The Pharmacogenomics Journal (2013) 13, 70–79
& 2013 Macmillan Publishers Limited. All rights reserved 1470-269X/13
Gray matter textural heterogeneity as a potential
in-vivo biomarker of fine structural abnormalities
in Asperger syndrome
E Radulescu1,2, B Ganeshan1,
L Minati1,3, FDCC Beacher4,
MA Gray5, C Chatwin6,
RCD Young6, NA Harrison1,2,7
and HD Critchley1,2,7
1
Department of Psychiatry, Brighton & Sussex
Medical School (BSMS), Brighton, UK; 2Sackler
Centre for Consciousness Science, University of
Sussex, Brighton, UK; 3Scientific Department,
Fondazione IRCCS Istituto Neurologico ‘Carlo
Besta’, Milano, Italy; 4Department of Biomedical
Engineering, Stony Brook University School of
Medicine, Stony Brook, NY, USA; 5Experimental
Neuropsychology Research Unit, School of
Psychology and Psychiatry, Monash University,
Melbourne, Victoria, Australia; 6Department of
Engineering and Design, University of Sussex,
Brighton, UK and 7Neurobehavioural Clinic,
Sussex Partnership NHS Foundation Trust, c/o
Sussex Education Centre, Mill View Hospital Site,
Hove, UK
Correspondence:
Dr E Radulescu, Department of Psychiatry,
Brighton and Sussex Medical School, Clinical
Imaging Sciences Centre (CISC), University of
Sussex, Falmer, Brighton BN1 9RR, UK.
E-mail: e.radulescu@bsms.ac.uk
Received 7 September 2011; revised 29
December 2011; accepted 10 January 2012;
published online 14 February 2012
www.nature.com/tpj
ORIGINAL ARTICLE
Brain imaging studies contribute to the neurobiological understanding of
Autism Spectrum Conditions (ASC). Herein, we tested the prediction that
distributed neurodevelopmental abnormalities in brain development impact
on the homogeneity of brain tissue measured using texture analysis (TA; a
morphological method for surface pattern characterization). TA was applied
to structural magnetic resonance brain scans of 54 adult participants (24 with
Asperger syndrome (AS) and 30 controls). Measures of mean gray-level
intensity, entropy and uniformity were extracted from gray matter images at
fine, medium and coarse textures. Comparisons between AS and controls
identified higher entropy and lower uniformity across textures in the AS
group. Data reduction of texture parameters revealed three orthogonal
principal components. These were used as regressors-of-interest in a voxel-
based morphometry analysis that explored the relationship between surface
texture variations and regional gray matter volume. Across the AS but not
control group, measures of entropy and uniformity were related to the
volume of the caudate nuclei, whereas mean gray-level was related to the
size of the cerebellar vermis. Similar to neuropathological studies, our study
provides evidence for distributed abnormalities in the structural integrity of
gray matter in adults with ASC, in particular within corticostriatal and
corticocerebellar networks. Additionally, this in-vivo technique may be more
sensitive to fine microstructural organization than other more traditional
magnetic resonance approaches and serves as a future testable biomarker in
AS and other neurodevelopmental disorders.
The Pharmacogenomics Journal (2013) 13, 70–79; doi:10.1038/tpj.2012.3;
published online 14 February 2012
Keywords: asperger syndrome; caudate; cerebellum; magnetic resonance imaging; texture
analysis; voxel-based morphometry
Introduction
Autism Spectrum Conditions (ASC) are neurodevelopmental syndromes char-
acterized by impairments in reciprocal social interaction, abnormal language use
and development and a restricted and repetitive range of behaviors and
interests.1 Etiologically, ASC appear to be multigenetic conditions emerging
from common genetic variants and rare mutations, each of small effect,2,3
combining to result in transcriptional dysregulation.4 Early environmental
factors (including intrauterine exposure to infection, hormones, medications,
metals, alcohol and smoking) are also linked to ASC, though broad epidemio-
logical surveys remain necessary to quantify the relative impact of these factors.5

One influential model proposes that genetic and environ-
mental factors ultimately shape aberrant patterns of con-
nectivity within and between regions during brain
development, such that ASC may be conceptualized as
‘disconnection syndromes’.6 Asperger syndrome (AS) is an
increasingly recognized, distinct type of ASC7 that shares
most of the clinical features of classical autism, but lacks
significant language delay.1
Neuroimaging methods have been applied extensively to
the in vivo characterization of gross structural and functional
brain abnormalities in ASC. Despite some discrepant find-
ings among structural studies with magnetic resonance
imaging (MRI), a set of gross anatomical findings are
noteworthy: ASC are typically associated with generalized
enlargement of the cerebral hemispheres, including lateral
prefrontal cortices, cerebellum and caudate nucleus and a
relative volume reduction in the size of the corpus callosum,
midbrain and cerebellar vermis.8 Volume reductions are also
reported within the amygdala-hippocampal complex and
medial parietal regions (precuneus).9 Although these find-
ings represent the main results of meta-analytic studies in
adult data, it is worth mentioning that different results are
found in children with ASC (that is, larger amygdala in
toddlers with autism10). In addition to these regional
changes, a general pattern of frontal-posterior undercon-
nectivity and increased occipital connectivity is suggested
from both functional MRI and structural (DTI (diffusion
tensior imaging)) connectivity studies.11,12 These differences
in regional structural integrity and interregional connectiv-
ity are likely to reflect neurodevelopment abnormalities
present at a finer, microstructural, level. However to date,
evidence from complimentary post-mortem neuropatholo-
gical studies to substantiate these findings are limited. Those
few post-mortem studies that have been completed however
have identified a number of important neuropathological
findings. These include evidence of cortical dysgenesis,
decreased Purkinje cell density in the cerebellum, lower
neuronal cell number in the amygdala13,14 and abnormal
number and width of cortical minicolumns.15
Faced with the inherent limitation of neuropathological
studies to the investigation of brain pathology at the end of
life, a number of in-vivo MRI techniques sensitive to finer
brain structure have been developed. These included
sulcation and gyrification measures, which may be more
sensitive macroscopic indicators of the integrity of cortical
microscopic architectural organization.16 Abnormalites in
cortical thickness and brain surface morphology are also
increasingly reported in autism, including temporal and
parietal cortical thinning in young adult males with high
functioning autism,17 and differences in the cortical folding
of the left inferior frontal gyrus and parietal operculum in
children and adolescents with ASC.18 Here we hypothesized
that microstructural anomalies in AS will translate into
quantifiable differences in a related surface pattern on MRI
(‘Image Heterogeneity’) assessed with texture analysis (TA).
Additionally, we hypothesized that differences in these
texture measures will relate to regional differences in gray
matter volume, the location of which may provide further
Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
71
insight into the origin and functional networks impacted by
ASC.
To test these hypotheses, we used a novel approach
combining whole brain selective-scale TA (‘Image Hetero-
geneity’) and voxel-based morphometry (VBM) to quantify
variation in surface intensity and its influence on regional
brain volume in AS and control participants. TA combines
the quantification of gray-level patterns, pixel interrelation-
ships and spectral properties of an image. When applied to
radiological scans, TA provides additional information about
surface patterns that can assist radiological diagnosis.19
Within neurology and psychiatry, TA has been applied to
the clinical and research characterization of brain tumors,
neurodegenerative disorders, epilepsy, multiple sclerosis19
and schizophrenia.20,21 VBM is an extensively used auto-
mated approach for the unbiased evaluation of regional
variations in brain structure (typically volume)22 related to
experimental group membership or effects of continuous
variables, for example, age, on brain structure.
In the present study, we therefore tested for hypothesized
differences between AS and control participants in: (1)
Textural features of the brain (heterogeneity as an indicator
of microstructrual anomalies in AS) and (2) Effects of
textural variations (heterogeneity) on regional gray-matter
volume that may indicate specific involvement of particular
neural systems (functional networks) within the brain.
Methods
Participants
Fifty-four individuals were recruited for this study, 24 with AS
and 30 controls (Table 1). The patients were diagnosed by a
multidisciplinary team including neuropsychiatrists (HDC,
NAH), a clinical psychologist and a speech and language
therapist in accordance with DSM-IV-R criteria for AS.
Clinical diagnosis was validated using the Diagnostic Inter-
view for Social and Communication Disorders.23 Exclusion
criteria for the entire sample were: (a) a history of neurolo-
gical disorders, (b) physical brain abnormalities, (c) general
learning disability (mental retardation), (d) a documented
history of significant head injury, (e) a history of psychosis (f)
a recent history of substance abuse/ dependence.
Due to the high rates of depressive symptoms among
people with AS,24 and the general population,25 a history of
depressive episodes in ‘full remission’ (based on DSM-IV-TR
specifiers) at the time of examination was permissible.
Maintenance treatment for past depression, medication
status (present/absent) was recorded (Table 1). The NART
(National Adult Reading Test)26 was used as an index of
intellectual function. Written informed consent was
obtained in accordance with the Declaration of Helsinki
(1991), and the procedures were approved by the National
Hospital for Neurology and Neurosurgery (NHNN) Research
Ethics Committee for studies of patients.
Procedure
Structural MR images were acquired with a Siemens
Magnetom Avanto 1.5T. system (Siemens AG, Erlangen,
The Pharmacogenomics Journal
 Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
72

Table 1 Demographic and structural data

Asperger syndrome (AS) (n ¼ 24) Controls (n ¼ 30)

Males Females Males Females

Sex 13 11 15 15
Age (years) 32.8±10.8 32.9±7.3 28.5±7.8 32.1±7.9
NART 30.7±8.8 30.2±7.3 35.2±5.9 35.5±6.8
Handedness 12 RH 9 RH 14 RH 13 RH
Medication status- antidepressants (p) 3 9 1 2
Gray matter volume (l) 0.8±0.06 0.7±0.04 0.8±0.06 0.68±0.03
Abbreviations: l, liters; NART, National Adult Reading Test; p, present; RH, right handed.

Germany). A sagittal 3D T1-weighted magnetisation-pre- Briefly, the TA comprises two stages:

pared rapid-acquisition gradient echo sequence (MPRAGE)
(parameters: 192 slices, TR (repetition time) ¼ 1160 ms, TE
(echo delay time) ¼ 4 ms, TI ¼ 600 ms, 0.9 mm isotropic
voxels) was used for acquisition of anatomical images. In
order to exclude gross brain pathology, T2-weighted images
were also acquired, by means of a coronal turbo spin-echo
sequence, having TR ¼ 6500 ms, TE ¼ 120 ms, 260  183 mm
FoV, 4.4 mm slice thickness.
Analysis of structural images
All T1- and T2-weighted images were reviewed by a
neuroradiologist to exclude macroscopic pathology and
white matter hyper-intensities or ventricular/sulcal enlarge-
ment beyond the level expected for age. All participants had
macroscopically normal imaging findings, and none had to
be excluded. Gray and white matter images, obtained by
automated segmentation implemented in SPM8 (Wellcome
Trust Centre for Neuroimaging, London, UK), were normal-
ized to the corresponding templates generated from all
participants using a diffeomorphic registration algorithm
(DARTEL).27 This nonlinear warping technique minimizes
inter-individual structural variance and thereby improves
the sensitivity of VBM analysis.28,29 A modulation step was
included to preserve voxel-wise information about local
tissue volume.27 Resulting tissue volume maps were
smoothed with an 8 mm FWHM Gaussian kernel and
entered into voxel-wise multiple regression analyses.
Individual total volumes of gray, white matter and CSF were
also computed. This processed MRI data set was also used in
a previous study (VBM and DTI) accepted for publication.30
For the present study, only results based on gray matter
images are reported.
Texture analysis
Segmented gray matter images from each participant were
converted into DICOM files using the ‘xMedcon’ medical
image conversion toolkit31 and individually analyzed with a
dedicated program written in MATLAB. This application
represents an extension of a TA methodology initially
developed to examine textural properties of tissues visua-
lized by medical imaging techniques.32 The method is
extensively described in a previous study on a psychiatric
population (schizophrenia).20
1. Whole brain filtration with a two-dimensional (2-D) LoG
(Laplacian of Gaussian) filter, which performs texture
extraction from individual MR image data comprising the
whole brain volume resulting in brain texture features of
different spatial frequency scales (feature width) corre-
sponding to filter values: fine (filter value ¼ 0.5, 2 pixels,
approx. filter width ¼ 2.0 mm), medium (filter value ¼
1, 4 pixels, approx. filter width ¼ 4.0 mm) and coarse
(filter value ¼ 1.5, 6 pixels, approx. filter width ¼
6.0 mm) (Figure 1).
2. Quantification of texture within the whole brain (compris-
ing individual MR image data) by using three statistical
and spectral parameters: mean gray-level intensity (MGL);
entropy (E), which indicates intensity and inhomogeneity
(irregularity); uniformity (U), which assesses the distribu-
tion of gray level. This process was performed on filtered
and unfiltered MRI-segmented whole brain gray matter.
These parameters are defined below:
1 X
Mean GrayLevel Intensity ðMGLÞ ¼ ½aðx; yÞ ð1Þ
N ðx;yÞ2R
X
k
Entropy ðEÞ ¼  ½pðlÞ log2 ½pðlÞ ð2Þ
l¼1
X
k
Uniformity ðUÞ ¼ ½pðlÞ2 ð3Þ
l¼1
where N is the total number of pixels in the region of
interest R within the whole brain a(x,y), l is the gray level
(l ¼ 1 to k for a k gray-level image) in R and p(l) is the
probability of the occurrence of the gray level l in R.
Without filtration, the above three parameters describe
the intensity distribution, whereas with the application of
the LoG filter they describe the distribution of gradient
magnitudes over different spatial frequency ranges. These
parameters represent quantitative metrics of features per-
ceived visually as texture.
Statistical analyses
Demographic data (age, NART) and global gray matter
volume were analyzed within a two-way ANOVA model

The Pharmacogenomics Journal


Figure 1 Three-dimensional (3-D) volume-rendered left sagittal view
of (a) conventional magnetic resonance whole-brain gray matter and
corresponding (b) fine, (c) medium and (d) coarse texture (intensity-
and gradient-based) in an Asperger syndrome female patient.
with diagnosis and sex as between-participants factors.
w2 tests were used for categorical variables (sex, handedness,
medication status). Four sets of analyses were performed to
investigate the effects of diagnosis on the texture parameters
(TPs):
1. Examination of the relationship between the variables of
interest (TPs) and potential interacting (confounding)
variables (that is, gray matter global volume, age,
medication status).
2. Group comparisons with diagnosis and sex as between-
participant factors (independent variables).
3. Data reduction technique (principal component analysis)
to generate independent textural regressors for the
structural imaging analysis.
4. ROC (Receiver Operating Characteristics) analysis to assess
the discriminatory ability of the most significant TPs.
All the above analyses were performed with the statistical
software SPSS 17.0 (SPSS, Chicago, IL, USA), MedCalc for
Windows, version 9.2.0.0 (MedCalc Software, Mariakerke,
Belgium) and MATLAB 7.5.0, The MathWorks, Natick, MA,
USA, 2007).
1. Bivariate correlations explored the associations between
global gray matter volumes, age and the twelve TPs (means
of gray-level intensities (MGLs), entropy (E), uniformity (U)
for unfiltered (unf) and filtered data- fine 0.5, medium 1,
coarse 1.5), separately in the controls and AS group. Results
revealed associations between age and filtered entropies and
uniformities in controls, but not in AS group; also, gray
matter volume was associated with unfiltered and filtered
MGLs in controls, but not in AS (see Supplementary Table 1).
Gray matter global volume was subsequently controlled for
in statistical models containing MGLs as variables of
interest. Similarly, age was controlled for in statistical
Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
73
models with entropies and uniformities as variables of
interest.
As w2 analyses revealed statistically significant differences
in medication status between the groups (see Results), linear
regression using this parameter as a predictive variable was
undertaken to check for potential effects on TPs. This
analysis revealed no significant effects of medication status
on any TP (data available upon request); consequently this
measure was not considered a major confounding factor.
2. (a) Univariate general linear models with MGLs as
dependent variables, diagnosis and sex as between-partici-
pant factors and gray matter global volume as a covariate
were performed after testing for homogeneity of slopes
(factor  covariate interaction). As the presence of signifi-
cant sex  gray matter volume interactions were found for
all MGLs, effects of diagnosis were tested separately within
each sex for these TPs.
(b) Similarly, we performed univariate analyses using
general linear models with entropy and uniformity as
dependent variables, diagnosis and sex as between-partici-
pant factors and age as a covariate (after testing for
homogeneity of slopes; factor  covariate interaction). Sig-
nificant slope heterogeneity (diagnosis-by-age interaction
terms) was found for three parameters: entropies medium
and coarse and uniformity coarse. For these parameters, we
examined differences among groups regression slopes using
the Wilcox modification of the Johnson–Neyman proce-
dure, which used Games–Howell test set at Po0.05 (Supple-
mentary Methods and Results).
3. Due to extensive correlations between TPs (Supplemen-
tary Table 1), a principal component analysis was performed
on these parameters using a varimax rotation and an
eigenvalue-based extraction criterion (eigenvalue 41). This
generated three factor scores per subject (see Results),
expressing the majority of independent variance captured
across our measures of brain texture. These scores were then
used within voxel-wise analysis of regional gray matter
volume to highlight where texture features were associated
with differences in localized volume between AS and
controls.
4. The ROC, including area under the curve (AUC), its
significance, sensitivity and specificity values were calcu-
lated for TPs found to differ significantly between AS
participants and controls.
Statistical analysis on imaging data
Normalized and smoothed gray matter images were ana-
lyzed using the General Linear Model as implemented in
SPM8 (Wellcome Trust Centre for Neuroimaging). Three
multiple regression models each with one of the three TP
factors as variables of interest and diagnosis, sex, age and
global gray matter volumes as nuisance covariates were used
to determine which gray matter regions were most sensitive
to differences in individual TP factors. A diagnosis-by-age
interaction was also modeled within the multiple regression
analysis, to test whether this interaction overlap with the
effects of TPs on gray matter volume.
The Pharmacogenomics Journal
 Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
74
Statistical inference was based on voxel-by-voxel T-tests at
the whole brain level. FWE (Family-wise error) rate estima-
tions were used to correct for multiple comparisons across
all voxels and results reported at a stringent FWE threshold
of Po0.05. Statistical parametric maps are displayed on the
customized template generated from all participants.
Finally, regional gray matter volumes were computed for
each subject from eigenvariates extracted from 8 mm radius
spheres centered on maximal voxels within all significant
clusters. These volume measures were entered as dependent
variables into regression analyses conducted in SPSS 17.0.
Age, diagnosis, sex and factor scores were regressed on the
regional gray matter volumes to investigate which of these
regressors better predict the extracted regional gray matter
volumes.
Results
The participants were matched by age (no significant
diagnosis, sex or diagnosis  sex effects on age, P ¼ 0.28,
P ¼ 0.43, P ¼ 0.46, respectively), by sex (w2 ¼ 0.093, P ¼ 0.761)
and by handedness (w2 ¼ 0.084, P ¼ 0.771). Table 1 displays
demographic and structural characteristics of the samples.
As expected, a main effect of sex on gray matter volume was
observed (FoM; F(1,50) ¼ 56.8, Po0.001). No diagnosis or
diagnosis  sex interaction effects were observed on total
gray matter volume (P ¼ 0.832, P ¼ 0.085, respectively). A
main effect of diagnosis on NART score was observed (AS o
Controls; F(1,50) ¼ 6.2, P ¼ 0.016). Medication status signifi-
cantly differed by sex (w2 ¼ 5.3, P ¼ 0.022) and by diagnosis
(w2 ¼ 10.63, P ¼ 0.001).
Results of the statistical analysis on TPs
Unfiltered textural parameters were significantly influenced
by diagnosis, with AS participants having significantly higher
Figure 2 Receiver Operating Characteristics (ROC) analysis results. Two texture parameters relatively discriminated between Asperger syndrome
and controls. AUC, area under the curve.
The Pharmacogenomics Journal
‘textural entropy’ compared with controls (mean±s.d. AS:
6.1±0.14, mean±s.d. controls: 5.96±0.15, F(1,49) ¼ 6.9,
P ¼ 0.011, Z2 ¼ 0.124).
Main effects of diagnosis and sex were observed on the
unfiltered uniformity: The control group had higher uni-
formity than AS (mean±s.d. AS: 0.045±0.006, mean±s.d.
controls: 0.049±0.008; F(1,49) ¼ 7.49, P ¼ 0.009, Z2 ¼ 0.133);
the female group had reduced uniformity compared with
males (mean±s.d. females: 0.045±0.007, mean±s.d. males:
0.05±0.008; F(1,49) ¼ 4.46, P ¼ 0.04, Z2 ¼ 0.083). These main
effects were qualified by an interaction diagnosis-by-sex:
F(1,49) ¼ 4.58; P ¼ 0.037; Z2 ¼ 0.086 (Supplementary Figure 1).
Diagnosis also influenced filtered TPs, respectively,
medium and coarse entropy. However, for these measures
we found significant slope heterogeneity (diagnosis-by-age
interaction; Supplementary Methods and Results) suggest-
ing that AS differ in these TPs in a manner dependent
on age values: for E medium and E coarse, the AS group
had higher values than controls within 19–45 years,
respectively, 25–51.7 years (Wilcox (J–N) procedure; for
methodological details see ‘Supplementary Methods and
Results’).
Complete results for diagnosis and sex effects on entropies
and uniformities are presented in Supplementary Table 2
(‘Supplementary Methods and Results’). No diagnosis effect
was observed on MGLs, for either males or females (data not
shown).
Principal component analysis on unfiltered and filtered
TPs yielded a three-factor solution (Supplementary Tables 3
and 4 and Supplementary Figure 2): Factor 1 (predominantly
filtered entropies and uniformities) explained 44.69% of the
total variance, factor 2 (mainly composed of MGLs) 32.89%
and factor 3 (unfiltered entropy and uniformity) 11.86%.
Factor scores were computed and used in the VBM analysis
as described below.
 Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
75

ROC analysis results
Two TPs also predicted patients with AS: E unfiltered
(criterion 45.91, sensitivity 91.67%, specificity 50%), U
unfiltered (criterion p0.049, sensitivity 83.33%, specificity
56.67%). Area under the curve and the significance levels are
presented in Figure 2.
VBM results
Gray matter volume differences were found in basal ganglia
and cerebellum based on factor 1 and factor 2 scores,
respectively. Specifically, factor 1 score predicted the varia-
tion of gray matter volume in the right caudate (MNI: 13 14
6; Z score ¼ 4.38; Po0.05) and left caudate (MNI: 13 14
5; Z score ¼ 4.23, Po0.05) (Figure 3). Factor 2 score
predicted the gray matter volume variation in cerebellum
(uvula) (MNI: 14 75 43, Z score ¼ 4.17, Po0.05) (Figure 4).
In AS group only, age affected gray matter volume within
bilateral parahippocampal gyrus (uncus): (MNI: 28 7
32; Z score ¼ 4.33, Po0.05; respectively, MNI: 23 4 33;
Z score ¼ 4.32, Po0.05).
The dependence of caudate volumes on the factor 1 score,
sex and age was examined using regression analysis. In the
AS group only, the significant effects were observed for the
three predictors, with the highest variance explained by the
factor 1 score (for the right caudate: partial correlation
coefficient 0.649, P ¼ 0.001, respectively, for the left
caudate: partial correlation coefficient 0.626, P ¼ 0.002;
see Supplementary Tables 5 and 6 for complete results, Figure
3 plotting the regression slopes in both groups). Similarly,
right cerebellum (uvula) volume showed a factor 2 score
effect (partial correlation coefficient 0.54, P ¼ 0.01; Sup-
plementary Table 6E, Figure 4), whereas bilateral parahippo-
campal gyri (uncus) showed only age and sex effects within
the AS group, but no factor 1 or factor 2 effects (Supple-
mentary Tables 6C, 6D). There was no significant correlation
between factor 3 score and any regional gray matter volume
at corrected threshold significance.
Discussion
This study provides the first evidence for abnormalities
(increased heterogeneity) in MRI brain texture in AS.
Further, by combining textural analysis with VBM we show
that MRI texture variations predicted the gray matter
volume of two brain structures (caudate and cerebellum)
frequently shown to be abnormal in ASC in neuropatholo-
gical studies. In particular, AS participants demonstrated
significantly greater entropy and significantly lower textural

Figure 3 Top panel: Effect of factor 1 score on gray matter volume of caudate nuclei. Statistical Parametric Maps are overlaid on subjects’
template (figure created with MRIcron62). Color bars: T score from the voxel-based morphometry (VBM) analysis; ‘x, y, z’: MNI coordinates.
Bottom panel: Scatter plots showing factor 1 by diagnosis group interaction on caudate volumes. Left: only within Asperger syndrome (AS)
group factor 1 predicted the volume of right caudate. Right: only within AS group factor 1 predicted the volume of left caudate. HC, healthy
controls.

The Pharmacogenomics Journal

 Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
76

Figure 4 Effect of factor 2 score on gray matter volume in the cerebellum. Statistical Parametric Maps are overlaid on subjects’ template (figure
created with MRIcron62). Color bars: T score from the voxel-based morphometry (VBM) analysis; ‘x, y, z’: MNI coordinates. Scatter plot showing
factor 2 by diagnosis group interaction on cerebellum. Results show an effect of factor 2 in Asperger syndrome (AS) group. HC, healthy controls;
R, right.

uniformity than controls. Both of these inversely correlated
textural features also predicted caudate gray matter volume.
Another TP, mean gray level additionally predicted cere-
bellar gray matter volume particularly within the vermis/
uvula. Interestingly, detailed exploration of these relation-
ships revealed that sensitivity of both caudate and cerebel-
lar volume to textural parameters was limited to the AS
group. Notably, although age and sex effects were also
demonstrated on the caudate gray matter volumes, the
highest proportion of variance was explained by entropy
and uniformity.
Diagnosis differences in some TPs (that is, entropy
medium and coarse) were mediated by age. Furthermore,
an age-by-diagnosis interaction predicted the gray matter
volume of bilateral parahippocampal gyri/uncus in AS
group. These results extend at clinical level previous
findings, which demonstrated heterogeneity in the devel-
opmental trajectories of subcortical structures in normative
population.33 Notably, textural parameters did not predict
the volumes of parahippocampal gyri, suggesting selective
associations with gray matter within ASC, which deserve
further exploration.
Two TPs namely: unfiltered uniformity and unfiltered
entropy additionally discriminated AS participants and
controls. These findings extend the set of in vivo structural
MRI analytic techniques more sensitive to neuropathol-
ogy.16 Other techniques in this class, for example, measures
of sulcation/gyrification and cortical thickness, quantify the
integrity of cortical features that emerge from microscopic
cellular organization. In distinction, textural analysis may
be more directly sensitive to histopathological variations,
which offers the potential to characterize and classify
macroscopic lesions across a broad range of neurological
disorders (that is, multiple sclerosis, brain tumors, mesial
temporal lobe epilepsy and traumatic brain injury).34–39 The
more diffuse and microscopic brain anomalies characteristic
of mental disorders require particularly fine TA methods in
order to ensure adequate sensitivity. The recent expansion of
fully automated algorithms, complex statistical methods of
feature extraction, high-resolution MR imaging and more

The Pharmacogenomics Journal


accurate segmentation techniques16 have converged to
enable identification of brain texture differences between
psychiatric populations (that is, schizophrenia) and healthy
controls.20,21 These previous results and the technical
refinements suggest that TA is a promising method for
generating and testing biomarkers for mental disorders.
Specifically, our TA technique utilizes a circularly symmetric
isotropic filter (LoG) to specifically enhance features
corresponding to a particular scale and variation in signal
intensity within the MR volume.20
A challenging question still remaining is what tissue
characteristics lay beneath each of the identified textural
parameters. We speculate that organization of cells and
neuropil composing the gray matter aggregate predomi-
nantly drive textural features expressed as MGL, inhomo-
geneity and distribution of the ‘gray’ on MR images.
Striatum/caudate architecture described as a mosaic of
predominantly medium spiny neurons is organized in a
patch-matrix model with compartments segregated based
on input-output connections.40 Direct post-mortem data
concerning architectural abnormalities of caudate is almost
lacking in ASC;13 however genetic animal models of autism
(that is, Shank3 mutant mouse) suggest a possible neuronal
hypertrophy and disorganization of dendritic arborization,
associated with striatal volumetric enlargement.41 MRI
studies in humans have documented an increase in caudate
volume in ASC, which is unrelated to psychotropic medica-
tion8,42 and conceivably attributable to neuronal hypertro-
phy and disorganization of local connections. Studies
showing a linear increase in caudate volume with age in
ASC (in contrast to the linear decrease see in typically
developing controls) arguably reflect a developmental
anomaly that continues into adult life.43,44 This continuing
increase in caudate size well into adult life appears distinct
from the more general brain volume trajectory observed in
ASC, which is characterized by an accelerated rate of
increase during childhood then premature arrest during
adolescence.45,46 In this context, it is tempting to speculate
that our results of abnormal entropy and uniformity
predicting caudate volume indicate an underlying disorga-
nized architecture with possibly maldistributed neurons and
disrupted local connectivity within the caudate volume.
However, future studies are necessary to explore the textural
characteristics of basal ganglia in normative population and
ASC. An altered texture impacting on the caudate volume
also supports the hypothesis of disrupted fronto-striatal
connectivity as fronto-striatal connections are topographi-
cally distributed within the striatum/ caudate.47
Cerebellum is characterized by a complex gross morphol-
ogy, a homogeneous three-layer cortex48 and has one of the
highest growth rates among brain structures.49,50 It is also
extremely vulnerable to insults during development, as has
been shown in studies on pre-term51 and term infants.52
Interestingly, various cerebellar malformations, including
vermal hypoplasia of genetic and environmental origin, are
also accompanied by autistic-like syndromes.51,52 The
trajectory of cerebellum size in autism appears to conform
with that seen in overall brain development, characterized
Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
77
by overgrowth during childhood, then premature arrest and
possible degeneration in adolescence.45 This is supported by
studies that have shown an increase in cerebellum size in
ASC children compared with neurotypical controls53 (how-
ever see ref. 54); but similar or even smaller cerebellum size
in ASC adults.55,56 Post-mortem studies have also shown
cerebellar pathology in ASC, in particular, a reduction in
Purkinje cell density predominantly in the cerebellar hemi-
spheres.13,57 These observations are relevant for interpreting
our results. For example, though our adult AS group did not
show differences in gross cerebellum volume, inter-indivi-
dual differences in volume within the ASC group were
sensitive to a texture factor composed of filtered and
unfiltered mean gray level.
Speculatively, although the cerebellum MGL (unfiltered)
is likely to be determined by the total quantity of gray
matter the gradients of intensity (filtered means) may well
be more sensitive to its organization into foliae separated by
fissures. From this viewpoint, our results indicate a possible
subtle perturbation of cerebellar architecture despite no
overall difference in volume with functional consequences
specific for ASC.
It has been proposed that the functional framework of the
cerebellum is supported by a precise and genetically
regulated structural organization along two coordinate
systems: lobules in the anterior-posterior axis and para-
sagittal molecular domains in the medial-lateral axis.48
Genetic animal58 and post-mortem human studies59 reveal
alterations in cerebellar architecture that are potentially
informative for autistic-like behavior. These alterations
include: a loss of intercrural fissure, impaired arborization58
and misplacement of Purkinje cells, decreased foliar width
and mild conformational changes in internal granular
layer.59 Overall, our results of gray matter texture differences
and the preferential relationship with volume of caudate
and cerebellum raise the appealing opportunity of using
textural features as potential biomarkers of ASC.
Nevertheless, some limitations should be highlighted: (a)
the diagnosis effects seem to be partially intricate with age
and sex effects; (b) also, the relationship between textural
parameters and regional gray matter volumes are yet indirect.
Therefore, a cautionary note is necessary concerning the
association with the subjacent biology, until future studies
provide specific textural profiles of various brain structures.
The need of further validation notwithstanding, several
directions seem promising:
(1) Establishing the trait- or state-like character of textural
ASC signatures. Our results confining the relationship
between texture and regional gray matter volume to the
AS group suggest a state-like characteristic. However, a
thorough examination of brain textural features across
ASC phenotypes compared with both neurotypical
controls and non-affected relatives may reveal differ-
ences with trait-like (endophenotype) characteristics.
(2) Establishment of textural features as state-markers
would then offer the opportunity to use them for testing
current and future therapies in ASC.
The Pharmacogenomics Journal
 Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
78
(3) Following recent breakthroughs in the use of surface
pattern variations as classifiers,60,61 textural features
could then also be used for re-defining ASC phenotypes
on the basis of biological rather than clinical features.
Our initial results from an ROC analysis indicate that
textural features, such as entropy and uniformity, are
highly sensitive for detecting AS albeit with a lower
specificity in discriminating AS participants. Such a
pattern of accuracy suggests potential utility in the
initial steps of patients’ screening with confirmation
based on subsequent pattern classifier approaches.
In conclusion, our study shows the utility of a combination
of TA and VBM as a novel in vivo approach for identifying
structural brain differences in ASD and helps bridge the gap
between classical MR imaging and neuropathological ap-
proaches. It also provides further evidence for pathology in
both the cerebellum and caudate in these conditions.
Conflict of interest
Only ER, LM, FDCCB, MAG, NAH, HDC had full access and
control of the data and have no conflict of interest to report.
BG, CC, RCDY provided TA software (described in the
manuscript) and have a commercial interest in the im-
plementation of this textural analysis software in oncology-
related applications. There are no other author disclosures.
Acknowledgments
ER and partly HDC are supported by a donation from the ‘Dr
Mortimer and Theresa Sackler Foundation’. HDC was supported by a
Wellcome Trust Programme Grant and LM was supported by the
Shirley Foundation during the acquisition of this data. Dr Nicholas
Dowell’s assistance with imaging processing is acknowledged.
References
1 American Psychiatric Association. Diagnostic and Statistical Manual of
Mental Disorders, Revised 4th edn. American Psychiatric Association:
Washington, DC, 2000.
2 Geschwind DH. Autism: many genes, common pathways? Cell 2008;
135: 391–395.
3 Abrahams BS, Geschwind DH. Connecting genes to brain in the autism
spectrum disorders. Arch Neurol 2010; 67: 395–399.
4 Voineagu I, Wang X, Johnston P, Lowe JK, Tian Y, Horvath S et al.
Transcriptomic analysis of autistic brain reveals convergent molecular
pathology. Nature 2011; 474: 380–384.
5 Newschaffer CJ, Croen LA, Daniels J, Giarelli E, Grether JK, Levy SE et al.
The epidemiology of autism spectrum disorders. Annu Rev Publ Health
2007; 28: 235–258.
6 Geschwind DH, Levitt P. Autism spectrum disorders: developmental
disconnection syndromes. Curr Opin Neurobiol 2007; 17: 103–111.
7 Woodbury-Smith MR, Volkmar FR. Asperger syndrome. Eur Child Adolesc
Psychiatry 2009; 18: 2–11.
8 Stanfield AC, McIntosh AM, Spencer MD, Philip R, Gaur S, Lawrie SM.
Towards a neuroanatomy of autism: a systematic review and meta-
analysis of structural magnetic resonance imaging studies. Eur Psychiatry
2008; 23: 289–299.
9 Via E, Radua J, Cardoner N, Happe F, Mataix-Cols D. Meta-analysis of
gray matter abnormalities in autism spectrum disorder: should Asperger
disorder be subsumed under a broader umbrella of autistic spectrum
disorder? Arch Gen Psychiatry 2011; 68: 409–418.
The Pharmacogenomics Journal
10 Schumann CM, Barnes CC, Lord C, Courchesne E. Amygdala enlarge-
ment in toddlers with autism related to severity of social and
communication impairments. Biol Psychiatry 2009; 66: 942–949.
11 Verhoeven JS, De Cock P, Lagae L, Sunaert S. Neuroimaging of autism.
Neuroradiology 2010; 52: 3–14.
12 Minshew NJ, Keller TA. The nature of brain dysfunction in autism:
functional brain imaging studies. Curr Opin Neurol 2010; 23: 124–130.
13 Amaral DG, Schumann CM, Nordahl CW. Neuroanatomy of autism.
Trends Neurosci 2008; 31: 137–145.
14 Palmen SJ, van Engeland H, Hof PR, Schmitz C. Neuropathological
findings in autism. Brain 2004; 127(Part 12): 2572–2583.
15 Casanova MF, Buxhoeveden DP, Switala AE, Roy E. Neuronal density
and architecture (Gray Level Index) in the brains of autistic patients.
J Child Neurol 2002; 17: 515–521.
16 Mangin JF, Jouvent E, Cachia A. In-vivo measurement of cortical
morphology: means and meanings. Curr Opin Neurol 2010; 23: 359–367.
17 Wallace GL, Dankner N, Kenworthy L, Giedd JN, Martin A. Age-related
temporal and parietal cortical thinning in autism spectrum disorders.
Brain 2010; 133(Part 12): 3745–3754.
18 Nordahl CW, Dierker D, Mostafavi I, Schumann CM, Rivera SM, Amaral
DG et al. Cortical folding abnormalities in autism revealed by surface-
based morphometry. J Neurosci 2007; 27: 11725–11735.
19 Kassner A, Thornhill RE. Texture analysis: a review of neurologic MR
imaging applications. AJNR Am J Neuroradiol 2010; 31: 809–816.
20 Ganeshan B, Miles KA, Young RC, Chatwin CR, Gurling HM, Critchley
HD. Three-dimensional textural analysis of brain images reveals
distributed grey-matter abnormalities in schizophrenia. Eur Radiol
2009; 20: 941–948.
21 Kovalev VA, Petrou M, Suckling J. Detection of structural differences
between the brains of schizophrenic patients and controls. Psychiatry
Res 2003; 124: 177–189.
22 Ashburner JT, Friston KJ. Voxel-based morphometry. In: Friston KJ,
Ashburner JT, Kiebel SJ, Nichols TE, Penny WD (eds). Statistical
Parametric Mapping. The Analysis of Functional Brain Images, 2007,
p 92. Academic Press, Elsevier; http://books.elsevier.com.
23 Leekam SR, Libby SJ, Wing L, Gould J, Taylor C. The Diagnostic
Interview for Social and Communication Disorders: algorithms for ICD-
10 childhood autism and Wing and Gould autistic spectrum disorder.
J Child Psychol Psychiatry 2002; 43: 327–342.
24 Lugnegard T, Hallerback MU, Gillberg C. Psychiatric comorbidity in
young adults with a clinical diagnosis of Asperger syndrome. Res Dev
Disabil 2011; 32: 1910–1917.
25 Hardeveld F, Spijker J, De Graaf R, Nolen WA, Beekman AT. Prevalence
and predictors of recurrence of major depressive disorder in the adult
population. Acta psychiatrica Scandinavica 2009; 122: 184–191.
26 Nelson HE, O’Connell A. Dementia: the estimation of premorbid
intelligence levels using the New Adult Reading Test. Cortex 1978; 14:
234–244.
27 Ashburner J, Friston KJ. Computing average shaped tissue probability
templates. NeuroImage 2009; 45: 333–341.
28 Bergouignan L, Chupin M, Czechowska Y, Kinkingnehun S, Lemogne C,
Le Bastard G et al. Can voxel based morphometry, manual segmenta-
tion and automated segmentation equally detect hippocampal volume
differences in acute depression? NeuroImage 2009; 45: 29–37.
29 Yassa MA, Stark CE. A quantitative evaluation of cross-participant
registration techniques for MRI studies of the medial temporal lobe.
NeuroImage 2009; 44: 319–327.
30 Beacher FD, Minati L, Baron-Cohen S, Lombardo MV, Gray MA,
Harrison NA et al. Autism attenuates sex differences in brain structure: a
combined voxel-based morphometry (VBM) and diffusion-tensor
imaging (DTI) study. Am J Neuroradiol 2012; 33: 83–89.
31 Nolf E. ‘XMedCon - An open-source medical image conversion toolkit’.
Eur J Nucl Med 2003; 30(Suppl 2): S246; (TP39).
32 Ganeshan B, Miles KA, Young RC, Chatwin CR. Three-dimensional
selective-scale texture analysis of computed tomography pulmonary
angiograms. Invest Radiol 2008; 43: 382–394.
33 Ostby Y, Tamnes CK, Fjell AM, Westlye LT, Due-Tonnessen P, Walhovd
KB. Heterogeneity in subcortical brain development: a structural
magnetic resonance imaging study of brain maturation from 8 to 30
years. J Neurosci 2009; 29: 11772–11782.
34 Theocharakis P, Glotsos D, Kalatzis I, Kostopoulos S, Georgiadis P, Sifaki
K et al. Pattern recognition system for the discrimination of multiple

sclerosis from cerebral microangiopathy lesions based on texture
analysis of magnetic resonance images. Magn Reson Imaging 2009;
27: 417–422.
35 Drabycz S, Roldan G, de Robles P, Adler D, McIntyre JB, Magliocco AM
et al. An analysis of image texture, tumor location, and MGMT
promoter methylation in glioblastoma using magnetic resonance
imaging. NeuroImage 2010; 49: 1398–1405.
36 Brown R, Zlatescu M, Sijben A, Roldan G, Easaw J, Forsyth P et al. The
use of magnetic resonance imaging to noninvasively detect genetic
signatures in oligodendroglioma. Clin Cancer Res 2008; 14: 2357–2362.
37 Juntu J, Sijbers J, De Backer S, Rajan J, Van Dyck D. Machine learning
study of several classifiers trained with texture analysis features to
differentiate benign from malignant soft-tissue tumors in T1-MRI
images. J Magn Reson Imaging 2010; 31: 680–689.
38 Jafari-Khouzani K, Elisevich K, Patel S, Smith B, Soltanian-Zadeh H. FLAIR
signal and texture analysis for lateralizing mesial temporal lobe epilepsy.
NeuroImage 2010; 49: 1559–1571.
39 Holli KK, Harrison L, Dastidar P, Waljas M, Liimatainen S, Luukkaala T
et al. Texture analysis of MR images of patients with mild traumatic
brain injury. BMC Med Imaging 2010; 10: 8.
40 Gerfen CR. The neostriatal mosaic: multiple levels of compartmental
organization in the basal ganglia. Annu Rev Neurosci 1992; 15: 285–320.
41 Peca J, Feliciano C, Ting JT, Wang W, Wells MF, Venkatraman TN et al.
Shank3 mutant mice display autistic-like behaviours and striatal
dysfunction. Nature 2011; 472: 437–442.
42 Haznedar MM, Buchsbaum MS, Hazlett EA, LiCalzi EM, Cartwright C,
Hollander E. Volumetric analysis and three-dimensional glucose meta-
bolic mapping of the striatum and thalamus in patients with autism
spectrum disorders. Am J Psychiatry 2006; 163: 1252–1263.
43 McAlonan GM, Daly E, Kumari V, Critchley HD, van Amelsvoort T,
Suckling J et al. Brain anatomy and sensorimotor gating in Asperger’s
syndrome. Brain 2002; 125(Part 7): 1594–1606.
44 Langen M, Schnack HG, Nederveen H, Bos D, Lahuis BE, de Jonge MV
et al. Changes in the developmental trajectories of striatum in autism.
Biol Psychiatry 2009; 66: 327–333.
45 Courchesne E, Pierce K, Schumann CM, Redcay E, Buckwalter JA,
Kennedy DP et al. Mapping early brain development in autism. Neuron
2007; 56: 399–413.
46 Redcay E, Courchesne E. When is the brain enlarged in autism? A meta-
analysis of all brain size reports. Biol Psychiatry 2005; 58: 1–9.
47 Draganski B, Kherif F, Kloppel S, Cook PA, Alexander DC, Parker GJ et al.
Evidence for segregated and integrative connectivity patterns in the
human Basal Ganglia. J Neurosci 2008; 28: 7143–7152.
48 Sillitoe RV, Joyner AL. Morphology, molecular codes, and circuitry
produce the three-dimensional complexity of the cerebellum. Annu Rev
Cell Dev Biol 2007; 23: 549–577.
Supplementary Information accompanies the paper on the The Pharmacogenomics Journal website (http://www.nature.com/tpj)
Gray matter textural heterogeneity in Asperger syndrome
E Radulescu et al
79
49 Volpe JJ. Cerebellum of the premature infant: rapidly developing,
vulnerable, clinically important. J Child Neurol 2009; 24: 1085–1104.
50 Limperopoulos C, Robertson RL, Sullivan NR, Bassan H, du Plessis AJ.
Cerebellar injury in term infants: clinical characteristics, magnetic
resonance imaging findings, and outcome. Pediatr Neurol 2009;
41: 1–8.
51 Knickmeyer RC, Gouttard S, Kang C, Evans D, Wilber K, Smith JK et al. A
structural MRI study of human brain development from birth to 2 years.
J Neurosci 2008; 28: 12176–12182.
52 Bolduc ME, Du Plessis AJ, Sullivan N, Khwaja OS, Zhang X, Barnes K
et al. Spectrum of neurodevelopmental disabilities in children with
cerebellar malformations. Dev Med Child Neurol 2011; 53: 409–416.
53 Palmen SJ, Hulshoff Pol HE, Kemner C, Schnack HG, Durston S, Lahuis
BE et al. Increased gray-matter volume in medication-naive high-
functioning children with autism spectrum disorder. Psychol Med 2005;
35: 561–570.
54 Scott JA, Schumann CM, Goodlin-Jones BL, Amaral DG. A comprehen-
sive volumetric analysis of the cerebellum in children and adolescents
with autism spectrum disorder. Autism Res 2009; 2: 246–257.
55 Hallahan B, Daly EM, McAlonan G, Loth E, Toal F, O’Brien F et al.
Brain morphometry volume in autistic spectrum disorder: a
magnetic resonance imaging study of adults. Psychol Med 2009; 39:
337–346.
56 Rojas DC, Peterson E, Winterrowd E, Reite ML, Rogers SJ, Tregellas JR.
Regional gray matter volumetric changes in autism associated with
social and repetitive behavior symptoms. BMC Psychiatry 2006; 6: 56.
57 Bauman ML, Kemper TL. Neuroanatomic observations of the brain
in autism: a review and future directions. Int J Dev Neurosci 2005; 23:
183–187.
58 Sadakata T, Kakegawa W, Mizoguchi A, Washida M, Katoh-Semba R,
Shutoh F et al. Impaired cerebellar development and function in mice
lacking CAPS2, a protein involved in neurotrophin release. J Neurosci
2007; 27: 2472–2482.
59 Greco CM, Navarro CS, Hunsaker MR, Maezawa I, Shuler JF, Tassone F
et al. Neuropathologic features in the hippocampus and cerebellum of
three older men with fragile X syndrome. Mol Autism 2011; 2: 2.
60 Ecker C, Marquand A, Mourao-Miranda J, Johnston P, Daly EM,
Brammer MJ et al. Describing the brain in autism in five dimensions–
magnetic resonance imaging-assisted diagnosis of autism spectrum
disorder using a multiparameter classification approach. J Neurosci
2011; 30: 10612–10623.
61 Jiao Y, Chen R, Ke X, Chu K, Lu Z, Herskovits EH. Predictive models of
autism spectrum disorder based on brain regional cortical thickness.
NeuroImage 2011; 50: 589–599.
62 Rorden C, Brett M. Stereotaxic display of brain lesions. Behav Neurol
2000; 12: 191–200.
The Pharmacogenomics Journal