-
neous, elastic material; blood is incompressible; flow is inviscid and in the axial
direction with uniform velocity profile; and the tube cross-sectional area (A) changes
little. These assumptions are most tenable for central arteries [1]. The governing
equations for this simple model can be obtained by invoking conservation of mass
and momentum and are equivalent to an ideal, lossless transmission line. It is well
known that solutions to these equations are waves that propagate with pulse wave
velocity (PWV, v) as follows:
l A dP
v= = , (10.2)
t r dA
dA
where ρ is the blood density, which is near that of water, and is the tube compli-
dP
ance (C). By applying Laplace’s law for a thin-walled tube and further assuming
that the tube wall has constant elastic modulus (E), C may be derived as follows:
dA 2p r 3
C= = , (10.3)
dP Eh
where r and h are the tube radius and wall thickness. Substituting Eq. (10.3) into Eq.
(10.2) yields another relationship for PWV as follows:
l Eh
v= = . (10.4)
t 2r r
Equation (10.4) is the popular Moens–Korteweg equation, while Eq. (10.2) is

another well-known equation that is often called the Bramwell–Hill equation in this
field. Both formulas indicate that PTT varies inversely with arterial elasticity. Note
that, even though compliance was assumed constant here, Eq. (10.2) is valid
dA
even when compliance depends on BP (i.e., = C ( P )) [1].
dP
The second mechanism may be quantified via curve fitting of simultaneous mea-
surements of an arterial elasticity metric and BP over a range of BP. A popular
formula is the Hughes’ model for the elastic modulus of the in vivo canine aorta as
follows:
10 Initialization of Pulse Transit Time-Based Blood Pressure Monitors 167
E = E0 ea P , (10.5)
where E0 is the elastic modulus at zero BP and α (>0) indicates the degree of the
elastic modulus dependency on BP [2]. Another useful formula is the Wesseling
model for the compliance of the ex vivo human aorta as follows:
é1 1 æ P - P0 ö ù
A ( P ) = Amax ê + tan -1 ç ÷ú (10.6)
êë 2 p è P1 ø úû
dA Amax
C (P) = = , (10.7)
dP é æ P - P ö2 ù
p P1 ê1 + ç 0
÷ ú
êë è P1 ø úû
where Amax is the maximum cross-sectional area at infinite BP; P0 is the BP at which
the compliance is maximal; and P1 reflects the BP range for which the compliance
is relatively constant [3]. The Wesseling model also includes a relationship between
its three parameters and age/gender as follows:
ì4.12 females
Amax = í (10.8)
î5.62 males
ì72 - 0.89age females

P0 = í (10.9)
î76 - 0.89age males
P1 = 57 - 0.44age, (10.10)
where Amax, Pj, and age are in units of cm2, mmHg, and years. Because of differences
between cardiovascular and chronological aging [4], these formulas may be thought
of as representing an average person for each age and gender. As indicated in
Fig. 10.2a, the overall Wesseling model indicates that the arterial cross-sectional
area becomes less dependent on BP with aging, while the compliance becomes
more dependent on BP with aging. Equations (10.8)–(10.10) also indicate that the
compliance is larger in males. The second mechanism may also be quantified via
mathematical modeling. Fung’s hyper-elastic model of the arterial wall and a cylin-
drical tube geometry without assumption of a thin wall was recently employed to
derive Ma’s formulas for A(P) and C(P) [5]. These model-based formulas have both
similarities and differences with the experimentally based Eqs. (10.6) and (10.7).
Due to the different quantitative formulations of the two underlying mechanisms,
multiple, theoretical models arise to relate PTT to BP. One popular model, which
may be obtained by substituting Eq. (10.5) into Eq. (10.4) [6], is a logarithmic rela-
tionship as follows:
æt ö
P = K1 ln ç ÷ + K 2 , (10.11)
èlø
(A) (B)
5.5 0.06 175
5
0.05 150
4.5
0.04 125
C [cm /mmHg]
4
P [mmHg]
A [cm 2]
3.5 0.03 100
2
3
0.02 75
2.5
30 years old
70 years old 0.01 50
2
1.5 0 25
25 50 75 100 125 150 175 25 50 75 100 125 150 175 0 5 10 15 20
P [mmHg] P [mmHg] l/ [m/s]
Fig. 10.2 A parametric model relating PTT to BP may be defined with the Wesseling model of the
aortic cross-sectional area (A)-BP relationship [3] integrated into the Bramwell–Hill equation (see
æ dA ö
Eq. 10.2). (a) The model A − P and C ç = ÷ - P relationships for a young and old male (see
è dP ø l
Eqs. 10.6-10.10). (b) The corresponding model P - relationships defined by two parameters
t
(see Eq. 10.14). The model indicates that the shape of the PTT-BP relationship changes with age
2 1 æ 2r r ö
where K1 = - and K 2 = ln ç ÷ and are thus person- and time-specific param-
a a è E0 h ø
eters. Another popular model, which may arise by substituting Eqs. (10.6) and
(10.7) into Eq. (10.2) and then assuming large P (i.e., (P>>P0 + P1) [1], is a line
relationship in PWV as follows:
l
P = K1 + K2 , (10.12)
t
2 r P1
where K1 = and K2 = P0 and are thus person- and time-specific parame-
p +2
ters. Ma’s A(P) and C(P) formulas were substituted into Eq. (10.2), and a reduced
model was derived as a quadratic relationship in PWV [5] as follows:
2
ælö
P = K1 ç ÷ + K 2 , (10.13)
èt ø
where K1 and K2 are functions of the material and geometric properties of the artery
and are thus person- and time-specific parameters. Equation (10.13) may also arise
by linearizing Eq. (10.5) and then substituting the linear model into Eq. (10.4) [7].
A more general model, which arises by substituting Eqs. (10.6) and (10.7) into Eq.
(10.2) without any assumptions on P [8], is as follows:
t 2819.7
= , (10.14)
l æ æ P - P ö2 ö æ 1 1 æ P - P0 öö
p P1 ç 1 + ç 0
÷ ÷ ç + tan -1 ç
ç ÷ ÷÷
ç è P1 ø ÷ è 2 p è P1 øø
è ø
where τ is in units of ms, l is in units of m, and P0 and P1 are person- and time-
specific parameters. Using the age-dependent P0 and P1 formulas in Eqs. (10.9) and
(10.10) and as illustrated in Fig. 10.2b, the model indicates that the shape of the
PTT-BP relationship may be age-dependent and becomes nearly a line relationship
in PWV in the elderly. (Note that at very low BP of a younger person, the PTT-BP
relationship may not be one-to-one due to the dominance of elastin, which is a linear
elastic fiber, over collagen in these conditions [1].) While all of the models here are
theoretically based, none may be strictly correct. In particular, Eq. (10.4) was
derived assuming constant E, whereas Eq. (10.2) was derived assuming small
changes in A.
Empirical Models
Empirical models are mainly based on PTT-BP data fitting considerations. A popu-
lar model is a line relating PTT to BP as follows:
t
P = K1 + K2 , (10.15)
l
where K1 and K2 are person- and time-specific parameters. Nonlinear, yet parsimo-
nious, functions have also been used to relate PTT to BP. Examples of such models
are as follows:
P = K1 x 2 + K 2 x + K 3 (10.16)
P = K1 x p + K 2 (10.17)
P = K1e K2 x , (10.18)
t l
where x = or , Kj are person- and time-specific parameters, and p is either a fixed
l t
value or likewise varies with the person and time [9, 10]. A more unique model
relates PTT to pulse pressure (PP = systolic BP – diastolic BP) rather than an abso-
lute level of BP as follows:
2
ælö
PP = K ç ÷ , (10.19)
èt ø
where K is a person- and time-specific parameter [11]. Empirical models have also
been proposed based on initialization considerations. One such model was specifi-
cally designed to prevent non-physiologic BP (i.e., negative or too high BP) and is
as follows:
K1
P= 2
+ K3 , (10.20)
æt ö
ç l - K2 ÷
è ø
where Kj are person- and time-specific parameters [12]. While this design may be
practical for initialization, note that physiologically correct limiting behavior is for
PTT to be finite at zero BP and approach zero as BP approaches infinity. See [1] for
a list of publications on the aforementioned and other PTT-BP models.
Comparison of Models as a Practical Calibration Curve Form
A model relating PTT to BP may constitute a practical calibration curve form, if it

fits actual PTT-BP data over a wide BP range using a minimal number of parameters
and does not output unrealistic BP. Several studies have compared PTT-BP models
in terms of fitting simultaneous measurements of PTT and BP over a range of
BP. These studies did not include single-parameter models, because such models do
not generally fit PTT-BP data (see, e.g., plots in [10, 13, 14]). Also note that Eq.
(10.19), a one-parameter model relating PWV squared to PP, may not hold in gen-
eral, as K depends on the difference in the arterial cross-sectional areas at systole
and diastole (see Eq. (10.34) below) and may thus vary considerably within a person
[15–17]. The comparative studies in humans have generally concluded that various
models fit the data similarly and increasing the number of model parameters hardly
improves the data fitting. The reason is that the data were obtained over too limited
a BP range due to ethical considerations. However, comparative studies in animals
wherein BP may be perturbed widely do indicate that Eq. (10.12), a line model in
PWV, fits PTT-BP data significantly better than Eq. (10.15), a line model in PTT
[18, 19]. A recent animal study indicated that Eq. (10.18), an exponential model with
t
x = , fits PTT-BP data well over a 100 mmHg range of BP and best amongst vari-
l
ous two-parameter models [10]. While two parameters appear to be the minimum
number needed for a model to fit data, models with more parameters such as Eq.
(10.20) can prevent non-physiologic BP. However, instead of adding more model
parameters, unrealistic BP could be averted by invoking the known BP range (see,
e.g., vast BP statistical data for different genders, ethnicities, and age groups [20]).
So, if a two-parameter model yielded BP outside this range, then the PTT-based
monitor could either not output BP or output the maximum BP for an unusually low
PTT or the minimum BP for an unusually high PTT. The former option makes more
sense when the unrealistic BP is due to gross error in PTT, whereas the latter option
might be reasonable when the unrealistic BP is due to a physiologic change in the

model parameters. Since the cause of the unrealistic BP would be unknown, it may
be safest to not output BP. If no BP outputs persist, then reinitialization of the moni-
tor may be indicated.
Model Parameter Determination
There are three methods to determine the multiple parameters of a PTT-BP model
and thereby initialize a PTT-based BP monitor for a person (see Fig. 10.1). A
person-specific method involves measuring cuff BP and PTT during interventions
that perturb BP in the person to determine all model parameters. A population-based
method involves using basic information about the person along with a training
dataset comprising cuff BP and PTT measurements from a cohort of different sub-
jects to determine all model parameters. A hybrid method involves measuring cuff
BP and PTT in the person to determine one model parameter and using the person’s
basic information and a similar training dataset to determine the remaining param-
eters. The three methods represent different trade-offs between accuracy and
convenience.
Person-Specific Methods
Figure 10.3 illustrates the person-specific method for determining the PTT-BP
model parameters. This method generally involves: (1) employing one or more
interventions to perturb BP in the person; (2) measuring cuff BP and PTT during the
baseline period and each intervention; and (3) fitting the model to the multiple
PTT-BP data pairs to determine all parameters. By yielding a calibration curve that
is specific to the person at the time of measurement, the method is most accurate but
least convenient.
Table 10.1 summarizes interventions that have been employed to perturb BP in
order to initialize and/or assess a PTT-based monitor. These interventions generally
represent different trade-offs between effect and convenience. Slow breathing [21]
and postural changes (e.g., from supine to upright [22]) are relatively convenient but
change BP little. Cold pressor (i.e., immersing a limb in ice water) [23] and different
types of exercise (e.g., cycling [24], sustained handgrip [25], and mental arithmetic
[26]) increase BP significantly but are less convenient. On the other hand, cycling
and similar exercises could be incorporated in a person’s daily life, and BP may also
fall below pre-exercise level sometime after the exercise [27]. The Valsalva maneu-
ver (i.e., exhaling with nose and mouth closed) [28] is fairly unique in that it
decreases BP appreciably but is inconvenient in that a non-standard, continuous BP
monitor such as a finger cuff, volume-clamp device [29, 30] is needed to detect the
momentary BP reduction.
One intervention that is both significant and relatively convenient is a hydrostatic

maneuver in which the vertical height (h) of the effective BP measurement site of a
PTT-based monitor with respect to the heart is varied. As shown in Fig. 10.4, due to
the weight of the blood column, such a maneuver will cause the BP at the measure-
ment site (i.e., “local BP”) to change by ρgh, where g is gravity. A change in h of
Fig. 10.3 The person-specific method for determining the PTT-BP model parameters involves
measuring cuff BP and PTT in the person during interventions that perturb BP and then optimally
fitting the model to the measured BP-PTT pairs to determine all parameters
Table 10.1 Blood pressure (BP) interventions

Intervention BP effect [mmHg] Convenience level
Slow breathing [21] <|5|/<|5|a High
Supine to standing [22] <|10|/+4 High
Cold pressor [23] +16/+14a Low
Exercise (e.g., cycling) +40/+40 to −9/−4a (1-h Moderate (could be incorporated in
[24, 27] later) daily life)
Sustained handgrip [25] +45−50/+40 Low
Mental arithmetic [26] +20/+11 Low (requires person adherence)
Valsalva maneuver [28] −15/−15 Low (requires special cuff to detect fast
change)
Hydrostatic maneuver: −50/−50 to +50/+50 High (but requires heart level detection)
Hand lowering/raising [31] +30/+30
Supine to seated [14]
All BP effect values for normal subjects
a
Indicates that BP effect is significantly greater in hypertensive subjects
Fig. 10.4 A relatively convenient intervention with large BP effect is a hydrostatic maneuver. This
intervention involves varying the vertical height (h) of the effective BP measurement site of a PTT-
based monitor with respect to the heart. Due to the weight of the blood column, such a maneuver
will cause the BP at the measurement site (PLocal) to change by ρgh, where ρ and g are respectively
the known blood density and gravity, relative to the BP at heart level (PHeart)
just 10 cm, will cause BP to change by over 7 mmHg. Implementation of a hydro-

static maneuver depends on how the monitor detects PTT. PTT detected from lower
arm and hand arterial waveforms is relatively convenient and most suitable to the
intervention. The effective BP measurement site would be at or near the hand, so the
hydrostatic maneuver could be implemented simply by hand lowering and raising
[31, 32]. For an average arm length, hand BP would change by about ±50 mmHg
relative to BP at heart level. PTT detected from ECG and finger arterial waveforms
(i.e., “pulse arrival time (PAT)”) [1] may be more convenient. However, the effective
BP measurement site of the PAT would be at about the midpoint of the arm such that
hand raising and lowering would change the local BP by only about half the extent.
PTT detected from carotid and femoral arterial waveforms (i.e., aortic PTT) may
best correlate with BP [1]. The effective BP measurement site would be at about the
abdomen, so the hydrostatic maneuver could be implemented by shifting posture
from supine to upright. In this case, while arm cuff BP (i.e., systemic BP) increases
by only a few mmHg, the local BP increase due to ρgh is almost 30 mmHg on aver-
age [14]. However, measurement of aortic PTT is inconvenient.
Another possible intervention is to leverage the natural BP variations that occur
over time in a person due to stress, physical activity, meals, and other factors. The
standard deviations of such diastolic BP and systolic BP variations are about 5 and
6–9 mmHg, respectively [33]. The systolic BP variations are higher in the elderly
and African Americans. Since this intervention could require many PTT-BP mea-
surements, it may be best suited to those patients who are already undergoing home
or 24-h ambulatory cuff BP monitoring as part of clinical care.
It is important to note that these interventions change BP via different physio-
logic mechanisms. For example, cold pressor may increase BP via augmented sys-
temic vascular resistance, whereas exercise may increase BP primarily via enhanced
cardiac output. A PTT-based BP monitor initialized using one intervention may thus
be less effective when tested using another intervention. Employing a set of inter-
ventions that invokes different physiologic mechanisms may provide a good bal-
ance for more effective initialization overall.
At the minimum, BP and PTT should be measured before the interventions and
during the steady-state of each intervention. Acquiring the BP-PTT data pairs dur-
ing the transient portion of an intervention may also be beneficial. A finger cuff,
volume-clamp device would allow full capture of transient changes. However, such
devices are expensive and not readily available. Use of an automatic arm cuff for the
BP measurement would be most convenient.
The BP value used to determine the model parameters from the measured data
depends on how the monitor detects PTT. For PTT detected at the proximal and
distal arterial waveform troughs or feet, which is at the level of diastole and com-
monly done to mitigate the impact of arterial wave reflection [1], it makes most
sense to use diastolic BP. For PAT, which includes the pre-ejection period (PEP) in
addition to PTT, it may be best to use systolic BP, which is the sum of diastolic BP
and PP. The reason is that the ventricle and arteries are common and strong determi-
nants of PEP and PP. Experimental studies have shown that PTT correlates better
with diastolic BP [18, 19] and that PAT correlates better with systolic BP [1].
However, the model parameters have often been determined to relate one time delay
to each of systolic BP and diastolic BP [1]. This common practice may be justified
by a popular, line model relating the two BP values in a person, which typically
yields correlation coefficients of 0.7 to 0.8 [34, 35]. The phenomenological model
has slope and intercept that are age- and gender-dependent as follows:
Ps = G1 Pd + G2 (10.21)
ì0.0067age + 0.90 females

G1 = í (10.22)
î 0.0083age + 0.75 males
ì -0.05age + 33.3 females

G2 = í , (10.23)
î-0.32age + 52.9 males
where Ps is systolic BP, Pd is diastolic BP and >50 mmHg, and age = 30 − 70 years

[8]. These formulas may similarly be thought of as representing an average person
for each age and gender. Hence, G1 and G2 are person- and time-specific parameters.
However, models relating one time delay to systolic BP and diastolic BP will obvi-
ously breakdown when the two BP values diverge.
Most of the PTT-BP models above are line equations in some function of PTT
per unit length (x; see Eqs. (10.11)–(10.13), Eqs. (10.15), (10.17), and (10.18) after
log transformation as well as Eq. (10.21)). The two parameters of these models can
thus be determined with basic linear least squares regression [36, 37] as follows:
N
å ( Pi - P ) ( xi - x )
K̂1 = i =1
N (10.24)
å ( xi - x )
2
i =1
Kˆ 2 = P - Kˆ 1 x , (10.25)
where ( ×) denotes estimate, N (>1) is the number of data pairs, and ( ×) indicates the
average. By assuming that the true PTT-BP relationship is as follows:
Pi = K1 xi + K 2 + ei , (10.26)
where Kj are the true parameters, xi is deterministic, and ei ~  0,s 2 , indepen- ( )

dent, and represents other contributions to BP (e.g., PEP and smooth muscle con-
traction [1]), the expected value and covariance matrix of the parameter error are as
follows:
æ é Kˆ - K1 ù ö é0 ù
Eçê 1 ú÷ = (10.27)
ç ê Kˆ - K ú ÷ êë0 úû
èë 2 2ûø
æ é Kˆ - K1 ù ö s2 é1 -x ù
1
Eçê 1
ç ê Kˆ - K ú
ú éë Kˆ 1 - K1 Kˆ 2 - K 2 ùû ÷ =
÷ N 1
ê
ê- x 1 N 2 úú . (10.28)
èë 2 2û ø
N
å(x - x)
2 å xi
N i =1 i
êë N i =1 úû

These formulas indicate that the parameter estimates are unbiased and become more
1
precise with increasing data pairs and variations in x (e.g., åiN=1 ( xi - x ) is the
2
sample variance of x). N

The statistics of the parameter error in Eqs. (10.27) and (10.28) can be mapped
to the expected value and variance of the error in predicting a new BP from a new x
as follows:

( ) (( )
E PˆN +1 - PN +1 = E Kˆ 1 x N +1 + Kˆ 2 - ( K1 x N +1 + K 2 + eN +1 ) = 0 )
(10.29)
æ ö
s 2 ç ( x N +1 - x )
2
÷
( )
2
E PˆN +1 - PN +1 = ç N + 1÷ + s 2 . (10.30)
N ç 1
ç å ( xi - x ) ÷÷
2
è N i =1 ø
These formulas indicate that the BP prediction is also unbiased with variance com-
prising two components, namely the variance due to the parameter error (the first
term on the right hand side of Eq. (10.30)) and the variance due to the unpredictable
error (the second term on the right hand side of Eq. (10.30)). The formulas quanti-
tatively underscore the need for interventions of marked effect. For instance, con-
sider the practical N = 2 case in which PTT-BP data pairs from baseline and one
intervention are obtained. Equation (10.30) then reduces to the following formula:
s2 ææ x - x ö
2
ö
( )
2
E Pˆ3 - P3 = ç ç 2 3 1 - 1÷ + 1÷ + s 2 . (10.31)
2 ç è x2 - x1 ø ÷
è ø
So, as long as x3 (the new measurement) is between x1 and x2 (the two measurements
used to determine the model parameters), the first variance will be less than the
intrinsic variance. Else, the first variance will grow quadratically as x3 deviates from
this range and thereby dominate the total BP prediction error variance. As an exam-
ple, let P1 = 80 mmHg, P2 = 90 mmHg, and P3 = 120 mmHg and assume values for
the true model parameters from Eq. (10.32) below (for any age) to compute the
corresponding three values for x. The standard deviation of the BP prediction error
would then be about 5σ, where σ is likely on the order of a few mmHg. Hence, con-
sistent with intuition, the formulas suggest that the average BP range elicited by the
baseline period and intervention should encompass the desired BP prediction range
of the PTT-based monitor. Interventions that increase and decrease BP a lot (e.g.,
±25–50 mmHg) would be ideal.
The PTT-BP model of Eq. (10.16) is a quadratic equation in x. Hence, its param-
eters can likewise be determined analytically with linear least squares regression.
However, at least three PTT-BP data pairs are needed to determine the three model
parameters. The models of Eqs. (10.14) and (10.20) are not linear in all parameters.
Hence, the parameters must be determined via a numerical search. For example,
nonlinear least squares with an initial guess for the parameters may be applied for a
fast solution to the local minimum [36]. Alternatively, since there are only two or
three parameters, least squares minimization by an exhaustive search over the phys-
iologic range of the parameters may be performed to ensure attainment of the global
minimum.
Advanced methods could be employed to determine the model parameters. For
example, maximum likelihood estimation for Laplacian distributed data may be
applied. Such estimation may provide superior parameter estimates than least
squares estimation (which assumes Gaussian distributed data) when there are outli-
ers in the data. Alternatively, the outliers could simply be discarded prior to least
squares estimation. Another potentially useful method is total least squares estima-
tion [36], which assumes that there is error in both BP and PTT. PTT error can be
nontrivial when the waveforms are measured with modalities such as ballistocardi-
ography [38] and electrical bioimpedance [1].
It may be possible to determine all PTT-BP model parameters without invoking
an intervention by exploiting the natural pulsatile variation in BP (i.e., BP nominally
varies from 80 to 120 mmHg with each heartbeat). However, note that simply detect-
ing multiple time delays between proximal and distal waveforms at different BP
levels may not be effective, because the pressure wave is reflected at the microcir-
culation and returns to heart typically during systole. Hence, because of wave reflec-
tion, the peak-to-peak time delay can even be negative. A recent study proposed to
employ a nonlinear model method instead [39]. The model accounts for the depen-
dency of arterial compliance on BP and wave reflection. The parameters of the model
are estimated so as to optimally couple proximal and distal BP waveforms in the least
squares sense. PTT as a function of BP is defined by the parameter estimates. In this
way, PTT for each BP level in the cardiac cycle is effectively estimated after mathe-
matically eliminating the reflected wave. The PTT-BP function can then serve as a
person-specific calibration curve. However, the requirement of BP waveforms as

input to the method is a practical disadvantage.
A number of studies have tested person-specific methods in terms of BP predic-
tion accuracy. The studies are too different methodologically to meaningfully aggre-
gate their results. However, a significant fraction of studies are common in that they
measured finger PAT or a similar time delay, employed a two-parameter PTT-BP
model, determined the model parameters using a BP intervention such as those in
Table 10.1 in conjunction with linear least squares regression, and tested the result-
ing calibration curve using unseen data [1]. For these studies, the unweighted aver-
age BP root-mean-squared-error is about 8 mmHg. The error for diastolic BP was
often lower than for systolic BP likely due to the fact that diastolic BP changed less
than systolic BP. This overall error level is within the bias and precision error limits
of 5 and 8 mmHg for automatic cuff devices set forth by the AAMI (Association for
the Advancement of Medical Instrumentation [40]). However, these methods and
other person-specific methods in the literature have not been tested comprehen-
sively enough to draw any definitive conclusions on their performance.
Population-Based Methods
The population-based method for determining the PTT-BP model parameters

involves using readily available information about the person in conjunction with a
training dataset comprising cuff BP and PTT measurements from a cohort of differ-
ent subjects to determine all parameters. By avoiding measurement of cuff BP or
application of a BP intervention in the person, the method is most convenient but
least accurate.
An extensive training dataset is needed to have a chance at producing a reason-
ably accurate, population-based calibration curve. The more extensive, the better. It
would also be ideal, if the data from each subject comprised cuff BP and PTT mea-
surements during a battery of interventions (see Table 10.1) to change BP widely
and in different ways. The model parameters could then be determined for a new
person by leveraging both the intervention-induced and inter-subject BP variations
for comparable subjects in the dataset. However, excluding interventions in the
training dataset is more practical and would likely allow for a larger and more
diverse subject cohort. In this case, only the BP variations between comparable
subjects could be used to specify the model parameters. A large training dataset
without interventions may yield relatively accurate BP readings amongst different
people, whereas a smaller training dataset with interventions may result in more
accurate tracking of BP changes in an individual person.
The model parameters must vary from person to person and should thus depend
on some information about the person in a quantitative way. Relevant but conve-
nient information include age, gender, weight, and race. Note that height may be
accounted for in the model by the measurement of l. Information about cardiovascular
health may also be included such as cholesterol level, smoking history, and presence
of diabetes.
The PTT-BP model of Eq. (10.14) along with Eqs. (10.9) and (10.10) provide an
exemplary basis for implementing the population-based method. Since the numeri-
cal values in the latter equations were determined to fit pressure-area data of ex vivo
human aortas, the values should be re-estimated based on a relevant training dataset
as described above. Nonlinear least squares could be performed to estimate the
values and thereby yield two model parameters that are fully defined by age and
gender. Equations (10.9) and (10.10) could also be extended to include other poten-
tially important regressors such as those described above.
Multiple groups, especially from industry, may have already attempted to imple-
ment the convenient, population-based method. However, few studies have been
published.
The most relevant study [41] was performed as follows. Training data compris-
ing serial measurements of invasive BP and ear and toe photo-plethysmography
(PPG) waveforms were collected from four cohorts of patients without cardiovascu-
lar disease during surgery. The four cohorts were: young, males; young, females; old,
l
males; and old, females. The parameters of the model of Eq. (10.18) with x = ,
t
where P is diastolic BP and τ is the foot-to-foot time delay between the two wave-
forms, was determined for each group. The population-based calibration curve was
then tested using the same measurements from new patients but of similar charac-
teristics. The bias and precision errors of the PTT-based diastolic BP measurements
were within AAMI limits. The model parameters were not reported, but the four
calibration curves are shown in the publication [41] (see Fig. 7 therein).
Two other studies provide insightful data pertaining to the population-based
method. One study involved detecting PTT as the foot-to-foot time delay between
invasive proximal and distal aortic BP waveforms in patients of various ages with
dilated cardiomyopathy but without coronary artery disease before and during nitro-
prusside administration to reduce BP [13]. A population-based PTT-BP model with
age dependency was presented as follows:
l
= ( 0.00131age - 0.0168 ) Pd + 3.35, (10.32)
t
where l/τ is PWV in units of m/s, age is in units of years, and R2 = 0.71 for the model
fit. Another study involved detecting PTT as the foot-to-foot time delay between
noninvasive carotid and femoral artery waveforms in diverse subjects (e.g.,
20–91 years, 49–97 mmHg for diastolic BP) before and after a hydrostatic maneu-
ver (supine to sitting) [14]. The following PTT-BP model was reported:
l
Pd = ( 22 ± 14 ) + K 2 , (10.33)
t
where the units are the same. The slope parameter not only varied widely but was
only modestly correlated with age, heart rate, BP, and weight (R2 = 0.25 for the
model fit). The K2 parameter also appeared to vary widely. Note that the validity of
the population-based method is supported more by Eq. (10.32) than Eq. (10.33).
While a few studies have been promising, the population-based method is essen-
tially unproven.
Hybrid Methods
The hybrid method for determining the PTT-BP model parameters generally
involves: (1) using readily available information about the person in conjunction
with a training dataset comprising cuff BP and PTT measurements from a cohort of
different subjects to determine all parameters but one and (2) using one pair of cuff
BP and PTT measurements from the person to determine the remaining parameter.
This method does not involve invoking an intervention in the person and is thus
more convenient than the person-specific method. At the same time, the method
does require measuring cuff BP in the person and is thus more accurate than the
population-based method. The hybrid method represents a balance between accu-
racy and convenience. It is thus popular not only for mapping PTT to BP but also
PTT plus covariates to BP (see below) and even for non-PTT modalities such as
single PPG waveform analysis [42].
Only a few basic embodiments of the hybrid method may have been described in
the literature. One method employed the model of Eq. (10.11) to relate finger PAT
to mean BP and experimental elastic modulus-BP data to set K1 = − 64.5 mmHg for
healthy subjects [43]. The intercept K2 may then be trivially determined from base-
line measurements of finger PAT and cuff BP in the person. This method was later
adapted for older, hypertensive subjects by setting K1 = − 22.2 mmHg [44]. Another
method employed a five-parameter model to relate finger PAT to systolic BP and
determined constant values for four of the parameters using a training dataset com-
prising finger PAT and cuff BP measurements before and after exercise [45]. The
remaining intercept parameter for a new person may then likewise be determined
using baseline cuff BP measurements. This latter method is apparently the basis for
the “Somnotouch” PTT-based BP monitor, which has notably been cleared by the
US Food and Drug Administration [46]. However, similar to the person-specific and
population-based methods, all hybrid methods require further assessment before
conclusions can be reached on their performance.
Other embodiments of the hybrid method may be envisaged. For example, the
population-based method could be implemented as described above to yield param-
eters that are functions of basic person information, and the parameter that is least
predictable from this information may then be determined using the pair of cuff BP
and PTT measurements from the person. As another example, the single pair of cuff
BP and PTT measurements could be used to constrain the model parameters instead
of specifying one parameter. One way to implement this embodiment is for the
model to be optimally fitted to the PTT-BP pairs of comparable subjects in the training
dataset in the least squares sense under the constraint that the model fits the baseline
cuff BP-PTT pair of the new person without error.
odels and Parameter Determination Using Additional

M
Waveform Features
There is increasing interest in incorporating additional waveform features in PTT-

based BP monitors. Exploiting features from the waveforms that are already being
measured to detect PTT may improve BP prediction accuracy from at least two
standpoints. Firstly, including additional features with PTT would allow the possi-
bility of independently tracking diastolic BP and systolic BP. Secondly, employing
other features may potentially help compensate for factors that obscure the PTT-BP
relationship such as PEP and smooth muscle contraction. However, in general,
model parameter determination becomes more difficult as the number of features in
the model increases.
As discussed above, PTT detected at the feet of proximal and distal arterial
waveforms may strictly correspond to diastolic BP, whereas PAT, which includes
PEP, may best correlate with systolic BP. While both time delays could be mea-
sured, it would be more convenient to obtain only one time delay and extract other
features from the two available waveforms in order to track PP and thereby enable
separate prediction of both BP values. Integrating Eq. (10.2) over the systolic inter-
val of the cardiac cycle (while assuming that PTT does not vary over this interval)
indicates that the absolute value of PP can be calculated from PTT and the arterial
cross-sectional areas at systole and diastole (As and Ad) as follows:
2 2
ælö æA ö æ l ö æ A - Ad ö
PP = r ç ÷ ln ç s ÷ » r ç ÷ ç s ÷. (10.34)
è t ø è Ad ø è t ø è Ad ø
This PP formula has been validated based on local measurements of PWV and arte-
rial areas in particular [15–17]. Although direct measurement of the arterial areas is
difficult (e.g., requiring expert-operated ultrasound), Eq. (10.34) suggests that con-
venient surrogates of the arterial areas along with PTT may allow tracking of PP. It
may be feasible to extract such features from PPG and electrical bioimpedance
(EBI) waveforms, which are often measured to detect PTT. The peak-to-peak ampli-
tudes of these waveforms are particularly sensitive to the pulsatile arterial volume
or area (i.e., As − Ad). However, as shown in Eq. (10.34), the absolute area is needed
to compute PP without any initialization. Another potentially convenient feature is
the peak-to-peak amplitude of the ballistocardiography (BCG) waveform. Through
mechanistic modeling, this so-called “J-K” amplitude was shown to be equal to PP
at the outlet of the descending aorta scaled by the descending aortic area [47].
Hence, use of the peak-to-peak amplitude of a convenient waveform should be
accompanied by an additional model parameter (pertaining to arterial area) to arrive
at PP in units of mmHg. Exemplary models relating PTT detected at the waveform

feet and a peak-to-peak PPG or EBI waveform amplitude (y) to both BP values may
be as follows:
l
Pd = K1 + K2 (10.35)
t
2
l ælö
Ps = K1 + K 2 + K 3 ç ÷ y. (10.36)
t èt ø
Note that the additional K3 parameter here could be determined simply with a single
cuff measurement of PP. Also note that the peak-to-peak amplitudes of convenient
waveforms could change due to other reasons (e.g., ambient lighting variations for
the PPG waveform) and that the model parameters obtained from one BCG instru-
ment may not generalize to another instrument due to variability in the BCG mor-
phology with respect to the recording instrument [48].
Several studies have sought to exploit the combination of a convenient peak-to-
peak waveform amplitude and PTT for cuffless BP measurement. In one study [49],
a proximal BCG waveform was measured in a subject standing on a high-
performance force plate, while a distal instep PPG waveform was also obtained with
a foot strap. A three-parameter, line model was defined to compute systolic BP from
PTT detected between the BCG I-wave and PPG trough and from the BCG J-K
amplitude. All model parameters were determined using systolic BP cuff measure-
ments before and after three BP interventions in conjunction with linear least
squares regression. The model predicted systolic BP significantly better than a
model including PTT alone. In another study [50], an EBI waveform from the wrist
and PTT detected as the time delay between wrist and finger PPG waveforms were
measured. A pair of one-parameter models was proposed to relate the EBI extrema
and PTT to diastolic BP and systolic BP. The model parameters were then deter-
mined using a single cuff measurement. The model predicted both BP values well
during exercise. In two studies by the same group, a pair of one-parameter models
was first proposed to relate the ratio of the peak to valley of the finger PPG wave-
form to diastolic BP and this ratio and PAT squared to systolic BP [11], and another
pair of one-parameter models was then proposed to relate the same ratio to mean BP
and the product of this ratio and PAT squared to PP [51]. After determining the
model parameters using a single cuff measurement, both models generally improved
the BP prediction accuracy over models involving PAT alone. However, there may
be some ambiguity regarding the use of the PPG amplitude as a predictor of dia-
stolic BP and mean BP instead of PP.
Aortic PTT measurements may change due primarily to BP over time periods in
which cardiovascular aging is not a factor [1], but practical PTT measurements such
as finger PAT are also impacted by a number of confounding factors. The most sig-
nificant confounding factors are PEP in the case of PAT and smooth muscle contrac-
tion in the case of PTT through peripheral arteries [1]. PEP can increase or decrease
with increasing BP [10] and may be thought of as an additive noise source, whereas
variations in smooth muscle contraction alter the PTT-BP relationship (e.g., the line
relating PWV to BP may become less steep as the smooth muscle contracts). Both
PEP and smooth muscle contraction can be modulated over the course of seconds to
minutes by the autonomic nervous system. So, it may be reasonable to use indices
of autonomic nervous function as features to compensate for these confounding fac-
tors. Heart rate (HR), RR interval or HR variability (HRV), and pulse rate variabil-
ity, which can be extracted from the waveforms being measured to detect PTT,
reflect autonomic nervous function. In particular, increases in HR indicate enhanced
sympathetic nervous function and/or reduced parasympathetic nervous function;
increases in the high frequency (> 0.15 Hz) power of HRV reflects augmented para-
sympathetic nervous function, while increases in the ratio of the low fre-
æ LF ö
quency(0.04–0.15 Hz) power to high frequency power of HRV ç ÷ indicates
è HF ø
higher sympatho-vagal balance [52]; and pulse rate variability correlates with HRV
[53] and can be obtained when an ECG waveform is unavailable. Beat-to-beat PTT
variability may reflect beat-to-beat BP variability, which is also a result of auto-
nomic nervous control. An exemplary model relating peripheral PTT detected at the
level of diastole and an autonomic nervous index to BP may be as follows:
æ LF ö l
Pd = ç K1 + K 2 + K3 , (10.37)
è HF ÷ø t
where K2 < 0. Determination of the model parameters here would require three
independent pairs of arterial waveforms and cuff BP measurements but could be
performed via linear least squares regression. Note that the adverse impact of PEP
and smooth muscle contraction may at best be partially mitigated via incorporation
of autonomic nervous system indices, because these confounding factors can be
modulated by other mechanisms (e.g., medications).
Several studies have sought to exploit the combination of autonomic nervous
system indices and PTT for cuff-less BP measurement. A few studies have exam-
ined HR and arm or finger PAT in a three-parameter line model. One study used cuff
measurements during baseline and exercise in conjunction with linear least squares
regression to determine the model parameters [54], while another study leveraged
invasive BP and natural BP variations in intensive care patients to determine the
parameters via the same regression method [55]. Each model predicted BP in test-
ing data better than a model with PAT alone as input. Note however that HR can rise
even when BP falls due to compensatory physiologic mechanisms, and PEP does
not change with pure chronotropic interventions such as atrial pacing [56]. A third
study proposed a pair of one-parameter models to compute diastolic BP and sys-
LF
tolicBP from , finger PAT, the ratio of the peak to valley of the PPG waveform,
HF
and other features [57]. After determining the model parameters with a single cuff
measurement, the model predicted both BP values on subsequent days better than
models using PAT alone or PAT and the ratio of the PPG peak to valley. The improve-
mentwas most profound in hypertensive subjects and attributed to the inclusion of
LF
,which was higher in these subjects. However, the proposed model was not
HF
dimensionally valid. A fourth study proposed a four-parameter line model to com-

pute both BP values from finger PAT and two metrics of PAT variability [58]. To
determine the model parameters, two of the parameters were set to constants, while
the other two parameters were determined using cuff measurements during sitting
and standing in conjunction with linear least squares regression. The model pre-
dicted diastolic BP in a separate sitting posture session significantly better than a
model of PAT alone.
In addition to exploiting select features based on physiologic principles, machine
learning has also been applied to discover waveform features beyond PTT for cuff-
less BP monitoring. Such data mining has been applied mainly to ECG and PPG
waveforms as described below but work on the BCG waveform can also be found
[59]. The general steps are to: (1) extract a large number of features from the wave-
forms and their derivatives (e.g., the entire waveforms and various time delays
between the two); (2) apply dimensionality reduction (e.g., principal or independent
components analyses) for a more tractable number of features; (3) define a paramet-
ric model (e.g., linear regression or support vector machine) to relate the features to
BP values; and (4) determine the model parameters from a training dataset that
includes cuff BP measurements. Since these models often contain more than two or
three parameters, the parameters are typically determined with the aid of a continu-
ous BP monitor for capturing more BP variations. One study employed invasive BP
and natural BP variations in intensive care patients to define the model [60], whereas
two other studies employed a finger cuff, volume-clamp device during exercise and/
or a Valsalva maneuver to build the model [61, 62]. The former study indicated that
nonlinear models are needed and that such models could yield BP precision errors
that are about one-third to one-half lower than the BP variances in unseen data.
However, the important waveform features beyond PAT were not reported. The
models in the latter two studies either did not significantly improve the BP predic-
tion accuracy relative to a model of PAT alone or required different features per
subject to improve the accuracy.
Figure 10.5 summarizes waveform features that could potentially improve the
accuracy of PTT-based monitors. However, much more evidence is needed to come
to a conclusion on whether or not these and other features actually add value
overall.
Required Time Period for Reinitialization
The PTT-BP relationship of a person evolves with aging and disease (see Fig. 10.2b).
To account for these phenomena, the PTT-BP model parameters for a person must
be periodically updated over time. In the case of the person-specific and hybrid
methods, each update would require at least making cuff BP measurements. As
discussed above, the PTT-BP relationship of a person may also change with smooth
muscle contraction. However, these changes can occur on the order of seconds to
Fig. 10.5 Additional features from the waveforms that are already being measured to detect PTT
or its popular pulse arrival time (PAT) surrogate may be incorporated into the PTT-BP model to
improve accuracy. Potentially useful features in commonly employed ECG, photo-plethysmography
(PPG), and ballistocardiography (BCG) waveforms include peak-to-peak amplitudes to track pulse
pressure (systolic BP – diastolic BP) and the standard deviation of the RR intervals (RRIV) or
pulse intervals and PTT or PAT (PTTV or PATV) over multiple beats to obtain autonomic nervous
system indices and thereby partially compensate for confounding physiology such as the pre-
ejection period in PAT
minutes. So, it does not make sense to make cuff BP measurements to try to com-
pensate for such fast phenomena. It would obviously be desirable to perform the
cuff initializations as infrequently as possible. However, there is a paucity of infor-
mation on the maximum reinitialization period that would not compromise accu-
racy. A few recent studies do shed some insight.
The PTT-BP model of Eq. (10.14) with Eqs. (10.9) and (10.10) were employed
to predict the maximum time period between cuff initializations that would not
introduce significant BP error [8]. This model indicates how the calibration curve
evolves with aging (see Fig. 10.2b). The maximum reinitialization period was thus
predicted by determining the longest time period for the calibration curve to change
by no more than 1, 2, or 3 mmHg over a physiologic range of BP for each age and
gender. Figure 10.6 illustrates the predicted maximum reinitialization period as a
function of age for males. The results for females were similar. As can be seen, the
maximum reinitialization period declined significantly with age. The maximum
reinitialization period is at least about 1 year for a 30 year old and about 6 months
for a 70 year old. This means that a 30 year old would require the next cuff initial-
ization at 31 years old, while a 70 year old would require the next cuff initialization
at 70.5 years old. Note that these predictions specifically pertain to aortic PTT.
The reinitialization period was also recently examined in two studies of human
subjects. In one of the studies [63], finger PAT was measured in 85 diverse subjects
and mapped to BP using a hybrid method. The hybrid calibration curve was then
tested 15 min, 2 weeks, and 1 month later. Overall, the BP precision errors increased
by 50% after 1 month. However, the systolic BP precision error was high even after
just 15 min (10.7 mmHg), and the BP bias errors did not change over the 1-month
study. Hence, these results suggest that the calibration curve changed due to varia-
Fig. 10.6 The maximum period between cuff initializations that would not compromise accuracy
as a function of age predicted using the model shown in Fig. 10.2b [8]. Here, 1, 2, and 3 mmHg
correspond to the largest tolerable BP error over a physiologic BP range for either gender. The
predictions indicate that the maximum reinitialization period declines with age and is >6 months
tions in PEP and/or smooth muscle contraction rather than 1 month of aging. In the
other study [61], finger PAT and additional waveform features were measured in 10
healthy subjects and mapped to BP using a hybrid method. The hybrid calibration
curve was then tested 1 day, 3 days, and 6 months later. The BP errors increased
significantly after just 1 day but then remained stable over the 6-month period.
These results likewise suggest that the calibration curve changed with PEP and/or
smooth muscle contraction but not 6 months of aging. Note that these data pertain
to measurements of PTT through peripheral arteries and that aging impacts periph-
eral PTT less than aortic PTT [1].
Based on the aforesaid studies, it may be reasonable to perform cuff initializa-
tions every few months. However, further studies are surely needed to arrive at a
conclusion.
Conclusions
In summary, PTT-based BP monitors have gained popularity, because they can

potentially permit cuffless and passive measurement of BP in a principled way.
However, initialization of these monitors so that PTT in units of ms can be mapped
to BP in units of mmHg is a serious challenge, as the PTT-BP relationship is defined
by at least two person- and time-specific parameters. Parametric models inspired by
theory or data are available to relate PTT to BP, with some constituting a practical
calibration curve form. There are three methods for determining the model
parameters that reflect different trade-offs between accuracy and convenience. A

person-specific method determines all of the parameters from multiple cuff BP
measurements during BP interventions in the person and is thus most accurate but
least convenient. A population-based method determines all parameters from only
basic information about the person in conjunction with a training dataset that
includes cuff BP measurements from other people and is thus most convenient but
least accurate. A hybrid method determines one parameter from a single cuff BP
measurement in the person and the remaining parameter(s) from basic person infor-
mation plus a similar training dataset and thus balances accuracy and convenience.
Additional features of the waveforms obtained to detect PTT such as peak-to-peak
amplitudes and autonomic nervous system indices may be incorporated in the model
to improve accuracy. Limited evidence seems to suggest that the model parameters
must be updated every several months to account for changes due to cardiovascular
aging. Hence, PTT-based monitors may provide ultra-convenient BP measurements
only in the periods between cuff initializations.
Our recommendations for initializing PTT-based BP monitors are as follows. We
currently advocate for Eq. (10.14) to serve as the calibration curve form, because it
can represent a linear in PWV relationship and change shape with aging using only
two parameters. In the event that this or another model yields unrealistic BP, we
further suggest that the monitor not output BP. This conservative approach assumes
that the PTT input is fraught with error. For the person-specific method to determine
the model parameters, we presently endorse exercise or a hydrostatic maneuver as
BP interventions. Exercise may be convenient (or necessary) for many and can raise
and then lower BP with respect to pre-exercise levels. Hand lowering and raising
and, to a lesser extent, postural changes are simple and can increase and decrease
local BP by a substantial and graded amount. Moreover, these interventions induce
BP changes that can be captured with a standard automatic cuff. For the population-
based and hybrid methods to determine the model parameters, it may possibly be
worthwhile to form a rich training dataset to maximize the chances for success by
applying a battery of interventions (cold pressor to increase BP via vasoconstric-
tion, mental arithmetic to increase BP via cardiac output, Valsalva maneuver to
decrease BP, and perhaps others) and a finger cuff, volume-clamp device to capture
all of the BP changes in large number of subjects with diverse BP. Such data could
be collected as part of a large, multicenter study and made available to the entire
research community for analysis. In addition, we believe that it is essential to
include a waveform amplitude feature with PTT to track PP and thereby indepen-
dently measure systolic BP and diastolic BP. For the person-specific method, we
advise to only add this one feature to maintain tractability of the parameter determi-
nation. Finally, we recommend reinitializing PTT-based BP monitors more fre-
quently than what may be necessary such as every week or month to ensure maximal
accuracy.
A key advantage of a PTT-based BP monitor is that it can yield many measure-
ments over time. These measurements can be averaged to not only mitigate the BP
prediction error but also eliminate intra-person BP variability. In this way, the moni-
tor may be useful for hypertension screening despite errors in individual measure-
ments that exceed the AAMI bias and precision limits [8]. Nevertheless, in our
opinion, it may still be difficult to map PTT to BP with enough accuracy under all
scenarios. Hence, it may be prudent to focus on niche areas such as nighttime and
24-h ambulatory BP monitoring, which are clinically important [64, 65] and for
which cuff usage is especially problematic. For example, PTT-based monitors could
be initialized in a person using cuff BP measurements before going to bed and after
getting up to potentially yield sufficient accuracy, while entirely avoiding disruptive
cuff inflations, during sleep [44].
In conclusion, PTT-based BP monitors may be promising, but substantial
research is still needed to initialize the monitors so as to yield absolute BP measure-
ments that are clinically useful. Our intent for this chapter is to serve as a foundation
for embarking on this research.
Acknowledgements This work was supported by the National Institutes of Health under Grant
EB-018818.
References
1. Mukkamala R, Hahn J-O, Inan O, Mestha L, Kim C-S, Toreyin H, Kyal S. Towards ubiquitous
2015;62:1879–901.
2. Hughes DJ, Babbs CF, Geddes LA, Bourland JD. Measurements of Young’s modulus of elas-
ticity of the canine aorta with ultrasound. Ultrason Imaging. 1979;1:356–67.
3. Wesseling KH, Jansen JRC. Computation of aortic flow from pressure in humans using a non-
linear, three-element model. J Appl Physiol. 1993;74:2566–73.
4. Tanaka H, Dinenno FA, Monahan KD, Clevenger CM, DeSouza CA, Seals DR. Aging, habit-
ual exercise, and dynamic arterial compliance. Circulation. 2000;102:1270–5.
5. Ma Y, Choi J, Hourlier-Fargette A, Xue Y, Chung HU, Lee JY, Wang X, Xie Z, Kang D, Wang
H, Han S, Kang S-K, Kang Y, Yu X, Slepian MJ, Raj MS, Model JB, Feng X, Ghaffari R,
Rogers JA, Huang Y. Relation between blood pressure and pulse wave velocity for human
arteries. Proc Natl Acad Sci. 2018;115:11144–9.
6. Pruett JD, Bourland JD, Geddes LA. Measurement of pulse-wave velocity using a beat-
sampling technique. Ann Biomed Eng. 1988;16:341–7.
7. Muehlsteff J, Aubert XL, Schuett M. Cuffless estimation of systolic blood pressure for short
effort bicycle tests: the prominent role of the pre-ejection period. In: Annual International
Conference of the IEEE Engineering in Medicine and Biology—Proceedings; 2006. p. 5088–
92. https://doi.org/10.1109/IEMBS.2006.260275.
8. Mukkamala R, Hahn J-O. Toward ubiquitous blood pressure monitoring via pulse transit time:
predictions on maximum calibration period and acceptable error limits. IEEE Trans Biomed
Eng. 2018;65:1410–20.
9. Thomas SS, Nathan V, Zong C, Soundarapandian K, Shi X, Jafari R. BioWatch: a noninvasive
wrist-based blood pressure monitor that incorporates training techniques for posture and sub-
ject variability. IEEE J Biomed Heal Inform. 2016;20:1291–300.
10. Gao M, Olivier NB, Mukkamala R. Comparison of noninvasive pulse transit time estimates
as markers of blood pressure using invasive pulse transit time measurements as a reference.
Physiol Rep. 2016;4:e12768.
11. Ding X, Zhang Y, Liu J, Dai W, Tsang HK. Continuous cuffless blood pressure estimation
using pulse transit time and photoplethysmogram intensity ratio. IEEE Trans Biomed Eng.
2016;63:964–72.
12. Wibmer T, Doering K, Kropf-Sanchen C, Rüdiger S, Blanta I, Stoiber KM, Rottbauer W,
Schumann C. Pulse transit time and blood pressure during cardiopulmonary exercise tests.
Physiol Res. 2014;63:287–96.
13. Carroll JD, Shroff S, Wirth P, Halsted M, Rajfer S. Arterial mechanical properties in dilated
cardiomyopathy. J Clin Invest. 1991;87:1002–9.
14. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio AP. Cuffless estimation of blood
pressure: importance of variability in blood pressure dependence of arterial stiffness across
individuals and measurement sites. IEEE Trans Biomed Eng. 2018;65:2377–83.
15. Seo J, Pietrangelo SJ, Lee H-S, Sodini CG. Noninvasive arterial blood pressure waveform
monitoring using two-element ultrasound system. IEEE Trans Ultrason Ferroelectr Freq
Control. 2015;62:776–84.
16. Beulen BW, Bijnens N, Koutsouridis GG, Brands PJ, Rutten MC, van de Vosse FN. Toward
noninvasive blood pressure assessment in arteries by using ultrasound. Ultrasound Med Biol.
2011;37:788–97.
17. Vappou J, Luo J, Okajima K, Di Tullio M, Konofagou EE. Non-invasive measurement of
local pulse pressure by pulse wave-based ultrasound manometry (PWUM). Physiol Meas.
2011;32:1653–62.
18. Geddes L, Voelz M. Pulse transit time as an indicator of arterial blood pressure.

19. Zhang G, Gao M, Xu D, Olivier NB, Mukkamala R. Pulse arrival time is not an adequate sur-
rogate for pulse transit time as a marker of blood pressure. J Appl Physiol. 2011;111:1681–6.
20. National Center for Health Statistics, Drizd T, Dannenberg AL, Engel A. Blood pressure levels
in persons 18–74 years of age in 1976–80, and trends in blood pressure from 1960 to 1980 in
the United States; 1986.
21. Joseph CN, Porta C, Casucci G, Casiraghi N, Maffeis M, Rossi M, Bernardi L. Slow breathing
improves arterial baroreflex sensitivity and decreases blood pressure in essential hypertension.
Hypertension. 2005;46:714–8.
22. Currens JH. A comparison of the blood pressure in the lying and standing positions: a study of
five hundred men and five hundred women. Am Heart J. 1948;35:646–54.
23. Hines EA Jr, Brown GE. The cold pressor test for measuring the reactibility of the blood pres-
sure: data concerning 571 normal and hypertensive subjects. Am Heart J. 1936;11:1–9.
24. Froelicher VF, Myers J. Exercise and the heart. Philadelphia, PA: Saunders; 2006.
25. Petrofsky JS, Lind AR. Aging, isometric strength and endurance, and cardiovascular responses
to static effort. J Appl Physiol. 1975;38:91–5.
26. Al’Absi M, Bongard S, Buchanan T, Pincomb GA, Licinio J, Lovallo WR. Cardiovascular
and neuroendocrine adjustment to public speaking and mental arithmetic stressors.
27. Kenney MJ, Seals DR. Postexercise hypotension. Key features, mechanisms, and clinical sig-
nificance. Hypertension. 1993;22:653–64.
28. Parati G, Casadei R, Groppelli A, Di Rienzo M, Mancia G. Comparison of finger and intra-
arterial blood pressure monitoring at rest and during laboratory testing. Hypertension.
1989;13:647–55.
29. Imholz BP, Wieling W, van Montfrans GA, Wesseling KH. Fifteen years experience with finger
arterial pressure monitoring: assessment of the technology. Cardiovasc Res. 1998;38:605–16.
30. Wesseling KH, De Wit B, van der Hoeven GMA, van Goudoever J, Settels JJ. Physical, cali-
brating finger vascular physiology for finapres. Homeostasis. 1995;36:67–82.
31. McCombie DB, Reisner AT, Asada HH. Motion based adaptive calibration of pulse transit time
measurements to arterial blood pressure for an autonomous, wearable blood pressure moni-
tor. In: 2008 30th Annual International Conference of the IEEE Engineering in Medicine and
Biology Society; 2008. p. 989–92. https://doi.org/10.1109/IEMBS.2008.4649321.
32. Wang Y, Liu Z, Ma S. Cuff-less blood pressure measurement from dual-channel photoplethys-
mographic signals via peripheral pulse transit time with singular spectrum analysis. Physiol
Meas. 2018;39:025010.
33. Rosner B, Polk BF. Predictive values of routine blood pressure measurements in screening for
hypertension. Am J Epidemiol. 1983;117:429–42.
34. Gavish B, Ben-dov IZ, Bursztyn M. Linear relationship between systolic and diastolic blood
pressure monitored over 24 h: assessment and correlates. J Hypertens. 2008;26:199–209.
35. Master AM, Lasser RP. The relationship of pulse pressure and diastolic pressure to systolic
pressure in healthy subjects 20–94 years of age. Am Heart J. 1965;70:163–71.
36. Keesman KJ. System identification—an introduction. Berlin: Springer; 2011. https://doi.
org/10.1007/978-0-85729-522-4.
37. Ljung L. System identification: theory for the user. Upper Saddle River, NJ: Prentice Hall;
1999.
38. Wiard RM, Inan OT, Argyres B, Etemadi M, Kovacs GTA, Giovangrandi L. Automatic detec-
tion of motion artifacts in the ballistocardiogram measured on a modified bathroom scale. Med
Biol Eng Comput. 2011;49:213–20.
39. Gao M, Cheng H-M, Sung S-H, Chen C-H, Olivier NB, Mukkamala R. Estimation of pulse
transit time as a function of blood pressure using a nonlinear arterial tube-load model. IEEE
Trans Biomed Eng. 2017;64:1524–34.
40. ISO 81060-2:2018 Non-invasive sphygmomanometers—part 2: clinical validation of auto-
mated manometer type. 2018. https://www.iso.org/standard/73339.html
41. Chen Y, Wen C, Tao G, Bi M, Li G. Continuous and noninvasive blood pressure measurement:
a novel modeling methodology of the relationship between blood pressure and pulse wave
velocity. Ann Biomed Eng. 2009;37:2222–33.
42. Watanabe N, Bando YK, Kawachi T, Yamakita H, Futatsuyama K, Honda Y, Yasui H, Nishimura
K, Kamihara T, Okumura T, Ishii H, Kondo T, Murohara T. Development and validation of a
novel cuff-less blood pressure monitoring device. JACC Basic Transl Sci. 2017;2:631–42.
43. Zheng Y-L, Yan BP, Zhang Y-T, Poon CCY. An armband wearable device for overnight and
cuff—less blood pressure measurement. IEEE Trans Biomed Eng. 2014;61:2179–86.
44. Zheng Y, Poon CCY, Yan BP, Lau JYW. Pulse arrival time based cuff-less and 24-h wearable
blood pressure monitoring and its diagnostic value in hypertension. J Med Syst. 2016;40:195.
45. Gesche H, Grosskurth D, Küchler G, Patzak A. Continuous blood pressure measurement
2012;112:309–15.
46. SOMNOtouch™ NIBP. https://somnomedics.eu/products/cardiology/24h-bloodpressure-
24h-ecg/somnotouch-nibp/
47. Kim C-S, Ober SL, McMurtry MS, Finegan BA, Inan OT, Mukkamala R, Hahn

J-O. Ballistocardiogram: mechanism and potential for unobtrusive cardiovascular health moni-
toring. Sci Rep. 2016;6:31297.
48. Wiens AD, Etemadi M, Roy S, Klein L, Inan OT. Toward continuous, noninvasive assessment
of ventricular function and hemodynamics: wearable ballistocardiography. IEEE J Biomed
Heal Inform. 2015;19:1435–42.
49. Kim C-S, Carek AM, Inan OT, Mukkamala R, Hahn J-O. Ballistocardiogram-based approach
to cuffless blood pressure monitoring: proof of concept and potential challenges. IEEE Trans
Biomed Eng. 2018;65:2384–91.
50. Huynh TH, Jafari R, Chung WY. Noninvasive cuffless blood pressure estimation using pulse
transit time and impedance plethysmography. IEEE Trans Biomed Eng. 2019;66:967–76.
https://doi.org/10.1109/TBME.2018.2865751.
51. Ding X, Yan BP, Zhang YT, Liu J, Zhao N, Tsang HK. Pulse transit time based continuous
cuffless blood pressure estimation: a new extension and a comprehensive evaluation. Sci Rep.
2017;7:11554.
52. Task Force of the European Society of Cardiology and the North American Society of Pacing
and Electrophysiology. Heart rate variability: standards of measurement, physiological inter-
pretation and clinical use. Circulation. 1996;93:1043–65.
53. Schäfer A, Vagedes J. How accurate is pulse rate variability as an estimate of heart rate vari-
ability?: a review on studies comparing photoplethysmographic technology with an electrocar-
diogram. Int J Cardiol. 2013;166:15–29.
54. Zhang Q, Zhou D, Zeng X. Highly wearable cuff-less blood pressure and heart rate monitor-
ing with single-arm electrocardiogram and photoplethysmogram signals. Biomed Eng Online.
2017;16:1–20.
55. Yoon YZ, Kang JM, Kwon Y, Park S, Noh S, Kim Y, Park J, Hwang SW. Cuff-less blood pres-
sure estimation using pulse waveform analysis and pulse arrival time. IEEE J Biomed Heal
Inform. 2018;22:1068–74.
56. Harris WS, Schoenfeld CD, Weissler AM. Effects of adrenergic receptor activation and block-
ade on the systolic preejection period, heart rate, and arterial pressure in man. J Clin Invest.
1967;46:1704–14.
57. Chen Y, Shi S, Liu YK, Huang SL, Ma T. Cuffless blood-pressure estimation method using a
heart-rate variability-derived parameter. Physiol Meas. 2018;39:095002.
58. Ma HT. A blood pressure monitoring method for stroke management. Biomed Res Int.
2014;2014:571623.
59. Yousefian P, Shin S, Mousavi A, Kim CS, Mukkamala R, Jang DG, Ko BH, Lee J, Kwon UK,
Kim YH, Hahn JO. Data mining investigation of the association between a limb ballistocardio-
gram and blood pressure. Physiol Meas. 2018;39:075009.
60. Kachuee M, Member S, Kiani MM, Member S. Cuffless blood pressure estimation algorithms
for continuous health-care monitoring. IEEE Trans Biomed Eng. 2017;64:859–69.
61. Miao F, Fu N, Zhang YT, Ding XR, Hong X, He Q, Li Y. A novel continuous blood pres-
sure estimation approach based on data mining techniques. IEEE J Biomed Heal Inform.
2017;21:1730–40.
62. Lin W-H, Wang H, Samuel OW, Liu G, Huang Z, Li G. New photoplethysmogram indica-
2018;39:025005.
63. Ding X, Zhang Y, Tsang HK. Impact of heart disease and calibration interval on accuracy of
pulse transit time–based blood pressure estimation. Physiol Meas. 2016;37:227–37.
64. Pickering TG, Shimbo D, Haas D. Ambulatory blood-pressure monitoring. N Engl J Med.
2006;354:2368–74.
65. Boggia J, Li Y, Thijs L, Hansen T, Kikuya M, Bjorklund-Bodegard K, Richart T, Ohkubo T,
Kuznetsova T, Torp-Pedersen C, Lind L, Ibsen H, Imai Y, Wang J, Sandoya E, O’Brien E,
Staessen J. Prognostic accuracy of day vs. night ambulatory blood pressure: a cohort study.
Lancet. 2007;370:1219–29.
Chapter 11
Key Regulatory Aspects
and the Importance of Consensus
Standards in Bringing Devices to Market
Carole C. Carey
Abstract Software and hardware innovations, smart devices, digitization in

healthcare, IoT (Internet of Things), and other emerging device technologies are
driving the medical device industry as one of the fastest growing markets around
the globe. Medical devices carry some measure of risk that can potentially cause
problems or result in harm impacting patient safety significantly. The challenge
manufacturers face is obtaining government marketing authorization for the pro-
posed new medical device before it can be legally sold in its respective jurisdic-
tion. This chapter examines key regulatory aspects and the importance of the use
of standards and their increasing importance as a regulatory tool in medical device
regulation. Five countries or regions with advanced medical device regulatory
systems were explored: the USA, European Union (EU), Canada, Australia, and
Japan.
Keywords Medical devices · Regulatory framework · Device classification · Risks

and controls · International consensus standards · Harmonization · Marketing
authorization
Background
The twenty-first century has brought a proliferation of innovative, unobtrusive, and

wearable personal devices both for clinical and consumer applications. The rising
trend is possible with the development of novel sensors, new materials, advances in
medicine, science, and biomedical engineering. Emerging technologies in nano-
technology, information technology, medical robotics, artificial intelligence, and
neurotechnologies in brain–computer interfaces—just to name a few, are impacting
C. C. Carey (*)
C3-Carey Consultants, LLC, Fulton, MD, USA
e-mail: c.carey@ieee.org

192 C. C. Carey
healthcare and personalized medicine. Wearable technology has the potential to

improve the management of various diseases with better clinical outcomes, reduced
costs (more affordable), and faster access to medical devices that can provide diag-
nostic, monitoring, and therapeutic solutions.
The medical devices industry is a highly regulated sector, particularly in jurisdictions
in different parts of the world with established regulatory systems. All devices carry a
certain degree of varying risk (low, moderate, and high) that can cause harm to the
human body and result in adverse event(s). Examples of devices range from thermom-
eters, stethoscopes to multi-function physiological monitoring systems to ventricular
bypass (assist) devices. Regulatory bodies share a common mission, to promote and
protect the public health of its citizens. Risks and benefits must be carefully weighed to
determine safety, effectiveness, performance, and quality before they can be sold and
distributed in the marketplace. The aim is to maximize benefit and minimize risk.
Introduction
This chapter focuses on five case studies in different parts of the world with well-
established regulatory systems: the USA, European Union (EU), Canada, Australia,
and Japan. It provides an overview of the regulatory frameworks and key aspects to
consider before placing a medical device in legal distribution. In all five different
jurisdictions, a risk-based classification scheme is used to guide regulators in the
evaluation of risks associated with a medical device. After the classification of the
device is determined, appropriate regulatory assessments and controls are applied.
The higher the risks to the patient and/or user, the greater is the level of control.
It is worthwhile mentioning that smaller countries are working together to har-
monize their medical device regulations in order to accelerate the development of
economic growth in the region, promote stability, and facilitate collaboration. Such
is the case with the ASEAN (Association of Southeast Asian Nations) member
states. The ten-member states are Brunei Darussalam, Cambodia, Indonesia, Laos,
Malaysia, Myanmar, Philippines, Singapore, Thailand, and Vietnam. They devel-
oped the ASEAN Medical Device Directive (AMDD) [1]. The Directive uses a
common format to simplify the registration process and application requirements
for manufacturers. Although implementation is still in process and some countries
are more developed than others, the cooperation allows for all the member states
provide a clearer regulatory pathway to market for manufacturers; thus, allowing
medical devices to reach the intended population sooner to benefit patients.
11 Key Regulatory Aspects and the Importance of Consensus Standards in Bringing… 193
Medical Device Definition
What is a medical device? In very broad terms, a device or equipment intended for
medical purposes is a medical device. The World Health Organization (WHO)
briefly described a medical device as,
An article, instrument, apparatus or machine that is used in the prevention, diagnosis or
treatment of illness or disease, or for detecting, measuring, restoring, correcting or modify-
ing the structure or function of the body for some health purpose. Typically, the purpose of
a medical device is not achieved by pharmacological, immunological or metabolic means.
This definition is an abridgement from the Global Harmonization Task Force (GHTF)
final document that was published by the Study Group 1 on May 12, 2016. [2] GHTF
no longer exists. However, their work is continuing under the International Medical
Device Regulators Forum (IMDRF) where the GHTF archived documents are kept for
preservation. While regulatory authorities across different jurisdictions have very simi-
lar definition (in content) of a medical device, it should be recognized that some prod-
ucts may not be considered medical devices in some jurisdictions. Devices in the gray
area zone may fall in different risk classification categories (borderline between two
different classes). In order to verify if a product meets the definition of a medical device
and its regulatory classification, manufacturers should consult with the appropriate
regulatory authority responsible for regulating medical devices in their jurisdiction.
The US Food and Drug Administration (FDA) regulatory definition of a medical
device can be found in Section 201(h) of the Food, Drug and Cosmetic Act. Health
Canada (HC) defines the term medical devices in the Food and Drugs Act, while
Australia’s Therapeutic Goods Administration (TGA) is stated in 41 BD of the
Therapeutic Goods Act 1989 (the Act). In the Government of Japan’s Ministry of
Health, Labour and Welfare (MHLW), it is identified in the Pharmaceuticals and
Medical Devices Act (PMD Act). The European Union medical device definition is
described in Article 1 of the Council Directive 93/42/EEC.
Regulatory Framework and Medical Device Classification
Each country or region recognizes the necessity to classify a medical device based
on its intended use and the level of risk posed to the patient and/or the user (Risk-
based Classification Scheme). There are several considerations when ensuring the
safety of a device, such as device features and characteristics, design complexity,
invasive or noninvasive, and local or systemic use. The environmental conditions of
use and whether it is radiation-emitting are other factors to consider as well as the
potential harm if misused. Medical devices can be as simple as a tongue depressor
to a wearable monitoring electrocardiograph (ECG) watch, an infusion pump, a
robotic surgery device—to more complex, high risk device that is life sustaining
and life supporting such as a cardiac implantable pacemaker/defibrillator.
194 C. C. Carey
The USA
The Center for Devices and Radiological Health (CDRH) is one of several Centers
within FDA responsible for regulating firms who manufacture, repackage, relabel,
and/or import medical devices sold in the USA. Additionally, CDRH also regulates
medical and non-medical radiation-emitting electronic products such as lasers,
X-ray systems, ultrasound equipment, microwave ovens, and color televisions. If a
product is both a medical device and an emitter, it must meet the regulations pertain-
ing to medical device as well as the requirements for radiation-emitting product.
Based on the laws set forth in the Food Drug and Cosmetic Act, most of FDA’s
medical device and radiation-emitting product regulations are detailed in Title 21
Code of Federal Regulations (CFR) Parts 800-1299 [1].
The FDA recognizes three levels of classification based on their risks and regula-
tory controls necessary to provide a reasonable assurance of safety and effective-
ness. The 1976 Medical Device Amendment Act created a risk-based classification
system and classified all medical devices into three classes, Class I, II, and III [2].
Device classification also considers the intended use of the device and the indica-
tions for use. For example, an external defibrillator is intended to restore a normal
heartbeat by sending an electric pulse or shock to the heart with an irregular and
abnormal heart beat (arrhythmia). It is indicated for use on victims of cardiac arrest
who are unconscious, not breathing and without circulation (without a pulse).
In Table 11.1 below are the three different classes in the FDA device classification
system, including the fact that a device may also fall under the exempt status. An
exempt device means the FDA will not require a premarketing application. However,
it is not exempt from the General Controls requirements. Most Class I devices are
Exempt. Some Class II devices may have also been exempted by regulation. In this
case, it can be viewed as three classes with four categories. General Controls are
basic requirements necessary for all devices irrespective of their classification, unless
exempted by regulation (example, a toothbrush). Devices must not be adulterated or
misbranded. Other general controls also include adherence to Good Manufacturing
Practices (GMP)/Quality System Regulations (QS Regulations), Premarket notifica-
tion (or 510(k)) unless exempt, Records and Reports (adverse event reporting),
device tracking, and UDI (unique device identification).
Table 11.1 US medical device classification

Classification Risk level Regulatory control/application
Class I, class II exempt Low General controls/exempt from 510(k)
Class I Low to moderate risk General controls/510(k)
Class II Moderate to high risk General controls, special controls/510(k)
Class III High risk General controls, premarket approval
(PMA
The classification of the device will determine the path to market and the premar-
keting application required. In general, devices under Class I or Class II (example,
blood pressure measuring devices) will require a premarket notification or 510(k)
submission to the FDA. An FDA clearance letter will be needed before going to
market. FDA Class III devices (example, automated external defibrillators) will
require a premarket approval (PMA) process. In this case, an FDA approval letter
will be needed before going to market. Scientific reviews of premarketing applications
are conducted by FDA. FDA has a Third-Party Review Program and accredits the
Third Party to review some Class II devices.
European Union
The EU does not have a “Food and Drugs Administration” but harmonizes their
efforts into one law that can be applied throughout the European Union. In close
cooperation with Member State’s Health Authorities, the European Commission
has the task of regulating medical devices and harmonizing requirements and legis-
lation adopted through a set of directives, collectively known as Medical Device
Directive (MDD). Manufacturers need to be aware of the current changes in the EU
legal regulatory framework for medical devices. Regulation (EU) 2017/745 [3]
repeals the existing Council Directives, Medical Devices (93/42/EEC) and Active
Implantable Medical Devices (90/385/EEC) and will officially replace them in
2020. EU Medical Device Regulation (MDR) 2017/745 was published on May 5,
2017 and came into force on May 25, 2017. Manufacturers with currently approved
medical device have 3 years transition period to meet the new EU MDR require-
ments. The implementation date for the European Union Medical Device Regulation
is May 22, 2020.
The device classification structure in the EU also follows a risk-based system.
However, determining classification and the path to market employs a different
structure. As a general rule, to gain market access in Europe will require obtaining
a CE (Conformité Européenne) marking approval. CE mark is the legal requirement
to place a product in the EU. With a CE mark, it signifies that a medical device
meets all relevant essential requirements of the European Directives that outline the
safety and performance requirements for medical devices in the European Union.
The body within the government of the EU Member States which has the authority
to ensure that the requirements of the CE marking directives are carried out in that
particular member state is known as Competent Authority. The certification organi-
zation that conducts the conformity assessment against the relevant EU Directives
(which the national Competent Authority of a member state designates to carry out
the procedure and issue the CE Certificate) is referred to as Notified Body (NB). The
conformity assessment normally involves an audit of the manufacturer’s quality
system and a review of manufacturer’s technical documentation on the safety and
performance of the device.
196 C. C. Carey
To decide what is required to CE mark a device, it is important to determine the

classification first. The EU rule-based classification scheme for medical devices is
set out in Annex IX of Directive 93/42/EEC within the MDD [4]. The European
Medicines Agency (EMA) is also involved in the assessment of certain categories of
medical devices. By following the rule-based system, each manufacturer could
determine the classification of its own device as early as possible in device develop-
ment; they could classify their own devices. Intended use determines the classifica-
tion. The classification rules consider criteria such as duration of contact with the
patient, degree of invasiveness, and the part of the body affected by the use of the
device. In the initial process of probable identification of device class, the devices
are segmented into broad categories: noninvasive, invasive, active, and special rules.
The conformity assessment routes lead to other routes (applicable certain rules)
within the directive. They are further segmented into classes as shown in Table 11.2.
The classification categories are Class I (including Is (sterile) & Im (measuring
function)), Class IIa, Class IIb and Class III, with Class III ranked as the highest risk.
The higher the classification, the greater is the level of assessment. With exception
of Class I, devices will generally require working with a Notified Body.
Canada
Health Canada’s Medical Devices Bureau of the Therapeutic Products Directorate

(TPD) is the government authority who reviews, evaluates, and monitors diagnostic
and therapeutic medical devices—to assess their safety, effectiveness and quality
before being commercialized in Canada. The regulatory and legal framework are
detailed under the authority of the Food and Drugs Act.
The medical device classification scheme is also based on risk and is rule-based
similar to the EU system. There are four device classifications using a set of 16 rules
found in the Canadian Medical Devices Regulations (CMDR) SOR/98-282 [5]
There is also a published Guidance on the Risk-based Classification System for
Non-In Vitro Diagnostic Devices [6]. Medical devices are classified into one of four
classes, Class I, II, III, and IV where I represents the lowest risk and IV represents
the highest risk (Table 11.3). In order to determine the appropriate classification of
the device, the Classification Rules for Medical Devices as detailed in Schedule 1 of
the Medical Device Regulations (Regulations) must be applied. As already men-
tioned, the rules were developed with strong similarities to the European Union’s
Table 11.2 The EU medical Classification Risk level

device classification system
Class I, including Is and Im Low to medium risk
(devices require CE mark)
Class IIa Medium risk
Class IIb Medium to high risk
Class III High risk
Table 11.3 Canada’s Classification Risk level Type of licence

medical device classification
Class I Lowest risk Establishment licence
system
Class II Low to medium risk Medical device licence
Class III Medium to high Medical device licence
Class IV Highest risk Medical device licence
Council Directive 93/42/EEC. It cannot be assumed, however, that a medical device

classified in one class according to the European Union’s classification system will
be classified in the same class based on the Canadian classification system or as in
other jurisdictions. Be aware that many Class I products in the USA may be assessed
as Class II products in Canada. Some USA Class II products could also fall into
Class III Canadian classification system.
With the exception of Class I devices, a Medical Device Licence is required for
Class II, III, and IV products before they can be sold in Canada. If the information
submitted in the Medical Device Licence Application meets the requirements of the
Medical Device Regulations, a License is issued. For Class I devices, they are moni-
tored through Establishment Licences. An Establishment Licence allows importers,
distributors and manufacturers to operate in Canada if they do not sell their medical
devices through a licensed importer or distributor.
Australia
At the time of this writing, the Australian Regulatory Guidelines for Medical
Devices (ARGMD) Version 1.1 (May 11, 2011) is currently under review.1
The Therapeutic Goods Administration (TGA) under the Australian govern-
ment’s Department of Health and Ageing oversees medical device regulation. A
major regulation is the Therapeutic Goods (Medical Devices) Regulations 2002 [7].
Australian TGA’s requirements are determined by the medical device intended pur-
pose and its classification category, similar to the risk-based management paradigm
in the USA, the EU, Canada, and Japan. It also adds separate fifth classification for
active implantable medical devices. A manufacturer determines classification of
their medical device using Schedule 2 of the Therapeutic Goods (Medical Devices)
Regulations 2002. The risk-based approach is matched to the level of regulation with
the risks presented by particular therapeutic goods. Table 11.4 lists the five main
classification for medical devices depending on the level of risk they pose, Class I,
Class IIa, Class IIb, Class III and Active Implantable Medical Devices (AIMD).
Every medical device is assessed against a set of Essential Principles (confor-
mity assessment)—such as, compliance with quality, safety and performance
principles as well as compliance with regulatory controls for manufacturing pro-
1
The information provided here is based on information available online (accessed: May 1, 2019)
https://www.tga.gov.au/publication/australian-regulatory-guidelines-medical-devices-argmd
198 C. C. Carey
Table 11.4 Australia’s medical device classification system

Classification Risk level
Class I Lowest
Class I—supplied sterile Low to medium risk
Class I—with measuring function
Class IIa
Class IIb Medium to high risk
Class III High risk
Active implantable medical devices High risk
cesses. For very low risk devices, these products can be self-certified. For all other
devices, conformity assessment can be conducted by TGA or a recognized indepen-
dent body, including notified bodies in Europe or US FDA-accredited bodies. TGA
can still choose to audit any device (application audit) as an added layer of assur-
ance that a device meets Australia’s regulatory requirement prior to marketing.
Although there is close alignment in the conformity assessment procedures between
the EU and Australia, it should be recognized that differences between Australian
Essential Principles and EU Essential Requirements exist. The upside is that with a
European CE marking, the TGA’s approval process will be easier. Australia recog-
nizes CE marking. Declarations of Conformity to the Australian Regulations
are also required in order to register with TGA.
Market authorization is by inclusion in the Australian Register of Therapeutic
Goods (ARTG). ARTG is an electronic, searchable registration system that holds
information about a product name, device classification, ‘sponsor’ and manufac-
turer details. Every ARTG entry belongs to a Sponsor. A Sponsor is a person or
company who is responsible for applying and for supplying a medical device in
Australia, as well as for maintaining the ARTG. All classes of devices must submit
a Medical Device Application (Intended purpose, Classification, and GMDN code2)
through TGA’s eBS (eBusiness Services) portal system. If TGA approves the appli-
cation, an ARTG listing number will be issued (Certificate of Inclusion). The listing
will be in the ARTG database on the TGA website and can begin marketing the
device in Australia.
More recently, TGA published a new guidance [8] on the use of Market
Authorization evidence from regulatory bodies by Australian market applicants to
abridge the TGA conformity assessment process, allowing speedier commercializa-
tion in the country. The regulatory bodies include the US Food and Drug
Administration, Health Canada, Japan’s Pharmaceutical and Medical Devices
Agency (PMDA), and Medical Device Single Audit Program (MDSAP) audit
organizations.
2
The sponsor obtains the classification and GMDN (Global Medical Device Nomenclature) codes
from the manufacturer. GMDN is a list of internationally agreed generic names/descriptors used to
identify all medical device products and managed by the GMDN Agency.
Japan
In Japan, the responsibility to regulate the sale and distribution of medical devices
to protect and promote the health of its citizens is performed by a close cooperation
and collaboration between the government and an independent administrative
agency. The Pharmaceutical Safety and Environmental Health Bureau (PSEHB)
under the Ministry of Health, Labor, and Welfare (MHLW)3 is in charge of develop-
ing regulatory policies, issuing Ministerial Ordinances, and implementing safety
standards. PSEHB is the final judge on approval decisions. On the other hand, the
Pharmaceutical Medical Devices Agency (PMDA) has been providing the technical
functions, e.g., approval reviews of medical device applications, consultations con-
cerning clinical trials, GMP inspections, monitoring of adverse events reports and
other administrative functions. The functional integration of MHLW and PMDA
allows them to handle a wider range of activities from clinical studies to approval
reviews to post-market measures throughout the device lifecycle. Japan has also
established the Registered Certification Bodies (RCB) third party program. RCBs
are certified by PMDA and can perform technical reviews for certain designated
products.
The law and regulations that govern MHLW’s authority to regulate medical
devices is written in the Pharmaceutical and Medical Devices Act (PMD Act). In
November 2014, it replaced the Japanese Pharmaceutical Affairs Law (PAL). PAL
was renamed as the Law for Ensuring Quality, Efficacy, and Safety of Drugs and
Medical Devices (commonly known as the Pharmaceutical and Medical Devices
Act [9]. Medical Device Registration is required before a medical device is placed
in the market.4
Japan has a Marketing Authorization Holder (MAH) system in which only local
companies in Japan with a valid MAH license may import and sell medical products
into the Japanese market. It is a mechanism to ensure that all imported medical
products have a local company in Japan that can take full regulatory responsibility
for the imported medical products. Thus, all foreign medical device companies who
want to sell products in Japan must first be approved as certified non-Japanese man-
ufacturers. Then, a Marketing Authorization Holder (MAH) or a designated
Table 11.5 Japan’s medical device classification system

Classification Risk level Registration type
Class I (general medical device) Extremely low risk Notification (self declaration)
Class II (controlled medical device) Low risk Certification or approval
Class III (specially controlled device) Medium/high risk Certification or approval
Class IV (specially controlled device) High risk Approval (PMDA)
3
Organization Chart, MHLW https://www.mhlw.go.jp/english/org/detail/dl/organigram.pdf
4
Be aware that the term Registration carry a different meaning in the US system. It is associated
with Establishment Registration and Device Listing; which is done only after the device has
obtained FDA market clearance or approval.
200 C. C. Carey
arketing authorization holder (DMAH) is appointed to manage their product reg-

m
istrations and interact with Japan’s regulatory authorities.
Regarding device classification, similar to the other jurisdictions discussed
above, the Japanese regulatory framework uses a risk-based classification system,
Class I, II, III and IV (Table 11.5) which aligns with the principles outlined in the
‘Principles of Medical Devices Classification’ Study Group 1, GHTF Final
Document, June 27, 2006.5 For a Class I general medical device, it does not require
an approval process; only a notification is required and the manufacturer does a self-
declaration. A Class II controlled device can be designated for RCB review or
reviewed by PMDA. A Class III and IV specially controlled device must undergo
PMDA and MHLW review for approval. However, under the new PMD Act some
Class III devices may be designated for RCB review and certification.
Consensus Standards
Good technical standards are built on globally accepted principles of consensus,

openness, balance, and transparency. The recognition and use of consensus-driven,
harmonized standards play a prominent and significant role in the regulation of
medical devices around the world. Standards should address important public health
issues. Whether the work was developed and consolidated at the national or interna-
tional level, the use of established standards provides regulatory authorities an addi-
tional tool to carry out its mission—to promote and protect public health. Recognized
published standards are used to satisfy certain premarket requirements to validate
device conformance to safety, quality and effectiveness or to demonstrate compli-
ance with Essential Principles.
Some commonly used international standards by medical device manufacturers are:
• ISO 13485—Quality management systems
• ISO 14971—Application of risk management to medical devices
• ISO 10993—Biological evaluation of medical devices
• ISO 60601—Series of technical standards for the safety and performance of
medical electrical equipment
Regulators have published guidance documents on the appropriate use of consen-
sus standards, how industry can use “Declarations of Conformity” and when addi-
tional data is needed. For example, the US FDA’s final guidance, Appropriate Use of
Voluntary Consensus Standards in Premarket Submissions for Medical Devices,
Guidance for Industry and Food and Drug Administration Staff, was issued September
14, 2018.6 The FDA also has a searchable Recognized Consensus Standards data-
5
GHTF archived documents are accessible at the IMDRF site, http://www.imdrf.org/ghtf/ghtf-
archives-sg1.asp
6
https://www.fda.gov/media/71983/download
base. These are standards recognized by the FDA, either wholly or in part.7 Health
Canada’s document, Guidance Document: Recognition and Use of Standards under
the Medical Devices Regulations, was effective on September 11, 2006 and pub-
lished by the authority of the Ministry of Health.8 Japan’s PMDA outlines the
Certification Criteria, Approval Criteria and Review Guideline for medical devices
in which they identify the use of technical standards.9 In the EU, harmonized stan-
dards on medical devices are referred to in Article 8 of Regulation (EU/MDR)
2017/745. A group of experts representing competent authorities of EU countries
was established as the Medical Device Coordination Group (the MDCG). They par-
ticipate in the development of proposals for standardization and provide assistance to
implement the regulation.10
Manufacturers should always check the government websites for updated ver-
sions and information related to standards and conformity assessments program and
policies.
Conclusion
Around the world, it is apparent that regulatory authorities in different jurisdictions

have something in common. As described in this chapter, they all apply the princi-
ples of risk-based regulatory paradigm in the management and supervision of medi-
cal devices. The methodology is grounded in risk and provides assurance that
products are safe, effective, and perform as they are intended prior to their sale and
distribution to the broader population. While this chapter focused on device classi-
fication and key aspects of regulations in bringing new devices in the marketplace,
it is important to recognize that the same risk-based concepts are applied across the
device life cycle from design to premarket to postmarket stages in order to accom-
plish the mission of promoting and protecting public health.
It is important for manufacturers to understand the differences and similarities in
the structure, regulatory approaches and regulatory pathways to market and lever-
age harmonized guidelines and policies. In all cases, the use of established industry
consensus standards and adherence to conformity assessments can meet certain
aspects of the regulatory requirements. Standards should promote and not hinder
innovation. However, they must also be periodically reviewed to stay in-step with
technological advances and real-world experience.
7
https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfStandards/search.cfm
8
https://www.canada.ca/content/dam/hc-sc/migration/hc-sc/dhp-mps/alt_formats/hpfb-dgpsa/pdf/
md-im/md_gd_standards_im_ld_normes-eng.pdf
9
http://www.std.pmda.go.jp/stdDB/index_en.html
10
https://ec.europa.eu/growth/sectors/medical-devices/regulatory-framework_en
202 C. C. Carey
References
1. ASEAN Medical Device Directive. 2015 [Online]. https://asean.org/wp-content/

uploads/2016/06/22.-September-2015-ASEAN-Medical-Device Directive.pdf.
2. Study Group 1 of the Global Harmonization Task Force (GHTF). Definition of the terms
‘Medical Device’ and ‘In Vitro Diagnostic (IVD) Medical Device’. GHTF Final Document
(revision of GHTF/SG1/N29:2005). 2012. [Online]. http://www.imdrf.org/docs/ghtf/final/sg1/
technical-docs/ghtf-sg1-n071-2012-definition-of-terms-120516.pdf.
3. U.S. Code of Federal Regulations (CFR) [Online]. https://www.accessdata.fda.gov/scripts/
cdrh/cfdocs/cfcfr/CFRSearch.cfm?CFRPartFrom=800&CFRPartTo=1299.
4. Public Law 94-295 94th Congress, Medical Device Amendments of 1976, amend the Federal,
Food, Drug, and Cosmetic Act, 28 May 1976.
5. Legislative Act, Regulation (EU) 2017/745 of the European Parliament and of the Council of
5 April 2017.
6. European Commission. Guidance document, classification of medical devices. Annex IX of
Council Directive 93/42/EEC, MEDDEV 2.4/1 rev.9, June 2010.
7. Canadian Medical Devices Regulations (CMDR) SOR/98-282, last amended March 4, 2019.
Current to March 27, 2019.
8. Minister or Health, Government of Canada. Guidance on the risk-based classification system
for non-In Vitro Diagnostic Devices (non-IVDDs). Effective date, June 12, 2015.
9. Therapeutic Goods (Medical Device Regulations 2002. Compilation No. 39), latest compila-
tion date is December 1, 2018.
10. Australian Therapeutics Goods Administration Health Safety Regulation. Use of market
authorisation evidence from comparable overseas regulators/assessment bodies for medical
devices (including IVDs). Ver 1.1, Nov 2018.
11. Regulatory Information Task Force, Japanese Pharmaceutical Manufacturers Association.
Pharmaceutical administration and regulations in Japan; 2017.
Chapter 12
Design of Clinical Trials to Validate
Cuffless Blood Pressure Monitors
Willem J. Verberk
Abstract Currently the market is flooded with cuffless blood pressure monitors.
However, these devices are not always as accurate as consumers and physicians
might expect.
Therefore, the clinical accuracy of these blood pressure monitors should be veri-
fied using a suitable standard. As the current standards for blood pressure monitors
are designed for cuff-based devices these are not suitable for verifying the accuracy
of monitors without cuffs due to essential differences between blood pressure moni-
tors with and without cuffs.
For example, almost all cuffless monitors must be calibrated, may be used as
wearables during activity and may measure blood pressure at body parts other than
upper-arm and wrist.
For this reason, there is an urgent need for a new standard to test cuffless moni-
tors that covers these different aspects. This chapter describes characteristics of
cuffless monitors, highlights the differences between cuff-based and cuffless moni-
tors, and proposes methods for testing. Additionally, attention is paid to the selec-
tion of participants in relation to good clinical practice and to statistical reporting of
the figures. Finally, some published clinical studies to the accuracy of cuffless blood
pressure monitors are discussed.
Keywords Cuffless blood pressure monitor · Cuff · Sphygmomanometer · Blood

pressure measurement · Validation protocol · Standards · Continuous blood
pressure measurement · Intermittent blood pressure measurement · Good clinical
practice
W. J. Verberk, Ph.D (*)

Microlife AG, Widnau, Switzerland
CARIM School for Cardiovascular Diseases, Maastricht University,
Maastricht, The Netherlands
e-mail: willem.verberk@microlife.ch

204 W. J. Verberk
Introduction
Blood pressure measurement is the most performed medical procedure. Usually it is

done by an upper-arm cuff inflating to supra-systolic pressure after which blood
pressure is determined by mean of Korotkoff sounds (auscultatory technique) or the
oscillometric technique. However, the cuff has numerous disadvantages. Patients
consider the cuff as inconvenient or causing pain, particularly among those with
high blood pressure. With frequent measurements such as 24-h blood pressure mea-
surement it can lead to skin irritation, bruising and sleep disturbance. In addition,
cuff pressure increases both systolic and diastolic blood pressure by approximately
7 and 5 mmHg, respectively [1]. Another serious challenge in clinical practice, can
be the finding of a cuff that fits correctly for some patients. This is not only related
to the size of the arm but also to deviating arm shapes (e.g., conical shape) and
composition of tissue (fat, muscles) can lead to incorrect blood pressure measure-
ment [2]. Finally, cuff blood pressure measurement on the upper arm only allows
intermittent blood pressure measurement and is therefore not capable of measuring
sudden changes in blood pressure which makes it unsuitable for use in acute care.
The above disadvantages related to upper-arm cuffs have encouraged many man-
ufacturers and researchers to develop noninvasive (continuous or intermittent) cuff-
less blood pressure monitors. Thus far, most of these cuffless monitors estimate
blood pressure transcutaneously using the principal of pulse transit time (PTT) or
pulse arrival time (PAT) [3]. This is based on the time it takes for a pulse wave to
travel through the artery from one place to another at a fixed distance apart. When
knowing the time delay and the distance between the two pulsatile arterial hemody-
namic related signals the pulse wave velocity (PWV) can be calculated [4]. With the
PWV the blood pressure value can be determined.
Cuff-less devices have high potential because the sensors can be easily imple-
mented in so-called wearables (watch, shirt, patch, etc.) that makes it suitable dur-
ing activity (e.g., 24-h ambulatory blood pressure measurement and fitness purpose).
In addition, the technique can be used in combination with, or implemented in
smartphones which enables easy blood pressure measurement to a wide population.
This has all led and will lead to numerous cuff-less blood pressure monitors enter-
ing the market. Obviously, not all these devices are as accurate as consumers and
physicians might expect. Because the devices are relatively new and the process of
making standards is time consuming, there is currently no internationally accepted
standard to verify the accuracy of these devices. However, a standard is urgently
needed for both clinicians and consumers who are overloaded with advertisements
for “quick and easy blood pressure measurement.”
In 2014, The Institute of Electrical and Electronics Engineers (IEEE) presented
a standard entitled “IEEE Standard for Wearable, Cuff-less Blood Pressure
Measuring Devices” [5]. However, this standard does not seem to cover all aspects
of cuffless blood pressure measurement, especially regarding validation of continu-
ous cuffless blood pressure monitors. Currently, the ANSI/AAMI/ISO Joint Working
Group is developing a new standard (ISO/CD 81060-3) for clinical investigation of
12 Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors 205
continuous automated measurement with noninvasive sphygmomanometers. This

chapter is aimed at highlighting certain aspects that need to be considered for
designing a clinical trial for validation of cuff-less blood pressure monitors.
uff or Cuffless Devices and Continuous or Intermittent

C
Devices
Blood pressure monitors can be distinguished according to two different classifica-

tions; cuff versus cuff-less devices and continuous compared to intermittent devices
as explained in Table 12.1. Although, the name “cuff-based monitors” seems self-
explaining, there is sometimes confusion about the definition of a cuff. A cuff is a
part of the blood pressure monitor that is wrapped around a limb (thus not necessar-
ily an arm) of the subject. It has a bladder that can be inflated to partially or fully
occlude the artery after which it is deflated. A cuffless blood pressure monitor gen-
erally works with a sensor (electrocardiographic, photoplethysmographic, and/or
phonocardiographic) that allows the assessment of the pulsation of arterioles (heart
beat) at different skin depths. The fact that a cuff-less blood pressure monitor
enables continuous blood pressure measurement does not necessarily mean that it is
a continuous blood pressure monitor. A continuous noninvasive automated sphyg-
momanometer is, according to the definition, “a device able to estimate blood pres-
sure from the pulse wave of each heart cycle without arterial puncture.” However,
the device does not need to display the blood pressure values per heartbeat (continu-
ously) but may display the blood pressure value averaged from multiple heart beats
instead. This means that cuffless monitors may provide intermittent blood pressure
values whereas on the other side cuff-based blood pressure monitors exist that mea-
sure blood pressure continuously. For instance, a device that measures finger arterial
pressure continuously using a finger cuff and infrared plethysmograph [6].
Table 12.1 Cuff or cuffless devices versus continuous or intermittent blood pressure monitors
Continuousa Intermittent
Cuff- Determines and presents blood Determines and presents blood pressure over
based pressure per heartbeat using a cuff multiple heartbeats
For example, by using a finger cuff For example, by using automated oscillometric
and infrared plethysmograph measurements
measurements
Cuff- Determines and presents blood Determines blood pressure per heart beat
less pressure per heartbeat without using (continuous) but presents blood pressure values
a cuff averaged over multiple heartbeats
For example, by mean of pulse For example, by mean of pulse wave velocity
wave velocity measurements measurements
Suitable for verifying sudden changes in blood pressure
a
206 W. J. Verberk
equired Tests for Validation of Cuffless Blood Pressure

R
Monitors
Current ANSI/AAMI/ISO standards are designed for testing the accuracy of auto-
mated cuff-based blood pressure monitors. However, cuffless devices essentially
differ from cuff-based automated oscillometric sphygmomanometers in design and
often also in intended purpose so current standards are not suitable to test cuffless
blood pressure monitors (Table 12.2). The following tests are required for the vali-
dation of a cuffless blood pressure monitor;
Static Test
Automated oscillometric devices provide intermittent blood pressure values in a static

condition. The validation protocol, therefore, is a relatively simple and static test veri-
fying the sphygmomanometer under testing against a reference upper-arm cuff or inva-
sive blood pressure monitor in a similar condition and environment [7, 8]. Reference
and test blood pressure measurement can be done either sequentially or simultane-
ously. The accuracy of intermittent cuffless blood pressure monitors, designed for static
measurement, can be tested almost in a similar way. However, an important difference
with cuff-based upper-arm devices is that almost all cuffless blood pressure monitors
need to be calibrated first. This process of determining subject specific parameters is
needed for estimating blood pressure with the cuffless monitor. It affects overall perfor-
mance of the device and therefore needs to be covered in the new standards.
Calibration
Most cuffless monitors estimate blood pressure from PTT obtained PWV using a
mathematical formula [9]. As the relationship of PTT to blood pressure (the calibra-
tion curve) depends on multiple characteristics such as arterial stiffness and body
Table 12.2 Standards for noninvasive blood pressure monitors

ANSI/AAMI/
ANSI/AAMI/ ISO 81060-2, ANSI/AAMI/ISO
Standard ISO 81060-1 BHS, IP 81060-3a IEEE cuffless wearable
Device to Non-automated Intermittent continuous blood Wearable cuffless blood
be tested (aneroid and automated pressure monitors pressure monitors
mercury blood blood pressure (can also provide (seems not suitable for
pressure monitors intermittent blood devices that show beat
monitors) pressure values) to beat blood pressure
values)
Standard under development
a
length it is highly variable from person to person. However, also within a person this
relationship fluctuates due to, for example, changes in heart rate, vasomotion, and
hydrostatic pressure [10]. Because many arterial and human factors affect the relation-
ship between PWV and blood pressure, calibration of the cuffless device seems indis-
pensable. Some investigators claim that there is no need for calibration when using
elastic arteries to measure blood pressure such as the aorta [11], but this needs further
investigation. Until then, calibration (or initialization) with a cuff-based upper arm or
invasive blood pressure monitor must be done for obtaining subject specific parame-
ters that are needed for estimating blood pressure using a cuffless monitor.
Stability Test
Currently, some cuffless blood pressure devices have been validated according to the
standards of ANSI/AAMI/ISO 81060-2:2013 [12] or the International Protocol of the
ESH 2010 [13]. For these validation studies the device under testing was calibrated
shortly before the test measurement was taken. A cuffless device generally appears to
be accurate for only a short time after calibration because there is no time for damping
or drifting and thus the blood pressure is almost similar to the moment of calibration.
However, the accuracy of the cuffless device generally deteriorates with blood pressure
changes and/or with time after the initial calibration so that the accuracy over time must
be investigated. Obviously, cuffless blood pressure measurement is useless if it requires
calibration with an upper-arm cuff monitor before every cuffless measurement.
Therefore, the manufacturer must disclose the maximum timeframe in which the
device is accurate before it needs recalibration. This means a “stability test” must be
done to verify accuracy over a longer period in which recalibration is not needed. The
stability of the displayed blood pressure values must be investigated during the whole
maximum period specified by the manufacturer. The IEEE standard suggests perform-
ing two test measurements within two calibration points; directly after initial calibra-
tion and before the next calibration is required. However, when performing only an
initial and end-testing the device might not be accurate in-between these tests.
Therefore, devices should be tested daily or even hourly depending on its disclosed
accuracy period. If the disclosed timeframe covers a long-time period, a higher test-
frequency in the beginning followed by less frequent testing towards the end should be
recommended. For instance, if the disclosed period is 1 month, twice-daily measure-
ments may be assessed for the first 3 days thereafter daily measurements for 10 days
followed by weekly performances until the end.
Test with Blood Pressure Change
Many cuffless devices are wearable and measure blood pressure beat by beat that
allows tracking acute blood pressure changes over time. This is a major advantage
because blood pressure variability is an important cardiovascular risk predictor
208 W. J. Verberk
[14]. Sometimes it is essential to verify acute changes in blood pressure such as in

the operation theater or acute care. However, this advantage of cuffless monitors
also adds an extra difficulty as its capability to track (quick) blood pressure changes
must be tested for accuracy.
Noninvasive Blood Pressure Change Test
The IEEE standard considers only auscultatory blood pressure measurement as a

reference device. Obviously, this method is unsuitable for testing quick blood pres-
sure changes. The performance of three sequential auscultatory blood pressure mea-
surements with 1-min interval time takes approximately 5 min. Nevertheless, the
auscultatory method may be used to test slow blood pressure changes (e.g., over
15 min) in intermittent cuffless monitors. In practice this procedure might not be
easy because the changed blood pressure value must be relatively stable for more
than 5 min.
Invasive Blood Pressure Change Test
The capability for the cuffless device to follow quick blood pressure changes can
only be verified by comparing to simultaneously performed invasive blood pressure
measurement. Blood pressure changes to be verified must be significant (15–
30 mmHg) with a defined minimum number of both changes and patients. This
means that if not enough blood pressure changes can be obtained, more patients
must be included. It may be recommended to present the number of blood pressure
changes in a histogram as suggested in the IEEE standards (Fig. 12.1).
Time Duration of Measurement Procedure
For intermittent cuffless blood pressure monitors, it is important that the manufac-
turer discloses the number of heart beats from which the blood pressure value is
averaged. Standards should provide a restriction to the time duration cuffless moni-
tors may take for assessing blood pressure. This time should not extend the time it
takes for an upper-arm cuff monitor to measure blood pressure (e.g., 30 s). The
lowest number of cardiac cycles from which blood pressure is calculated should be
used for the validation study. There are two reasons for this: First, it makes the vali-
dation procedure shorter and easier to perform, second the 30 s may be indicated as
a sort of “worst case scenario.” If the short period of the device under testing appears
accurate, one may expect that the average taken over a longer period with more
heart beats is more accurate. The latter also indicates that the allowance of a larger
time frame might make it too easy to pass the standards.
Fig. 12.1 Histogram of blood pressure changes
For continuous cuffless blood pressure measurement validation, a clear time

frame should be defined. It may seem logical to use approximately 5 min as this is
the same time that is normally taken for intermittent blood pressure comparison.
This will lead to approximately 300 measurements (5 × 60 heart beats) to
compare.
Sequential or Simultaneous Comparison
The accuracy of intermittent cuffless blood pressure monitors can be verified with
either sequential or simultaneous blood pressure measurement.
Sequential
Sequential blood pressure measurement only seems plausible for auscultatory mea-
surement reference of intermittent (limb) monitors and therefore the existing stan-
dard [8] may be followed; Triplicate test device measurements should follow
reference measurements. At least 30 s should be allowed after reference measure-
ment to avoid venous congestion, but not more than 60 s to reduce differences due
to blood pressure variability. Differences are calculated from the test device mea-
surement and the average of the adjacent two reference readings. Preferably, the
same limb should be used to avoid lateral blood pressure differences.
210 W. J. Verberk
Simultaneous
For simultaneous comparison using an auscultatory reference the test device should
start with the cuff deflation of the reference sphygmomanometer. If both the refer-
ence blood pressure monitor and test device are (upper) arm devices and compared
with simultaneous measurement, then the inter-arm blood pressure difference must
be verified with simultaneous measurements [15]. If the difference exceeds a certain
threshold (e.g., 5 mmHg) then the patient should be excluded from testing. It could
also be decided to compensate for the differences. Then the reference and test
devices should be changed between arms and the difference related to inter-arm dif-
ference can be subtracted [7]. For verifying a longer time-period the usefulness of
compensating for inter-arm blood pressure difference seems not a good idea as
reproducibility of (nonpathological) inter-arm difference in the long-term may be
questioned [16].
For testing a cuffless intermittent device using invasive blood pressure measure-
ment, the average blood pressure value should be calculated from the same number
of heartbeats for both the reference and test method. Obviously, cuffless continuous
blood pressure measurement must be validated with simultaneous invasive
measurements.
Blood Pressure Measurement at Different Parts of the Body
Wearable cuffless devices measure blood pressure at different sites of the body.
Therefore, the reference measurement site should be disclosed by the manufacturer.
As different sites of the body have different blood pressure values the test measure-
ment site should be as close as possible to the reference site. In addition, new stan-
dards must define accepted reference sites, or state that any site may be used as a
reference if appropriately measured. Current standards state that the same limb, the
aorta and subclavian or femoral arteries are appropriate references for simultaneous
validation [7].
Statistical Analysis and Reporting
Published clinical studies to the accuracy of cuffless devices show a wide heteroge-
neity in design (sample size, number of measurements, reference device), statistical
analysis and reporting of data. This makes it difficult to interpret the outcome. A
universal protocol with proper design, described statistics and graphs can help to
understand the outcome and makes it easy to compare studies. In this section, a list
of important variables to report is provided.
Population Sample
Blood Pressure Ranges
Most studies that investigated the accuracy of cuffless blood pressure measurement
thus far, used a limited number of (healthy) patients who are within an average
blood pressure range. However, Ding et al. showed that blood pressure from a nor-
motensive group agreed better with the reference blood pressure than those of the
hypertensive group [17]. This underscores the importance of selecting patients with
a wide blood pressure range distribution as is required in current standards.
Gender Distribution
Almost all validation guidelines require an equal distribution for women and men
(e.g., IEEE n = 20/20; ANSI/AAMI/ISO > 30%) despite the lack of convincing
evidence that gender affects blood pressure measurement. However, gender appears
to be of significant influence on vessel composition [18, 19]. Because most cuffless
blood pressure monitors directly depend on PWV an equal distribution between
men and women seems essential for clinical investigation of measurement accuracy.
For invasive blood pressure measurement, it may be difficult to find a perfect distri-
bution. Therefore, the current requirements of a least 30% of either gender in the
total population seems appropriate.
Sample Size and Measurement Frequency
Sample size determination is essential in the design of clinical trials; less patients
reduce power, but more patients may cause unnecessary inconvenience. The latter
applies certainly for invasive blood pressure measurement but also for the duration
of the procedure required for testing noninvasive monitors as this can take several
days to weeks.
Sample Size for Noninvasive Testing
The ANSI/AAMI/ISO standard requires a minimum of 85 participants and a mini-

mum of three paired tests (test and reference). This number was based on what was
considered an acceptable difference between the average reference and test blood
pressure value, which is 5 mmHg and an acceptable standard deviation of 8 mmHg.
The statistical approach was a t-test with a 5% chance that a proper device was
rejected (α = 0.05) and a 2% chance that an inaccurate device would be approved
2% (β = 0.02) [20]. See Formula 12.1.
212 W. J. Verberk
Formula 12.1
Subjects calculation for ANSI/AAMI/ISO standards:
2
  σ2
n = 2  z α + z1− β  2 .
 1− 2  δ
(n = sample size Ζ = statistic for normal distribution σ = Sample standard deviation;

δ = mean difference)
2 (8 )
2
With the in the text described values this results in: 2 (1.96 + 2.06 ) = 83
( 5)
2
subjects
The IEEE standard proposes 45 subjects based on an analysis by Yan et al. [21]
who claim that the ANSI/AAMI/ISO sample size calculation is based on the errone-
ous assumption that the error rate of the 85 included subjects have a normal distribu-
tion. According to the authors, blood pressure variability is related to the height of
blood pressure values which is not covered by the included three blood pressure
ranges. Therefore, the authors consider a t-distribution (four degrees of freedom
[t4]) to be a better model than the normal distribution on prediction of the errors for
most blood pressure monitors [22].
Sample Size for Invasive Testing
According to the guidelines for good clinical practice, one should not perform inva-
sive blood pressure measurement for the sole purpose of a validation study [23].
Therefore, it may be difficult to find patients so that less patients (15 instead of 85
as in ANSI/AAMI/ISO) and less strict selection criteria should be allowed. For
example, there are no age range requirements for intra-arterial verification (as in the
first edition of the AAMI standard). This all leads to the fact that the range of
observed blood pressures can be less evenly distributed. The lower number of
patients can be partly compensated by increasing the measurement procedure fre-
quency (e.g., 10 instead of 3 performances). This in total does not lead to the same
measurement number as for noninvasive comparison (n = 85 × 3 measurements) but
because invasive blood pressure measurement is performed simultaneously, differ-
ences because of blood pressure variability are low.
Special Subject Population
If a cuffless blood pressure monitor is intended for use in a special-subject popula-

tion (e.g., children, pregnant women, end stage renal disease patients) a special vali-
dation is needed. As physiological characteristics (vessel composition, respiration
rate, heart rate etc.) of these patients may influence the accuracy of the device.
However, if the device is already validated in a general population there is no need
to investigate the entire number of subjects again. Then an additional 35 (with
auscultatory reference) or 5 subjects (with invasive reference) as recommended in

current standards should suffice [7].
Performance Criteria
Mean Difference and Standard Deviation
The simplest and most commonly used method to present differences between refer-
ence and test blood pressure measurement is by calculation of the mean difference and
standard deviation. The current standard for noninvasive validation of automated
sphygmomanometers prescribes the mean value and standard deviation of the differ-
ences from the average of three test determinations and three reference determinations
[7]. This standard has defined an acceptable difference from “the gold standard”
(Mercury) of up to 5 mmHg. Due to the variation in blood pressure it is also agreed
that a single deviation of more than 5 mmHg should not immediately lead to rejection
of a device. Therefore, a standard deviation of 8 mmHg from at least 255 measure-
ments is allowed. However, the criteria of 5 ± 8 mmHg could still lead to a high num-
ber of subjects that fall outside the acceptable rate. Therefore, the standard added a
second criteria; the mean error of the average measurements from at least 85 subjects
should be 5 mmHg with acceptable SD-values (from 4.79 to 6.95) that decrease as the
mean errors increase. This schedule must ensure that at least 85% of all subjects fall
within the 10-mmHg error range (and approximately 50% of subjects have errors that
fall within the commonly accepted error of 5 mmHg).
The acceptable differences of 5 ± 8 mmHg between the reference and test device
were determined a few decades ago. It may be expected that at that time the quality
of the automated blood pressure monitors were lower than is currently the case.
Nowadays, it may be questioned if we should continue accepting this error or per-
haps it is time to lower the acceptance threshold considering the method of valida-
tion. However, if a device is manually tested by two observers in sequential order, it
might not be possible to decrease this threshold due to inter-observer bias and blood
pressure variability. On the other hand, use of current developed techniques that
support observers with their readings might decrease observer bias [24].
Corrected Standard Deviation for Repeated Measurements
In continuous blood pressure measurement validation, many more readings are

compared to the reference than with intermittent measurement. Whereas the latter
requires 3 or 10 measurement procedures, simultaneous continuous measurement
over a period of 5 min leads to approximately 300 measurement comparisons for
one subject. This means that many pairs of measurements are made in relatively few
patients, which needs to be considered in the statistical analysis. For this Bland and
Altman suggested a corrected experimental standard deviation. This considers both
214 W. J. Verberk
the within and between subject variance. The measure of correction correlates to the
ratio number of measurements per subjects to number of study participants [25].
Following this, the previously defined acceptable error of 5 mmHg and standard
deviation of 8 mmHg might not be acceptable anymore. Therefore, the accepted
differences may need to be reconsidered based on current clinical evidence.
Error-Bands BHS and ESH Protocols
Instead of using the mean and standard deviation some standards use so-called error
bands. The mean error of the subjects from ANSI/AAMI/ISO is in line with the
BHS-protocol that requires that 60% of the cumulative readings fall within 5 mmHg,
85% within 10 mmHg and 95% within 15 mmHg [26]. Later, the BHS protocol was
replaced by the logistically easier ESH protocol that requires 33 instead of 85 sub-
jects. The ESH protocol demands 65% within 5 mmHg, 81% within 10 mmHg and
96% within 15 mmHg [27]. Recently, it has been decided to discontinue the ESH
protocol because of concerns regarding its statistical power [20] and to design a new
joint protocol of ANSI/AAMI/ISO and ESH [28]. The new “universal protocol”
mostly resembles the ANSI/AAMI/ISO protocol which indicates that mean and
standard deviations is a preferred outcome above error-bands.
Absolute Differences
Mean absolute differences (MAD) and mean absolute percentage difference (MAPD)
is suggested in the IEEE standard. Because differences may also relate to blood pres-
sure levels, the accuracy is also calculated within the four different blood pressure
classifications they propose. MAD at different blood pressure classification categories
should be within 6 mmHg for systolic and diastolic measurement analyzed separately.
According to the designers of the IEEE standard MAD has the advantage that the
standard deviation is not needed as the absolute values also consider the spread and
the range. As compared to standard deviation MAD does not put so much weight on
outliers which relevance could be discussed (a coincidental error or important mea-
sure). Yan et al. showed that a MAD of approximately 5–6 is considered acceptable as
this comes close to the criteria of ANSI/AAMI/ISO and BHS [29].
Statistical Figures
Bland–Altman Plot and Histogram of Differences
For comparing the agreement between two types of measurements the Bland–
Altman plot (Fig. 12.2) is often used [30]. This plot, that shows the 95% limits of
agreement between two methods, was introduced in 1986 to replace the, until then
Fig. 12.2 Bland–Altman plot; BPref indicates blood pressure values of the reference technology;
BPDUT blood pressure values of the Device Under Testing
most used outcome, correlation coefficient. The latter was considered misleading
because a high correlation does not necessarily imply a good agreement between
two methods. The Bland–Altman plot shows the difference of two paired measure-
ments against the mean of these two measurements. It depends on the assumptions
that the mean and standard deviation of the differences are constant throughout the
range of measurements, and that these differences have an approximately normal
distribution [31]. To check the latter a histogram of the differences is proposed as
demonstrated in Fig. 12.3. If the histogram is skewed the Bland–Altman plot may
lead to misinterpretation. The 95% limits of agreement are shown as the mean of the
two values ±1.96 standard deviations. These limits are expected to cover the general
difference between the method measurements for 95% of pairs.
Four Quadrant Plot (for Blood Pressure Changes)
To visualize how the device under testing behaves in blood pressure changes
(∆PDUT) as compared to the blood pressure changes of the reference (∆PREF) device
a Four-quadrant plot (FQP) may be helpful (Fig. 12.4). When both the studied tech-
nology and the reference technology indicate an increase in blood pressure, the data
points will appear in the upper right quadrant of the FQP. Similarly, the lower left
quadrant contains data points resulting from decreases in blood pressure for both
the test and reference device. Therefore, the upper right and the lower left quadrants
of the FQP represent concordant measurements of the studied and reference tech-
nology regarding direction of changes. The more dots in these areas the more con-
cordant a device is. However, this does not necessarily mean that the device is
216 W. J. Verberk
Fig. 12.3 Histogram of differences between reference device and the device under testing
Fig. 12.4 Four quadrant plot of the continuous noninvasive sphygmomanometer versus the refer-
ence sphygmomanometer. The values on the X-axis refer to ∆P values of the reference technology
(REF), whereas the y-axis refers to the ∆ values of the Device Under Testing (DUT)
accurate. For accuracy the points should be located close to the 45° diagonal (Unity
line). In addition, the acceptable maximum error could be defined in advance and
presented as lines (PERRORmax) to either site of the 45° diagonal of the concordant
quadrants between which the data should fall [32].
xecute and Monitor the Measurement Campaign; Good

E
Clinical Practice
For the performance of a validation study it is of essential importance that the stan-
dards for Good Clinical Practice (GCP) for the clinical investigation of medical
devices for human subjects [23] are followed and the study is approved by the local
Medical Ethical Committee. For this some items, in particular, apply to cuffless
blood pressure device validation and therefore deserve extra attention.
Invasive Blood Pressure
Regarding the performance of an invasive blood pressure validation the ANSI/

AAMI/ISO standard states that invasive blood pressure measurement may not be
assessed just for the purpose of validating a blood pressure monitor [7]. The studies
should be conducted on clinical patients in whom an intra-arterial line has already
been placed for reasons other than sphygmomanometer verification. This could
include patients undergoing carotid arteriography or cardiac catheterization, or
patients undergoing clinical research studies approved by an institutional review
board that involve intra-arterial blood pressure monitoring [33].
Duration of the Blood Pressure Measurement Procedure
It should be predetermined how long a person will be measured, and this period
should not be longer than strictly needed to ensure the convenience and/or safety of
the study participant. Furthermore, all requirements according to GCP and the dec-
laration of Helsinki should be respected, such as obtaining the informed consent
from all participants and all adverse events must be reported.
Reference Device
According to ANSI/AAMI/ISO standard a reference measurement can be per-

formed auscultatory (both aneroid and mercury) and invasively [7, 33]. The IEEE
standard only considers auscultatory blood pressure measurement with a mercury
218 W. J. Verberk
device as a reference [34]. Auscultatory reference measurement should be provided

by two trained observers and measured simultaneously with one reference sphyg-
momanometer (using a “Y” connector). Invasive blood pressure measurement shall
comply with the IEC standards [35], the maximum allowable error is ±2 mmHg and
the intra-arterial transducer must be kept at the level of the heart.
Observer Measurement
For the noninvasive auscultatory evaluation of the cuffless blood pressure measure-
ment two well-trained observers are needed. Measurements must be determined
using the first and fifth Korotkoff sound for systolic and diastolic blood pressure,
respectively. All measurements should be recorded to the nearest 2 mmHg. If mea-
surements from the two observers are no more than 4 mmHg apart, the mean value
of the two is used as the reference measurement. Otherwise, the measurement
should be taken again.
Test Environment
A validation study should preferably be performed in a recognized center. The test

environment room should have a good climate (room temperature) and should be
quiet (no disturbing noises). Finally, the test device must be used according to the
manufacturer’s instruction.
Phase Yes or No?
Although the ANSI/AAMI/ISO standard does not suggest using phases/interim

analyses, it deserves consideration from an ethical perspective. Most cuffless
blood pressure monitors are at an initial stage of development and tested for the
first time on real subjects. Having a phase 1 may prevent unnecessary inconve-
nience to the participants and wasting time and money. Phase 1 should be consid-
ered as an interim analysis, as is performed in many studies. If the device fails at
that stage, it is useless to continue. For IEEE in Phase 1, only 20 subjects are
recruited for which the MAD of the collected data must be within 7 mmHg for
both systolic and diastolic pressure. If the MAD is higher than 7 mmHg further
testing is not needed anymore. If lower, then Phase 2 can start in which another
25 subjects are recruited.
ublished Studies to Investigate the Accuracy of Cuffless

P
Thus far, several studies have aimed to “validate” cuffless blood pressure monitors
in real patients. However, these accuracy studies contain flaws, did not pass the tests
of accuracy or used inappropriate statistics. Therefore, it seems that currently, none
of the cuffless devices can be considered accurate for their intended purpose. Some
examples are discussed below.
Reference Devices Used
Most cuffless continuous blood pressure monitors were not tested against a proper
reference measurement but against another noninvasive automated device that mea-
sured blood pressure continuously (e.g., a Finapres (Medical Systems, the
Netherlands) device) [17, 36]. This device provides beat to beat measurements by
means of finger-cuff blood pressure measurement [37]. The fact that a device has
been validated against invasive blood pressure does not make it suitable as a refer-
ence. The device under testing may have an acceptable difference with the Finapres
but not with invasive blood pressure measurement due to the difference between the
latter two. Studies to the accuracy of intermittent cuffless blood pressure measure-
ment often compared it against another cuff-based oscillometric blood pressure
monitor [11, 38–41]. Although it is practically more convenient than auscultatory
blood pressure measurement, it is not a reliable reference.
Static, Stability and Blood Pressure Change Tests
The few studies that performed a validation study according to one of the existing
validation protocols [12, 13] must be questioned for its usefulness in clinical prac-
tice as the devices were validated directly after calibration and no stability test was
performed. Perhaps a minimal time delay of 1 h or so before starting the validation
procedure may be suggested to exclude unstable cuffless devices. Multiple other
studies also presented static tests only (in either supine [39, 41] or sitting position
[40–42]). Dependent on the intended purpose of a wearable cuffless device this may
not be enough. Obviously, it is easier to determine blood pressure accurately if the
patient is not moving as there is no chance of movement artifacts and auscultatory
reference blood pressure measurement during movement is nearly impossible.
However, a cuffless wrist monitor that provides continuous measurement, may need
to be tested in different arm positions and walking with a normal arm swing [43].
The IEEE standard recommends that blood pressure should be measured in “chang-
ing conditions.” In studies to the accuracy of cuffless monitors with changing blood
220 W. J. Verberk
pressure conditions several methods were used to induce blood pressure changes,
such as: holding breath, mental stress test, pedalling while sitting on a chair [38] or
on a bike [44], rope skipping [45], simple leg-stretching [46], the use of intra-arterial
nitroglycerin [47] and handgrip exercise [48]. Although, this all undoubtedly leads
to significant blood pressure changes, it also causes large blood pressure fluctua-
tions. It, therefore, remains uncertain if these changes can be accurately verified
using an auscultatory blood pressure measurement that takes approximately 30 s.
Schoot et al. [41] verified a cuffless monitor with a modified International Protocol
[27]. The interesting part of this study was that the cuffless device was verified with
the subjects in both supine and sitting position, which leads to different blood pres-
sure values [49]. First, the calibration and test measurements were performed in
supine position, thereafter subjects were also measured in sitting position. As may
be expected, measurements taken in the “calibration-position” led to smaller errors
than measurements obtained in the sitting position. However, in both positions the
device failed to pass the defined criteria for accuracy.
Subjects Investigated
Most of the currently performed studies to the accuracy of cuffless blood pressure
monitors showed limitations regarding the population sample. Often the selected
number of patients was (too) low and blood pressure ranges were not considered.
That the latter is important for clinical testing was shown in the study of Ding et al.
where the cuffless blood pressure monitor showed higher errors in the hypertensive
than in normotensive subjects [17].
Statistical Outcomes of Performed Validations Studies
Generally, studies to cuffless blood pressure measurement accuracy provided no or

minor information of how the average blood pressure value of the cuffless device
under testing was calculated, that is, the number of measured heart beats used and/
or the duration of measurement. An exception is the study of Ding et al. who com-
pared 5024 beats from 33 subjects obtained in sitting and supine position [17]. The
authors were also one of the few who compared the test methods over an extended
calibration interval of 24 h. This showed that a longer duration after calibration led
to higher errors.
Obviously, new studies to cuffless blood pressure monitors follow, more or less,
the existing standards for cuff-based blood pressure monitors resulting in mean and
standard deviation and correlation coefficients as the preferred outcome measures.
For graphic presentation, Bland–Altman and correlation plots were often used.
Studies that performed different statistics used, for example, repeatability of mea-
sured variables over continuous cardiac cycles presented as the ratio of standard
deviation to mean value of a number of cycles expressed in percentage [42]. The

feasibility for hypertension screening (yes or no) was investigated using receiver
operating characteristic curves [42]. A t-test was performed to determine significant
difference between reference blood pressure device and the device under testing
[50]. Another study presented the results following different standards: mean abso-
lute difference (IEEE), mean error and standard deviation of the error (ANSI/
AAMI/ISO), and the cumulative percentage of readings falling within 5, 10, and
15 mmHg (BHS) [51]. Finally, Sola et al. investigated the capability of a cuffless
monitor to track continuously blood pressure changes against invasive blood pres-
sure measurement using the four-quadrant plot for presenting the results [52].
Calibration Free Cuffless Blood Pressure Measurement
Matsumara et al. claimed to have developed a cuffless blood pressure measurement

method that requires no calibration. The device works with a smartphone and a
traditional finger photo plethysmograph [53]. According to the authors the calibra-
tion is not needed because the used method is based on heart rate and modified
normalized pulse volume instead of pulse wave velocity. Because the device is
meant as a replacement of the intermittent cuff-based monitor it might have been
validated according to the current ANSI/AAMI/ISO standard, but the authors fol-
lowed their own study design and expressed the outcome as a correlation coefficient
larger than 0.7 compared to a standard brachial cuff sphygmomanometer.
Summary and Conclusion
The quick development of cuffless blood pressure monitors and the published stud-
ies to their accuracy suggest an urgent need for a new widely accepted standard to
guarantee the quality of these devices. The existing standards are insufficient, and
the lack of new standards has led to large heterogeneity in clinical tests. A new stan-
dard should consider the following items in particular:
1. Details of how the cuffless device calculates blood pressure should be

provided.
2. A stability test should be performed to verify the time at which a cuffless blood
pressure monitor remains accurate after calibration (until recalibration).
3. Cuffless monitors intended for medical purpose must also be tested for their
capability to track (quick) blood pressure changes. This means that a static test is
not enough.
4. Continuous cuffless devices that are meant to track quick blood pressure changes
can only be validated with simultaneous invasive blood pressure measurement.
5. Acceptable errors from the reference devices may need to be redefined considering
more measurement comparisons and other circumstances (blood pressure changes).
222 W. J. Verberk
6. Regarding the population sample: a minimal number of patients is needed to

guarantee enough power. The population needs to be balanced for gender, age
and blood pressure level.
7. Results should be presented as mean and standard deviation and the results
should be shown in Bland–Altman and/or 4-quadrant plots if the device is tested
for tracking blood pressure changes. These plots can reveal correlation and
agreement but also show the bias between the devices.
8. The standards for Good Clinical Practice should be followed. Unnecessary sub-
ject measurements should always be prevented.
Until devices are appropriately tested for their intended purpose, considering the
above points, continuous or intermittent cuffless blood pressure monitors should not
be used for healthcare purposes.
References
1. Gazzola K, Honingh M, Truijen J, Zuliani G, Van Den Born BH. Effect of cuff inflation on
blood pressure during self-measurement. J Hypertens. 2018;36(9):1798–802.
2. Palatini P, Benetti E, Fania C, Saladini F. Only troncoconical cuffs can provide accurate blood
pressure measurements in people with severe obesity. J Hypertens. 2019;37(1):37–41.
3. Zhang G, Gao M, Xu D, Olivier NB, Mukkamala R. Pulse arrival time is not an adequate
surrogate for pulse transit time as a marker of blood pressure. J Appl Physiol (1985).
2011;111(6):1681–6.
4. Bramwell JC, Hill AV. The velocity of the pulse wave in man. Proc R Soc B Biol Sci.
1922;93(652):298–306.
5. IEEE standard for wearable cuffless blood pressure measuring devices. IEEE std 1708-2014;
2014. p. 1–38.
6. Chin KY, Panerai RB. Comparative study of Finapres devices. Blood Press Monit.
2012;17(4):171–8.
7. Non-invasive sphygmomanometers — Part 2: clinical investigation of the intermittent auto-
mated measurement type. ISO/CD 81060-2:2013(E); 2013.
8. AAMI. Association for the Advancement of Medical Instrumentation; non-invasive sphygmo-
manometers -part 2: clinical investigation of automated measurement type, ANSI/AAMI/ISO
81060-2013; 2013.
9. Wang R, Jia W, Mao Z-H, Sclabassi RJ, Sun M. Cuff-free blood pressure estimation using
pulse transit time and heart rate. In: International conference on signal processing proceedings
international conference on signal processing 2014; 2014. p. 115–8.
10. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio A. Cuffless estimation of blood pres-
sure: importance of variability in blood pressure dependence of arterial stiffness across indi-
viduals and measurement sites. IEEE Trans Biomed Eng. 2018;65(11):2377–83.
11. Pm N, Joseph J, Karthik S, Sivaprakasam M, Chenniappan M. Bi-modal arterial compli-
ance probe for calibration-free cuffless blood pressure estimation. IEEE Trans Biomed Eng.
2018;65(11):2392–404.
12. Boubouchairopoulou N, Kollias A, Chiu B, Chen B, Lagou S, Anestis P, et al. A novel cuffless
device for self-measurement of blood pressure: concept, performance and clinical validation.
J Hum Hypertens. 2017;31(7):479–82.
13. Bilo G, Zorzi C, Ochoa Munera JE, Torlasco C, Giuli V, Parati G. Validation of the Somnotouch-
NIBP noninvasive continuous blood pressure monitor according to the European Society of
Hypertension International Protocol revision 2010. Blood Press Monit. 2015;20(5):291–4.
14. Webb AJS, Mazzucco S, Li L, Rothwell PM. Prognostic significance of blood pressure
variability on beat-to-beat monitoring after transient ischemic attack and stroke. Stroke.
2018;49(1):62–7.
15. Verberk WJ, Kessels AG, Thien T. Blood pressure measurement method and inter-arm differ-
ences: a meta-analysis. Am J Hypertens. 2011;24(11):1201–8.
16. Grossman A, Weiss A, Beloosesky Y, Morag-Koren N, Green H, Grossman E. Inter-arm
blood pressure difference in hospitalized elderly patients--is it consistent? J Clin Hypertens
(Greenwich). 2014;16(7):518–23.
17. Ding X, Yan BP, Zhang YT, Liu J, Zhao N, Tsang HK. Pulse transit time based continuous
cuffless blood pressure estimation: a new extension and a comprehensive evaluation. Sci Rep.
2017;7(1):11554.
18. Borlotti A, Khir AW, Rietzschel ER, De Buyzere ML, Vermeersch S, Segers P. Noninvasive
determination of local pulse wave velocity and wave intensity: changes with age and gender in
the carotid and femoral arteries of healthy human. J Appl Physiol (1985). 2012;113(5):727–35.
19. Eikendal ALM, den Ruijter HM, Haaring C, Saam T, van der Geest RJ, Westenberg JJM, et al.
Sex, body mass index, and blood pressure are related to aortic characteristics in healthy, young
adults using magnetic resonance vessel wall imaging: the AMBITYON study. Magma (New
York, NY). 2018;31(1):173–82.
20. Friedman BA, Alpert BS, Osborn D, Prisant LM, Quinn DE, Seller J. Assessment of

the validation of blood pressure monitors: a statistical reappraisal. Blood Press Monit.
2008;13(4):187–91.
21. Yan R. Chapter 6: proposal for the evaluation of wearable cuff-less devices. Thesis. In:
Evaluation of the wearable cuff-less blood pressure measuring devices; 2009.
22. Yan IR, Poon CC, Zhang YT. Evaluation scale to assess the accuracy of cuff-less blood pres-
sure measuring devices. Blood Press Monit. 2009;14(6):257–67.
23. Clinical investigation of medical devices for human subjects — good clinical practice ISO
14155:2011(E); 2011.
24. Alpert BS. The Accutension Stetho, an automated auscultatory device to validate automated
sphygmomanometer readings in individual patients. J Hum Hypertens. 2018;32(6):455–9.
25. Bland JM, Altman DG. Agreement between methods of measurement with multiple observa-
tions per individual. J Biopharm Stat. 2007;17(4):571–82.
26. O’Brien E, Petrie J, Littler W, de Swiet M, Padfield PL, Altman DG, et al. An outline of the
revised British Hypertension Society protocol for the evaluation of blood pressure measuring
devices. J Hypertens. 1993;11(6):677–9.
27. O’Brien E, Atkins N, Stergiou G, Karpettas N, Parati G, Asmar R, et al. European Society of
Hypertension International Protocol revision 2010 for the validation of blood pressure measur-
ing devices in adults. Blood Press Monit. 2010;15(1):23–38.
28. Stergiou GS, Alpert B, Mieke S, Asmar R, Atkins N, Eckert S, et al. A universal standard
for the validation of blood pressure measuring devices: association for the Advancement
of Medical Instrumentation/European Society of Hypertension/International Organization
for Standardization (AAMI/ESH/ISO) Collaboration statement. Hypertension (Dallas, Tex:
1979). 2018;71(3):368–74.
29. Yan IR, Poon CC, Zhang YT. A protocol design for evaluation of wearable cuff-less blood
pressure measuring devices. In: Conference proceedings: annual international conference of
the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and
Biology Society annual conference 2009; 2009. p. 7045–7.
30. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of
clinical measurement. Lancet (London, England). 1986;1(8476):307–10.
31. Bland JM, Altman DG. Applying the right statistics: analyses of measurement studies.

Ultrasound Obstet Gynecol. 2003;22(1):85–93.
32. Saugel B, Dueck R, Wagner JY. Measurement of blood pressure. Best Pract Res Clin

Anaesthesiol. 2014;28(4):309–22.
33. Association for the Advancement of Medical Instrumentation. American National Standard.
ANSI/AAMI/ISO 81060-2:2013 non-invasive sphygmomanometers—part 2: clinical investi-
gation of automated measurement type. Arlington: AAMI; 2013.
224 W. J. Verberk
34. Non-invasive sphygmomanometers—part 1: requirements and test methods for non-automated

measurement type. ANSI/AAMI/ISO 81060-1:2007; 2007.
35. Medical electrical equipment—part 2-34: particular requirements for the basic safety and
essential performance of invasive blood pressure monitoring equipment. IEC 60601-2-
34:2011; 2011.
36. Lin WH, Wang H, Samuel OW, Liu G, Huang Z, Li G. New photoplethysmogram indica-
2018;39(2):025005.
37. Silke B, McAuley D. Accuracy and precision of blood pressure determination with the
Finapres: an overview using re-sampling statistics. J Hum Hypertens. 1998;12(6):403–9.
38. Tang Z, Tamura T, Sekine M, Huang M, Chen W, Yoshida M, et al. A chair-based unobtrusive
cuffless blood pressure monitoring system based on pulse arrival time. IEEE J Biomed Health
Inform. 2017;21(5):1194–205.
39. Shin H, Min SD. Feasibility study for the non-invasive blood pressure estimation based on ppg
morphology: normotensive subject study. Biomed Eng Online. 2017;16(1):10.
40. Huynh T, Chung W-Y. Radial electrical impedance: a potential indicator for noninvasive cuff-
less blood pressure measurement. J Sens Sci Tech. 2017;26(4):239–44.
41. Schoot TS, Weenk M, van de Belt TH, Engelen LJ, van Goor H, Bredie SJ. A new cuff-
less device for measuring blood pressure: a real-life validation study. J Med Internet Res.
2016;18(5):e85.
42. Nabeel PM, Joseph J, Karthik S, Sivaprakasam M, Chenniappan M. Bi-modal arterial compli-
ance probe for calibration-free cuffless blood pressure estimation. IEEE Trans Biomed Eng.
2018;65(11):2392–404.
43. Wang Y, Liu Z, Ma S. Cuff-less blood pressure measurement from dual-channel photoplethys-
mographic signals via peripheral pulse transit time with singular spectrum analysis. Physiol
Meas. 2018;39(2):025010.
44. Zhang Q, Zhou D, Zeng X. Highly wearable cuff-less blood pressure and heart rate monitor-
ing with single-arm electrocardiogram and photoplethysmogram signals. Biomed Eng Online.
2017;16(1):23.
45. Miao F, Fu N, Zhang YT, Ding XR, Hong X, He Q, et al. A novel continuous blood pres-
sure estimation approach based on data mining techniques. IEEE J Biomed Health Inform.
2017;21(6):1730–40.
46. Watanabe N, Bando YK, Kawachi T, Yamakita H, Futatsuyama K, Honda Y, et al. Development
and validation of a novel cuff-less blood pressure monitoring device. JACC Basic Trans Sci.
2017;2(6):631–42.
47. Xiao-Rong D, Yan BP, Yuan-Ting Z, Jing L, Peng S, Ni Z. Coherence analysis of invasive
blood pressure and its noninvasive indicators for improvement of cuffless measurement accu-
racy. Conf Proc IEEE Eng Med Biol Soc. 2017;2017:2255–8.
48. Kim S, Lee JD, Park JB, Jang S, Kim J, Lee SS. Evaluation of the accuracy of a new cuff-
less magnetoplethysmography blood pressure monitor in hypertensive patients. Pulse (Basel,
Switzerland). 2018;6(1-2):9–18.
49. Cicolini G, Pizzi C, Palma E, Bucci M, Schioppa F, Mezzetti A, et al. Differences in blood pres-
sure by body position (supine, Fowler’s, and sitting) in hypertensive subjects. Am J Hypertens.
2011;24(10):1073–9.
50. Krisai P, Vischer AS, Kilian L, Meienberg A, Mayr M, Burkard T. Accuracy of 24-hour ambu-
latory blood pressure monitoring by a novel cuffless device in clinical practice. Heart (British
Cardiac Society). 2019;105(5):399–405
51. Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival
Time (PAT) in anesthetized patients Presented at EMBEC2017 - European Medical and
Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017.
52. Solà J, Ghamri Y, Proença M, Braun F, Lemkaddem A, Pierrel N, et al. Tracking blood pressure
changes in anesthetized patients: the optical blood pressure monitoring (oBPM) technology;
Conference Paper 40th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society, At Honolulu, Honolulu, Hawaii, USA. July 17-21, 2018.
53. Matsumura K, Rolfe P, Toda S, Yamakoshi T. Cuffless blood pressure estimation using only a
smartphone. Sci Rep. 2018;8(1):7298.
Chapter 13
Cuffless Blood Pressure Monitoring:
The Future for the Evaluation
and Management of Hypertension
George S. Stergiou
Abstract The upper arm-cuff blood pressure measurement using the manual aus-
cultatory or automated oscillometric method remains the cornerstone for the diag-
nosis and management of hypertension. However, this method has major deficiencies
which seriously undermine the accurate evaluation of blood pressure, including the
fact that it estimates rather than measures blood pressure, differs from intra-arterial
measurement, might be affected by cuff inflation and arterial occlusion, and assesses
only intermittently and only static blood pressure, ignoring thereby the continuous
dynamic blood pressure variability. The development of reliable continuous cuffless
blood pressure measurement technology can provide a complete picture of the 24-h
blood pressure profile and behaviour for long periods, which will reflect the true
burden of blood pressure on the cardiovascular system. However, there are several
technical research questions that need to be addressed regarding the accuracy of
cuffless blood pressure measurement and its persistence with time and in conditions
of acute and chronic changes in blood pressure and other hemodynamic parameters.
There are also clinical research questions, including the optimal sampling method,
reproducibility, clinical relevance and thresholds for hypertension diagnosis. When
the abovementioned research questions for this novel method are adequately
addressed, the conventional methods for office, ambulatory and home blood pres-
sure measurement will become absolute and replaced by the evaluation of the com-
plete evaluation of the true blood pressure profile.
Keywords Accuracy · Blood pressure measurement · Cuffless · Methodology ·

Validation
G. S. Stergiou (*)
Hypertension Center STRIDE-7, National and Kapodistrian University of Athens,
Third Department of Medicine, Athens, Greece
e-mail: gstergi@med.uoa.gr

226 G. S. Stergiou
Century of Clinical Use of Cuff-Based Blood Pressure

A
Measurement in Hypertension Management
The history of blood pressure measurement evolved in 1733 when Reverend Stephen
Hales conducted his famous experiment by placing a glass tube in the artery of horse
and observed the blood to rise high due to blood pressure [1]. However, it was in
1896 that Scipione Riva-Rocci first estimated blood pressure using an inflatable cuff
to encircle the upper arm aiming to occlude the brachial artery [1]. In 1905, Nicolai
Sergeivich Korotkof who was surgeon at the Russian army founded the auscultatory
measurement of systolic and diastolic blood pressure using a Riva-Rocci upper arm
cuff [1, 2]. The oscillometric cuff-based blood pressure measurement was introduced
in 1876 by the French physiologist Etienne Marey [2, 3]. The upper arm-cuff blood
pressure measurement has been the cornerstone for the evolution of clinical
hypertension.
In the last 100 years the evolution of clinical hypertension has been an excellent
model for evidence-based medicine. A total of 61 prospective observational out-
come studies including one million adults with 12.7 million person-years at risk and
56,000 cardiovascular deaths during follow-up showed that blood pressure is
strongly and directly related to cardiovascular and total mortality, without no evi-
dence of a threshold up to at least 115 mmHg systolic and 75 mmHg diastolic [4].
More importantly, a total of 122 interventional randomized outcome clinical trials
including almost 350,000 subjects demonstrated the benefits of treatment induced
blood pressure lowering using different antihypertensive drug classes in reducing
the risk of fatal and nonfatal cardiovascular events [5]. This enormous database of
mega-trials, which is the basis of the current management of hypertension, has been
almost exclusively based on auscultatory upper-arm cuff blood pressure measure-
ments. In the last 20 years the automated oscillometric blood pressure measurement
method is progressively replacing the manual auscultatory method and has been
increasingly used in hypertension outcome trials [6]. The oscillometric blood pres-
sure measurement method aims to replicate the auscultatory method, which is taken
as reference in empirical development of estimation algorithms [2].
Problems of Cuff-Based Blood Pressure Measurement
Despite the indisputable value of the upper-arm cuff blood pressure measurement
method in the identification and management of hypertension worldwide, this
method has major deficiencies which seriously undermine the accurate evaluation
of blood pressure. The main issues with the upper-arm cuff blood pressure measure-
ment method (auscultatory and oscillometric) are listed below.

(a) Provides an “estimation” of the blood pressure level rather a true
measurement.
13 Cuffless Blood Pressure Monitoring: The Future for the Evaluation… 227
A true “measurement” can only be obtained directly using the invasive intra-
arterial method.
(b) Underestimates the systolic and overestimates the diastolic blood pressure.
Picone et al. performed a meta-analysis of studies comparing intra-arterial
brachial versus cuff-based brachial blood pressure and showed that the latter
considerably underestimates intra-arterial systolic blood pressure (by about
6 mmHg) and overestimates diastolic blood pressure (by about 6 mmHg), lead-
ing thereby to a large underestimation of pulse pressure by about 12 mmHg [7].
From the scientific and the physiological point of view, the invasive intra-
arterial blood pressure measurement is the meaningful and relevant one [8].
(c) Provides intermittent rather than continuous measurement of blood pressure.
The current methodology applied for the evaluation of blood pressure in
clinical practice can give only a very limited picture (snapshots) of the 24 h
blood pressure profile. Thus, the widely used methods for blood pressure mea-
surement in the office, at home and with 24 h ambulatory monitoring provide
only 2–100 blood pressure readings out of 100,000 heart cycles per 24 h (each
one generating a blood pressure value). This is a serious limitation of the cuff-
based blood pressure measurement method as the dynamic variation of blood
pressure during routine daily challenges cannot be demonstrated.
(d) Is unable to evaluate beat-to-beat blood pressure variability.
The dynamic variation of blood pressure during routine physical and mental
activities certainly puts addition stress on the heart and vasculature independent
of the average blood pressure [9, 10]. As mentioned above, the currently used
conventional methods for the evaluation of blood pressure in the office, at home
and with 24 h ambulatory monitoring obtain a very small sample of the 24 h
beat to-beat profile and therefore provide very limited information on the bur-
den of blood pressure variability. Thus, at present the management of hyperten-
sion is based on the average value of few blood pressure measurements and
inevitably the impact of variability has been neglected [9, 10].
(e) Blood pressure is measured in static conditions only and fluctuations are not
assessed.
A major disadvantage of both the auscultatory and the oscillometric blood
pressure measurement method is their inability to obtain a valid reading during
body movement. Thus, the blood pressure reactivity and instability, and the epi-
sodic blood pressure rise during usual daily activities cannot be evaluated [9, 10].
(f) The blood pressure level may be affected by the cuff inflation.
The cuff inflation and the resulting compression of the upper arm is sensed
by the user and may by itself affect the blood pressure level during wakefulness
and during sleep [11] giving thereby a distorted estimate of the true blood
pressure.
(g) The blood pressure level may be affected by the vascular occlusion.
It is not clear whether and in what extent the cuff inflation induced arm com-
pression which alters the circulation by occluding the brachial artery and vein
affects the level of measured blood pressure.
228 G. S. Stergiou
he Promise of Cuffless Blood Pressure Measurement

T
Technology
Blood pressure is a continuous variable with dynamic characteristics of variability

during routine physical and mental activities. Therefore, the crucial question for the
medical engineer and the crucial need for the clinical doctor is how to quantify
accurately the burden of the continuous beat-to beat blood pressure fluctuation on
the cardiovascular system. The full picture of the dynamic blood pressure behaviour
cannot be obtained by intermittent cuff-based blood pressure measurement as usu-
ally obtained in clinical practice. It can only be provided by continuous beat-to-beat
blood pressure measurement which at present can be obtained only invasively
(intra-arterial) and therefore is impractical for clinical use and unethical for clinical
research (unless in case of medical indication for intra-arterial monitoring and for
short-term only). Continuous beat-to-beat measurement of ambulatory blood pres-
sure in the finger has been developed using the volume-clamp method, which has
been shown to underestimate blood pressure and is restricted to non-invasive evalu-
ation of beat-to-beat changes in research studies [12, 13].
The ultimate approach for blood pressure evaluation should be (1) non-invasive and
cuffless, so that the individual does not sense the measurement procedure and therefore
an unbiased evaluation is feasible in multiple repeated measurements, (2) continuous so
as the full 24-h profile of blood pressure can be evaluated rather than a small sample of
measurement and (3) unaffected of body movements so that the dynamic fluctuation of
blood pressure during routine daily activities can be accurately recorded.
The technologies for cuffless blood pressure measurement such those described
in this book (see previous chapters) have the potential to satisfy all the requirements
for optimal evaluation of the 24 h blood pressure profile. An accurate blood pressure
measuring device that can be used for several days and record a large sample of
blood pressure measurements without affecting the individual’s routine behaviour
will provide a complete picture of the true blood pressure level and variability. Thus,
for the first time in the history of clinical hypertension the true impact of the blood
pressure on the cardiovascular system will be accurately quantified.
Technical Research Questions
Further to the technical and research debates presented in the previous chapters of
this book, several research questions arise regarding the accuracy of cuffless blood
pressure measurement and its persistence in conditions of acute and chronic changes
in blood pressure and other hemodynamic parameters. International standards
which are specific for the evaluation of the accuracy of cuffless continuous blood
pressure monitors should be urgently developed and agreed.
• Which individual patient parameters need to be provided to improve the accu-
racy of cuffless blood pressure estimation.
• Which reference blood pressure measuring device and method to be used for
static calibration.
13 Cuffless Blood Pressure Monitoring: The Future for the Evaluation… 229
• Which calibration method can be accurately replicated in general practice.

• Whether recalibration is necessary for retaining blood pressure measurement
accuracy and at what conditions and time intervals.
• Whether shortly after calibration accurate cuffless blood pressure measurement
is obtained in static conditions.
• Whether accurate cuffless blood pressure measurement is retained in conditions
of acute change in blood pressure and heart rate (increase or decrease) and other
hemodynamic parameters (physical or mental activity, emotional change, sleep,
disease state, etc.).
• Whether accurate cuffless blood pressure measurement is retained when the blood
pressure level is changed (increased or decreased) in response to drug effect or
other factors (disease state, body weight change, temperature change or others).
Clinical Research Questions
Accurate non-invasive continuous cuffless blood pressure measurement has never

been obtained in clinical practice. This method is not the same as intra-arterial con-
tinuous blood pressure monitoring which is performed in bedridden patients, nei-
ther is the same as 24 h ambulatory blood pressure monitoring which is intermittent,
takes measurements only in static conditions and is affected by cuff inflation and
arm compression. As a novel method for blood pressure measurement, several fun-
damental clinical research questions need to be addressed.
• Reproducibility compared to 24 h cuff-based ambulatory blood pressure
monitoring.
• Relationship with indices of preclinical target organ damage and cardiovascular
risk compared to 24 h cuff-based ambulatory blood pressure monitoring.
• Optimal 24 h blood pressure sampling and number of monitoring days that give
the highest reproducibility and correlation with indices of preclinical target organ
damage.
• Thresholds for hypertension diagnosis for 24 h, daytime and night-time average
blood pressure values compared to conventional ambulatory blood pressure.
• Optimal indices and thresholds for blood pressure variability.
Conclusion
The development of a reliable continuous cuffless blood pressure measurement

technology is expected to end the era of conventional blood pressure measurement
and change the practice of hypertension management. This method promises to pro-
vide a complete picture of the 24 h blood pressure profile and behaviour for long
periods, which will represent the true burden of blood pressure on the cardiovascu-
lar system. Thus, the conventional methods currently used for blood pressure evalu-
ation in the office and at home, and for 24 h ambulatory blood pressure monitoring
230 G. S. Stergiou
will probably become obsolete. When the abovementioned research questions on

the measurement accuracy and the clinical relevance of this novel method are ade-
quately addressed, then the chapter on blood pressure measurement for hyperten-
sion diagnosis and management will have to be rewritten.
References
1. O’Brien E, Fitzerald D. The history of indirect blood pressure measurement. In: O’Brien
E, O’Malley K, editors. Handbook of hypertension, Vol. 14: blood pressure measurement.
Amsterdam: Elsevier; 1991. p. 1–54.
2. Pickering TG. Blood pressure measurement. In: Ambulatory monitoring and blood pressure
variability, Part 1. London: Science Press; 1990. p. 2.1–2.16.
3. Marey EJ. Pression et vitesse du sang. In: Masson G, editor. Physiologie experimentale, Vol.
2. Paris; 1876. p. 307.
4. Lewington S, Clarke R, Qizilbash N, Peto R, Collins R, Prospective Studies Collaboration.
Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of indi-
vidual data for one million adults in 61 prospective studies. Lancet. 2002;360(9349):1903–13.
5. Thomopoulos C, Parati G, Zanchetti A. Effects of blood pressure-lowering treatment on car-
diovascular outcomes and mortality: 14—effects of different classes of antihypertensive drugs
in older and younger patients: overview and meta-analysis. J Hypertens. 2018;36(8):1637–47.
6. Giorgini P, Weder AB, Jackson EA, Brook RD. A review of blood pressure measurement pro-
tocols among hypertension trials: implications for “evidence-based” clinical practice. J Am
Soc Hypertens. 2014;8:670–6.
7. Picone DS, Schultz MG, Otahal P, Aakhus S, Al-Jumaily AM, Black JA, Bos WJ, Chambers
JB, Chen CH, Cheng HM, Cremer A, Davies JE, Dwyer N, Gould BA, Hughes AD, Lacy PS,
Laugesen E, Liang F, Melamed R, Muecke S, Ohte N, Okada S, Omboni S, Ott C, Peng X,
Pereira T, Pucci G, Rajani R, Roberts-Thomson P, Rossen NB, Sueta D, Sinha MD, Schmieder
RE, Smulyan H, Srikanth VK, Stewart R, Stouffer GA, Takazawa K, Wang J, Westerhof
BE, Weber F, Weber T, Williams B, Yamada H, Yamamoto E, Sharman JE. Accuracy of
cuff-measured blood pressure: systematic reviews and meta-analyses. J Am Coll Cardiol.
2017;70(5):572–86.
8. Stergiou GS, Kollias A, Protogerou AD. Evidence on blood pressure measurement methodol-
ogy and clinical implementation: research agenda for the 21st century. J Am Coll Cardiol.
2017;70(5):587–9.
9. Parati G, Stergiou GS, Dolan E, Bilo G. Blood pressure variability: clinical relevance and
application. J Clin Hypertens (Greenwich). 2018;20(7):1133–7.
10. Rothwell PM. Limitations of the usual blood-pressure hypothesis and importance of variabil-
ity, instability, and episodic hypertension. Lancet. 2010;375:938–48.
11. Sheshadri V, Tiwari AK, Nagappa M, Venkatraghavan L. Accuracy in blood pressure moni-
toring: the effect of noninvasive blood pressure cuff inflation on intra-arterial blood pressure
values. Anesth Essays Res. 2017;11:169–73.
12. Imholz BP, Wieling W, van Montfrans GA, Wesseling KH. Fifteen years experience with finger
arterial pressure monitoring: assessment of the technology. Cardiovasc Res. 1998;38:605–16.
13. O’Brien E, Parati G, Stergiou G, Asmar R, Beilin L, Bilo G, Clement D, de la Sierra A, de
Leeuw P, Dolan E, Fagard R, Graves J, Head GA, Imai Y, Kario K, Lurbe E, Mallion JM,
Mancia G, Mengden T, Myers M, Ogedegbe G, Ohkubo T, Omboni S, Palatini P, Redon J,
Ruilope LM, Shennan A, Staessen JA, vanMontfrans G, Verdecchia P, Waeber B, Wang J,
Zanchetti A, Zhang Y, European Society of Hypertension Working Group on Blood Pressure
Monitoring. European Society of Hypertension position paper on ambulatory blood pressure
monitoring. J Hypertens. 2013;31:1731–68.
Index
A aorta/arm complex arterial system, 79, 80

Acceleration plethysmogram (APG), 118 catheter sensor positions, 79, 81
Active implantable medical devices digital/radial, 76
(AIMD), 197 distinct pulse-like features, 78, 79
Aging, 164, 167, 181, 183–186 features, 77
Analytical models, 143, 146, 147 noninvasive method, 81
Android-based tablet, 102 propagation velocities, 77
ANSI/AAMI/ISO standard, 217, 221 pulse-like protrusions, 77
Aortic PTT measurements, 181 radial pulse signatures
Applanation tonometry, 109, 110, 120–124, high-fidelity, 77
128–131 qualitative comparison, 77, 78
Architecture of a cuffless BPM Valsalva maneuver, 77–79
advantages, 37 second systolic/diastolic peak, 81
arterial-occlusion monitors, 37–39 Arterial stiffness (AS), 4
characteristics, 37, 39 aging, 26
disadvantages, 37 cath lab patient, 90
initialization layer, 41 coronary heart disease risk, 26
invasive BPM, 37, 38 digital pulse, 91
limitation, 39 featuredness, 91
non-pressure units, 37 flexible arteries, 90
principles and output units, 37, 38 identification, distinct inversions, 92
processing layer, 40, 41 pancreaticoduodenectomy surgery, 91
pulsatility signals, 37 parameter, 92
transducer layer, 40 PDA framework, 91
user’s skin surface, 37 pressure pulses, 90
Arm cuff inflation, 10 pulse analysis, 92
Arterial catheters, 53, 109 second derivative analysis, 92
Arterial hemodynamics, 5 vasodilation, 90
Arterial length, 62 wall structure, 90
Arterial lines, 53 young athlete, 90
Arterial-occlusion BPM, 35 Arterial system, 16
Arterial-occlusion monitors, 37–39 Arterial tonometry, 61
Arterial periphery, 76 Artificial neural network, 155
Arterial pulse, 2, 16 ASEAN Medical Device Directive
Arterial reflections (AMDD), 192

Monitoring, https://doi.org/10.1007/978-3-030-24701-0
232 Index
Association for the Advancement of Medical BP variability

Instrumentation (AAMI), 177 assessment, 10
Association of Southeast Asian cardiovascular regulatory mechanisms,
Nations (ASEAN) member 10, 11
states, 192 physiological conditions, 10
Asymmetric T-shaped model, 76 Brachial-ankle pulse wave velocity, 117
Augmentation index (AIx), 112, 115 Bramwell–Hill equation, 66, 166, 168
Auscultation, 51, 52
Auscultatory method, 10, 16, 23, 226
Australia’s medical device classification C
system, 197, 198 Calibration, 62–65, 71, 206, 207
Australian Register of Therapeutic Goods method, 229
(ARTG), 198 parameters, 35, 41
Australian Regulatory Guidelines for Medical PTT to BP, 69, 70
Devices (ARGMD), 197 Calibration curve
Automated oscillometric devices, 206 cardiovascular aging, 164
Autonomic nervous index, 182 construction, 164
hybrid, 184, 185
parameters, 164
B person-specific, 177
Ballistocardiography (BCG), 48, 81 population-based, 177, 178
measurement, 68 practical form, 170, 171
waveform, 180, 181 Canada’s medical device classification system,
Basic glossary of cuffless BP monitoring, 196, 197
32–35 Cardiac output (CO), 67
Beat-by-beat oscillations, 9 Cardiac signals, 139, 149
Beat-to-beat BP variability, 227 CareTaker (CT), 76, 96, 99, 101–103
B-H equation, 143 Carotid artery, 62
Big data, 3, 4 Center for Devices and Radiological Health
Bioimpedance method, 68 (CDRH), 194
Biopac system, 57 Central arterial line comparisons, 98–100
Bland–Altman plot, 214 Chronophotography, 2
Blood pressure (BP) Clinical and consumer applications, 191
estimation, 139 Clinical hypertension, 226, 228
horse, measurements, 19 Clinical studies, 115, 119–128
indicators, 140, 148, 150 CNSystems, 62
insurance examination, 24, 25 Cold pressor, 171
measurement, 139 Communication modalities, 102
pulse waveforms, 112 Competent Authority, 195
PWA, 118, 119 Complex chip design, 6
variability (see BP variability) Computer vision, 147, 148
Blood pressure monitoring (BPM) Consumer devices, 4
clinical practice, 24 Continuous BP, 84, 103, 205, 206, 213
cuffless (see Cuffless BPM) Continuous BP monitoring, 71
graphic recordings, 16 arm cuff inflation, 10
physiological terms, 32–33 auscultatory method, 10
technological terms, 32–35 beat-by-beat oscillations, 9, 10
BP change tracker, 71 clinical diagnostic, 10
BP measurement clinical issues, 11
challenge, clinical practice, 204 CPAP, 11
disadvantages, 204 in daily practice, 10
intermittent, 204 invasive recordings, 11
upper-arm cuff inflating, 204 noninvasive, 11
wearable cuffless devices, 210 oscillometric method, 10
Index 233
prognostic information, 10 Deep RNN, 154–156

techniques, 11 Definition of a cuffless BPM
Continuous/intermittent vs. cuff/cuffless arterial-occlusion, 35
devices, 205 classification, 35, 36
Continuous measurement, 43, 50, 52–56 noninvasive, 35
Continuous noninvasive automated perimeter, 35
sphygmomanometer, 205 PWA, 37
Continuous, noninvasive blood pressure PWV, 37
(cNIBP), 56, 62 Designated marketing authorization holder
Continuous positive air pressure (CPAP), 11 (DMAH), 199–200
Cuff Devices
definition, 205 hemodynamic information, 28
inflation, 227 mobile/mHealth, 27
Cuff-based BP measurement optical sensors, 27
hypertension management, 226 pulse tonometry, 27
issues, 226, 227 sensor technology, 27
Cuff-based upper-arm devices, 206 volume clamp/plethysmography, 27
Cuff/cuffless vs. continuous/intermittent wrist and finger sensors, 27
devices, 205 Device validation, 6
Cuffless BP, 164, 181–183 Diastolic blood pressure (DBP), 52, 142, 147,
sensor, 205 155, 173
machine learning (see Machine learning) Diastolic pressure (DP), 16
validation (see Validation protocol) Dynamic component (DC), 45, 48
Cuffless BPM, 39
Cuffless devices, 3, 5, 6, 204, 206
Cuffless techniques E
monitoring, BP, 5, 6 Electrical bioimpedance (EBI) waveforms,
wearable devices, 5 180, 181
Electrocardiogram (ECG), 44, 46, 49–51,
56–58, 63, 139
D measurement methods, 68
Data-driven based machine learning method Electronic devices, 16
analytical model, 146 Empirical model, 143
artificial intelligence, 143 EU medical device classification system,
big data analysis techniques, 143 195, 196
deep learning, 146 European Medicines Agency (EMA), 196
feature and value, 146
M-K equation, 140
PTT and BP relationship, 141, 142 F
PTT-based BP estimation, 143–145 Femoral artery, 62
PWV recording, 140, 141 Finapres, 62
static analytical model, 146 Finger cuff, 102, 103
vs. theory-driven analytical model, 146, 147 Food and Drugs Act, 193, 196
Data mining, 183 Four-quadrant plot (FQP), 215
Deep learning, 148 Framingham Heart Study, 24, 26
advantages, 152 Fung’s hyper-elastic model, 167
artificial neurons, 152
classification, 152
cuffless BP, 153, 154 G
DNN, 153, 154 Generalized transfer functions (GTF), 114, 128
vs. machine learning, 152, 153 Global Harmonization Task Force (GHTF), 193
neural networks, 153 Good clinical practice (GCP)
properties, 152 clinical investigation, medical devices, 217
Deep neural networks (DNNs), 153, 154, 159 duration, BP measurement procedure, 217
234 Index
Good clinical practice (GCP) (cont.) autonomic nervous function

invasive BP, 217 and index, 182
observer measurement, 218 BCG, 180, 181
phases/interim analyses, 218 data mining, 183
reference device, 217 diastolic and systolic BP, 180
test environment, 218 EBI, 180, 181
Good Manufacturing Practice (GMP), 194 J-K amplitude, 180
Green emitters, 46 PAT, 182–184
peak-to-peak amplitudes, 180, 181
PEP, 180
H PP formula, 180
Haemodynamometer, 19 PPG, 180
Harmonization, 192, 195, 200, 201 proximal and distal arterial, 180
Health Canada (HC), 193 single cuff measurement, 181
Health management, 2, 3, 6 smooth muscle contraction, 180
Heart rate (HR), 67, 115 Valsalva maneuver, 183
PDA, 93 Institute of Electrical and Electronics
Hemodynamic parameters, 229 Engineers (IEEE), 204, 212, 214,
Homeostasis, 10 217, 219
Hot-wire sphygmographs, 56 Intermittent BP measurement, 204, 205
Hughes model, 166 International consensus standards, 200, 201
Hybrid method, 179, 180 International Medical Device Regulators
Hydrostatic maneuver, 172, 173 Forum (IMDRF), 193
Hypertension, 187 International standards, 228
ageing, 26 Inter-subject BP variations, 177
central aortic SP and PP, 17, 18, 26 Intervention-induced BP variations, 177
conventional brachial cuff Intra-arterial catheters, 53
sphygmomanometric Intra-thoracic pressure, 97
measurements, 26 Invasive BP, 217
cuff-based BP measurement, clinical Invasive BPM, 37, 38
use, 226 Invasive measurements, PAT, 53, 54
Framingham Heart Study, 24, 26 Inverse model, 67
management, elevated BP, 27 Inverse square model, 67
risk evaluation, 24
SHEP, 26
stiffening, 26 J
Japan’s medical device classification system,
199, 200
I J-K amplitude, 180
Ice stimulus experiments, 96–98
Impedance cardiography (ICG), 46–48
Impedance plethysmography (IPG), 46–48 K
Infrared (IR), 46, 68 Korotkoff sounds, 4, 16, 52
Initialization algorithm, 39, 41 Kymograph, 20
Initialization layer, 39, 41
Initialization procedure, 39
Initialization, PTT-based BPM L
calibration curve, 164 Laplace’s law, 166
measurements, 164 Laplacian distributed data, 176
parametric models, 164–170 Least squares estimation, 176
properties, 163 Left ventricular ejection time (LVET), 90
reinitialization period, 183–185 Linear least squares regression, 174, 176
waveform features Linear model, 66
accuracy, 183, 184 Linear/nonlinear regression models, 143
aortic PTT measurements, 181 Linear regression, 150, 151
Index 235
Logarithmic model, 66 disadvantages, 67

Long-short-term-memory (LSTM), 148, 154 ECG, 68
PATs, 67
PPG, 68
M PPT, 68
Machine learning VPG, 69
advancement, 155 The Medical Device Coordination Group
artificial neural network, 155 (the MDCG), 201
black box technique, 158 Medical Device Directive (MDD), 195
cuffless BP Medical Device Regulations, 196, 197
analytical model-based method, 148 Medical devices
application, 149 definition, 193
calibration, 150 industry, 192
computer vision, 147, 148 in legal distribution, 192
deep learning, 152–154 regulations, 192
features, 147 Medical Device Single Audit Program
indirect estimation, 149 (MDSAP), 198
linear regression, 150 Mercury manometer, 19, 20
measurement, 149 Methodology, 227
natural language processing, 147, 148 Miniaturization, 6
polynomial regression, 150 Ministry of Health, Labor and Welfare
PPG signals, 147, 148 (MHLW), 199
random forests, 151 Mobile health BP monitors, 11
regression trees, 151 Mobile/mHealth, 27
speech recognition, 147, 148 Moens–Korteweg (M-K) equation, 65, 82,
SVR, 151, 152 140, 143, 166
target function, 149 Motion artifacts, 50, 51
univariate/multivariate linear Multi-electrode system, 46
regression, 148
data-driven method (see Data-driven based
machine learning method) N
deep RNN, 154–156 Natural language processing, 147, 148
development, BP models, 155, 157–158 Neural networks, 153
DNNs, 159 Nocturnal cuff inflations, 11
hand-designed features, 154 Noninvasive, 43, 46, 50, 52–56, 81, 84, 91, 99,
labelled data, 158 100, 102
limit and challenges, 158 Noninvasive cuffless surrogate signal, 4
methods, 71 Non-PTT modalities, 179
MSE, 154 Notified Body (NB), 195, 196
signal modalities, 159 Novel cuffless methodologies, 5
techniques, 5
Marey’s sphygmograph, 20, 22
Marketing Authorization Holder (MAH) O
system, 199 Optical measurements, 131
Mean absolute differences (MAD), 214 Optical sensors, 27
Mean absolute percentage difference Oscillometric method, 10
(MAPD), 214 Oscillometry, 52, 53
Mean arterial pressure (MAP), 52, 97
Mean BP (MBP), 141, 155
Mean pressure, 20 P
Mean squared error (MSE), 154 Parameter error, 175
Measurement methods, PWV Parametric models, PTT-BPM
BCG, 68 empirical models, 169, 170
bioimpedance, 68 hybrid method, 179, 180
calibration, 69, 70 person-specific methods, 171–177
236 Index
Parametric models, PTT-BPM (cont.) Population-based methods, 177–179

population-based methods, 177–179 Practical calibration curve form, 170, 171
practical calibration curve form, 170, 171 Predictive analytics, 149
theoretical models Predictive modelling, 149
arterial vessel, 166 Pre-ejection period (PEP), 50, 64, 174
Bramwell–Hill equation, 166, 168 Premarket approval (PMA) process, 195
cardiovascular and chronological Principles, cuffless methodology, 3, 4
aging, 167 Processing layer, 39–41
Fung’s hyper-elastic model, 167 PTT-based BPM
Hughes’ model, elastic modulus, 166 initialization (see Initialization,
Laplace’s law, 166 PTT-based BPM)
M–K equation, 166 Pulsatility-based algorithm, 39–41
person- and time-specific parameters, Pulsatility energy, 40
168, 169 Pulsatility sensors, 37, 39, 40
physiologic mechanisms, 166 Pulse arrival time (PAT), 63, 204
PWV, 166 auscultation, 51, 52
Wesseling model, 167, 168 BCG, 48
Peak-to-peak amplitudes, 180, 181 Biopac, 57
Peak-to-peak time delay, 176 BP measurement site, 173
Periodic fluid pump, 16 cNIBP, 56
Peripheral arterial line comparisons, 99, 101, 102 continuous, noninvasive BP measurement, 43
Peripheral resistance and cuff BP, 179
arterial pressure pulse, 93, 94 definitions, 44
arterial pressure reflections, 93 ECG-PPG plot, 44
heart rate/LVET/cardiac output, 93 handheld/wrist-worn devices, 57
implementation, 94, 95 hot-wire sphygmographs, 56
left ventricular ejection, 93 invasive measurements, 53, 54
Person-specific calibration curve, 177 IPG/ICG, 46–48
Person-specific methods, 171–177 limitation, 64
Pharmaceutical Affairs Law (PAL), 199 measurements, 44, 55
Pharmaceutical and Medical Devices Act motion artifacts, 50, 51
(PMD Act), 193, 199 NIBP, 56
Pharmaceutical and Medical Devices Agency oscillometry, 52, 53
(PMDA), 198–200 PCG, 48, 49
Pharmaceutical Safety and Environmental PEP, 50, 174
Health Bureau (PSEHB), 199 PPG (see Photoplethysmogram (PPG))
Phases/interim analyses, 218 PWV, 49, 56, 57
Phonocardiogram (PCG), 48, 49 S1/S2 heart sounds, 49
Photoplethysmography (PPG), 110, 111, 117, SCG, 48
118, 125–128, 130, 131, 139, 147, and smooth muscle contraction, 181
149, 155, 178–184 somnomedics, 57
applied pressure, 46 Sotera Wireless markets, 56
DC, 45 time delay, 56
ECG waveform, 44 time difference, 64
green emitters, 46 tonometry, 54, 55
IR, 46 variability, 183
photodetector, 45 ViSi Mobile, 56
pulse wave, 44 volume clamp method, 53–55
reflective sensors, 45 Pulse contour analysis, 75
SC, 45 Pulse decomposition analysis (PDA)
transmissive sensors, 45 central arterial line comparisons, 98–100
zero transmural pressure, 46 central reflection sites, 76–82
Poiseuille manometer, 20 component, 76
Polynomial regression, 150 hardware platform, 102–104
Index 237
hemodynamic parameters, 75 physiological mechanism, 140

HR, 93 practical approach, 64, 65
ice stimulus experiments, 96–98 reciprocal/PWV, 62
implementation regulatory agencies, 52
AS, 90–93 ViSi’s cNIBP, 62
modeling of pulses, 87–89 wearable objects, 62
physiological confounders, 89, 90 Pulse wave analysis (PWA), 37
pulse reflections, 82–87 advantages, 132
peripheral arterial line comparisons, 99, AIx, 112
101, 102 amplification of pressure waveform, 113
peripheral resistance (see Peripheral aortic pressure, 112
resistance) applanation tonometry, 109, 110, 130, 131
physiological model, 76 arterial catheters, 109
principal mechanism, 76 BP, 112, 118, 119
radial/digital arterial pressure pulse, 82 challenges, 132
second systolic peak, 76 classification, 112, 113
Valsalva experiments, 96–98 clinical evidence
Pulse plethysmography (PPG), 63, 68 applanation tonometry, 120–124, 128, 129
Pulse pressure (PP), 16, 26 PPG, 125–128, 130
Pulse reflections DBP, 112
arrival time, 82 development, 132
arterial paths, 82 elastic and geometric properties, 111
arterial pulse velocity profiles, 85 endogenous components, 132
delay times, 84 exogenous components, 132
model’s equations, 82 frequently encountered features, 115–117
Moens–Korteweg equation, 82 GTF, 114
preliminary tests, 84 history, 108
pressure dependence, 82 inter-subject variability, 131
pressure response, 85, 86 measurement site, 131
propagation velocity, 85–87 measurement system, 132
QRS complex, 85 morphological analysis, 107
renal reflection coefficient, 83, 84 morphological differences, 115
second systolic, 82 optical measurements, 131
timing considerations, 87 peripheral pressure, 113
Valsalva, 84 PPG, 110, 111, 117, 118, 131
Young’s modulus, 82 SBP, 112
Pulse tonometry, 27 sphygmograph recording, 108
Pulse transit time (PTT), 4, 5, 11, 27, 204 sphygmomanometer, 108
advantages, 70 systolic pressure, 113
arterial length, 62 TFs, 113, 114
arterial sites, 63 Pulse waveform, 16
arterial stiffness, 63 Pulse wave velocity (PWV), 4, 37, 166,
BP estimation, 143 168–170, 178, 180, 182, 186, 204
and BP relationship, 141–142 advantages, 70
calibration, 69, 70, 141 and arterial BP, 141
continued miniaturization, 58 BP change tracker, 71
description, 63 brachial-to-radial, 142
estimation, cuffless BP, 140 Bramwell–Hill equation, 66
infrared, 46 cardiac pulse, 50
limitation, 63, 64 computerized measurement system, 56, 57
mathematical models, 66, 67 continuous BP monitoring, 71
mechanical measurements, 44 continuous measurement, 50
and PAT (see Pulse arrival time (PAT)) description, 63
PEP, 50, 64 ECG, 63
238 Index
Pulse wave velocity (PWV) (cont.) S

hydrostatic pressure effects, 56 Second systolic peak, 76
limitation, 63, 64 Seismocardiogram (SCG), 48
machine learning methods, 71 Sensors, 5
mathematical modeling, 65–67 Set point, 62
measurement, 62 (see also Measurement Single cuff measurement, 181
methods, PWV) Slow breathing, 171
Moens–Korteweg equation, 65 Smooth muscle contraction, 180
PAT and PTT, 56 Somnomedics, 57
PPG, 63 Somnotouch PTT-based BPM, 179
predictor, vascular stiffness, 27 Sotera Wireless markets, 56
principle, 62, 140 Speech recognition, 147, 148
recording, 140 Sphygmocardiography, 23, 24
set point, 62 Sphygmograph recording, 108
time delay, 56 Sphygmography, 2
trigger for absolute BP measurement, 71 accurate estimation, BP, 21
volume clamp method, 62 asymmetrical, pulse waves, 20
volume clamp/plethysmography, 27 clinical medicine, 23
Young’s modulus for zero arterial modification, 21, 23
pressure, 65 non-invasive BP measurement, 21, 23
recording, pulse, 20
Vierordt’s instrument, 20
Q Sphygmomanometer, 21, 23, 24, 108
Quadratic equation, 176 clinical parameter, 3
Quality System Regulations, 194 clinical use, 2
cuff-based automated oscillometric, 206
detection, markers, 3
R management, hypertension, 5
Radial artery, 62 noninvasive, 205
Random Forest Tree method, 147 static test, 206
Random forests, 151 technological advances, 2
Recurrent neural network (RNN), 148 Stability test, 207
Red optical wavelengths, 68 Standard deviation, 176
Reflectance-mode PPG, 68 Static component (SC), 45, 47
Reflective PPG sensors, 45 Stethoscopes, 192
Registered Certification Bodies Stretch–strain relationship model, 11
(RCBs), 199 Sub-endocardial viability ratio (SEVR), 115
Regression trees, 151 Support vector machine (SVM), 151–152
Regulatory framework and device Support vector regression (SVR), 151, 152
classification Systemic BP, 173
Australia, 197, 198 Systolic BP (SBP), 52, 142, 147, 155, 173
Canada, 196, 197 Systolic Hypertension in the Elderly Project
environmental conditions, 193 (SHEP), 26
European Union (EU), 195, 196 Systolic pressure (SP), 16, 112
Japan, 199, 200
USA, 194, 195
Regulatory systems, 192 T
Reinitialization period, 183–185 Technology, cuffless BP measurement
Ridge linear regression, 147 accuracy, 228, 229
Risk-based classification system, 192, accurate device, 228
193, 200 approaches, 228
Risks and controls, 192, 194, 197 clinical practice, 229
Riva-Rocci technique, 23 continuous beat-to-beat measurement, 228
Riva-Rocci upper arm cuff, 226 continuous variable, 228
Index 239
Test environment, 218 MAPD, 214

Theory-driven analytical model vs. data-driven reference vs. test device, 213
based machine learning method, standard, defined, 213
146, 147 sequential, 209
Therapeutic Goods Administration (TGA), simultaneous, 210
193, 197, 198 stability test, 207
Therapeutic Products Directorate (TPD), 196 standards noninvasive BP monitors, 206
Thermometers, 192 static test, 206
Time-domain profiles, 147 statistical analysis
Tonometry, 54, 55 BP ranges, 211
Tonometry sensor array, 62 gender distribution, 211
Transducer layer, 39, 40 invasive testing, sample size, 212
Transfer functions (TFs), 110, 113, 114, 121 measurement frequency, 211
Transmission-mode PPG measurements, 68 noninvasive testing, sample size,
Transmissive PPG sensors, 45 211, 212
sample size, 211
special-subject population, 212
U time duration, measurement procedure,
Univariate/multivariate linear regression, 148 208, 209
Unpredictable error, 175 Valsalva, 84
US Food and Drug Administration (FDA), 193 Valsalva experiments, 96–98
US medical device classification, 194, 195 Valsalva maneuver, 171, 183
Vascular unloading/method of Penaz, 53
Videoplethysmography (VPG)
V method, 69
Validation protocol Vierordt’s instrument, 20
accuracy studies ViSi Mobile, 56
BP change tests, 219, 220 Volume clamp method, 11, 53–55, 62
calibration, 221
reference device, 219
stability, 219, 220 W
static, 219, 220 Wave propagation, 3
statistical outcomes, 220, 221 Wearable, 43, 50, 55, 58
subjects investigated, 220 Wearable devices, 5
ANSI/AAMI/ISO standards, 206 Wearable objects, 62
Bland–Altman plot, 214, 215 Wearable/unobtrusive system, 159
BP change test Wesseling model, 167, 168
histogram, 209 World Health Organization (WHO), 193
invasive, 208 Wrist and finger sensors, 27
noninvasive, 208 Wrist-worn devices, 4, 57
calibration, 206
GCP (see Good clinical practice (GCP))
histogram, differences, 215, 216 Y
mean difference and standard deviation Young’s modulus, 82
correction, measurements, 213 for zero arterial pressure, 65
error-bands BHS, 214
ESH protocols, 214
FQP, 215, 216 Z
MAD, 214 Zero transmural pressure, 46

2019 Book TheHandbookOfCufflessBloodPres

Uploaded by

Document Information

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

2019 Book TheHandbookOfCufflessBloodPres

Uploaded by

Copyright:

Available Formats

Josep Solà · Ricard Delgado-Gonzalo

The Handbook of Cuffless

ISBN 978-3-030-24700-3 ISBN 978-3-030-24701-0 (eBook)

© Springer Nature Switzerland AG 2019

Let us start with a typical scenario from clinical practice today.

comprehensive publication providing an overview of the emerging field of cuffless

Neuchâtel, Switzerland Josep Solà

AAMI Association for the Advancement of Medical Instrumentation

1 Cuffless Blood Pressure Monitoring and the Advent of a

10 Initialization of Pulse Transit Time-Based Blood

Audrey Adji St Vincent’s Clinical School, University of New South Wales, Sydney,

Jin-Oh Hahn Department of Mechanical Engineering, University of Maryland,

Alberto Avolio, Fatemeh Shirbani, Isabella Tan, and Mark Butlin

Keywords Arterial stiffness · Pulse wave velocity · Pulse transit time ·

A. Avolio (*) · F. Shirbani · I. Tan · M. Butlin

© Springer Nature Switzerland AG 2019 1

 uffless Monitoring of Blood Pressure: A Disruptive Step

concepts in human history, the description of electromagnetic radiation by James

Underlying Principles of Cuffless Methodology

An important overarching concept underlying cuffless measurement of blood pres-

The Handbook of Cuffless Blood Pressure Monitoring

sensor technology, device development and fabrication, signal processing, deep

Is “Cuffless” the Future of Blood Pressure Monitoring?

Abstract The assessment of dynamic features of blood pressure, which represent a

Keywords Continuous blood pressure monitoring · Blood pressure variability ·

© Springer Nature Switzerland AG 2019 9

Keywords Systolic pressure · Diastolic pressure · Mean pressure · Pulse pressure

© Springer Nature Switzerland AG 2019 15

From Pulse Wave to Systolic and Diastolic Pressures

The Ancient Times

The Mercury Manometer

The Kymograph and the Sphygmograph

In 1847, Carl Ludwig (1816–1895) made another modification to Poiseuille’s

Independent from the development of sphygmography, the first truly accurate

The use of sphygmogram to detect elevated pressure was further emphasised by

Clinical Practice Today

Hypertension as a Risk Factor in Cardiovascular Disease

on to conclude the importance of systolic hypertension in the middle-aged and

The Need for Cuffless Blood Pressure Measuring Devices

There is now a demand and a sufficiently advanced technology to develop cuffless

isolated systolic hypertension in the elderly: meta-analysis of outcome trials. Lancet.

Abstract This chapter introduces a systematic framework to navigate through the

Keywords Basic glossary of cuffless blood pressure monitoring · Definition of a

© Springer Nature Switzerland AG 2019 31

Defining the Perimeter of Cuffless Blood Pressure Monitors

Cuffless Blood Pressure monitors

The Architecture of a Cuffless Blood Pressure Monitor

Body Transducer layer Processing layer Output

Invasive blood pressure monitor

Body Transducer layer Processing layer Output

Non-invasive arterial occlusion blood pressure monitor

Body Transducer layer Processing layer Initialization layer Output

Non-invasive cuffless blood pressure monitor

 xamples of Implementation of Cuffless Blood Pressure

A more detailed description of the generic architecture of Fig. 4.4 is provided in this

Table 4.5 Examples of Pulsatility-based algorithms Described in this book

Table 4.6 Examples of Initialization algorithm Described in this book

Introduction to the Technical Chapters of this Book

M. S. Dhillon · M. J. Banet (*)

© Springer Nature Switzerland AG 2019 43

Fig. 5.1 ECG-PPG plot of PAT

Transmissive and Reflective PPG Sensors

Fig. 5.3 Reflective optical

Impedance Plethysmography and Impedance Cardiology

Neuchâtel, Switzerland Josep Solà

1 Cuffless Blood Pressure Monitoring and the Advent of a

10 Initialization of Pulse Transit Time-Based Blood

uffless Monitoring of Blood Pressure: A Disruptive Step

Underlying Principles of Cuffless Methodology

The Handbook of Cuffless Blood Pressure Monitoring

Is “Cuffless” the Future of Blood Pressure Monitoring?

From Pulse Wave to Systolic and Diastolic Pressures

The Ancient Times

The Mercury Manometer

The Kymograph and the Sphygmograph

Clinical Practice Today

Hypertension as a Risk Factor in Cardiovascular Disease

The Need for Cuffless Blood Pressure Measuring Devices

Defining the Perimeter of Cuffless Blood Pressure Monitors

The Architecture of a Cuffless Blood Pressure Monitor

xamples of Implementation of Cuffless Blood Pressure

Introduction to the Technical Chapters of this Book

Transmissive and Reflective PPG Sensors

Impedance Plethysmography and Impedance Cardiology

Other Sensing Technologies

Challenges of PAT-Based Blood Pressure Monitors

The Effect of Pre-ejection Period

PAT = PEP + PTT (5.1)

ositioning PAT-Based Monitors Among Conventional

Technical Details on Measurement Methods

Calibration of PTT to BP

The Future of Technology

Pulse Decomposition Analysis

Evidence of Central Reflection Sites

Modeling of Pulse Reflections

Arterial Stiffness Considerations

Heart Rate Considerations

Peripheral Resistance Considerations

Final Considerations on Implementation

Benchmarks and Clinical Evidence

Ice Stimulus Experiments

Central Arterial Line Comparisons

Peripheral Arterial Line Comparisons

Hardware Implementation and Future Developments