Directory UMM :Data Elmu:jurnal:B:Biological Psichatry:Vol48.Issue6.2000:
A Geometric Morphometric Assessment of Change in
Midline Brain Structural Shape following a First
Episode of Schizophrenia
Waleed S. Gharaibeh, F. James Rohlf, Dennis E. Slice, and Lynn E. DeLisi
Background: Previous reports indicate that brain structural abnormalities may be progressive in some patients
with schizophrenia. Our study was designed to determine
deviations in the shape of midline brain structures at the
time of onset of symptoms of schizophrenia and 3–5 years
later.
Methods: Eleven landmarks were located on the midsagittal magnetic resonance imagery brain scans of 55
patients with schizophrenia and 22 nonpsychiatric control
individuals. Geometric morphometric methods were used
for the extraction of shape variables from landmark
coordinates. Permutation tests were used to test the effects
of gender, diagnosis, time elapsed since illness onset, and
age on brain shape.
Results: The diagnosis-by-time interaction and the effect
of gender were significantly different from zero (p , .027
and p , .039, respectively). The effect of time was
significant in patients (p , .002), but not in control
subjects. Some anatomical abnormalities in mean patient
brain morphology seem to be present both at the time of
diagnosis and at follow-up. These are similar to anomalies
reported by previous geometric morphometrics studies.
Conclusions: Some previously identified brain abnormalities are detectable at the time of first hospitalization. The
rapid change in midline brain morphology in patients with
schizophrenia during the subsequent 3–5 years is consistent with either a neurodegenerative disease process or an
effect of treatment with psychiatric drugs. There is a
sexual dimorphism in brain morphology that might be
reduced by schizophrenia. Biol Psychiatry 2000;48:
398 – 405 © 2000 Society of Biological Psychiatry
Key Words: First-episode schizophrenia, MRI, brain
morphology, corpus callosum, Procrustes, thin-plate spline
From the Departments of Ecology and Evolution (WSG, FJR, DES) and Psychiatry
(LED), SUNY, Stony Brook, New York.
Address reprint requests to F. James Rohlf, SUNY, Department of Ecology and
Evolution, Stony Brook NY 11794-5245.
Received January 27, 2000; revised April 20, 2000; accepted May 2, 2000.
© 2000 Society of Biological Psychiatry
Introduction
O
ver the past 25 years, neuroimaging studies have
associated schizophrenia with a number of anatomic
brain abnormalities with varying degrees of reproducibility (for a review, see Pearlson and Marsh 1999); however,
only a few have investigated the timing of emergence of
these neuroanatomic abnormalities (for a review, see
DeLisi 1999). Establishing the temporal progress of pathology can be of value in testing competing hypotheses
for the etiology of schizophrenia. A model in which the
clinical manifestation of the disease is the result of a much
earlier fixed derailment of normal brain development
(Weinberger 1987) would predict a different temporal
profile for structural anomalies than a model of schizophrenia as a continuing active neurodegenerative process
(DeLisi 1999; Lieberman 1999). Identifying the most
accurate disease model has implications for early detection
and treatment. The corpus callosum, in particular, could be
a focus of the progressive brain change in schizophrenia
since it is known to increase in size during normal
adulthood in a process that might involve changes in the
number and size of fibers connecting the cerebral
hemispheres.
Most studies of the neuroanatomy of schizophrenia
have quantified volumes, areas, or linear dimensions of
different brain structures from computed tomography or
magnetic resonance imaging (MRI) images. Recently, a
number of studies have employed landmark-based approaches to address some of the same questions (e.g.,
Bookstein 1995, 1997; DeQuardo et al 1996, 1999). The
methods are based on the rapidly developing field of
geometric morphometrics and have the advantage of
allowing a geometrically correct partitioning of size and
shape information and preserving all of that information
for use in statistical tests and graphical representation.
An overview of the theory of geometric morphometrics,
its merits as a tool for shape analysis, and examples of
application are discussed in Marcus et al (1996) and
Rohlf (1999). Bookstein (1998) is an introduction to
0006-3223/00/$20.00
PII S0006-3223(00)00916-1
Morphometrics of First-Episode Schizophrenia
399
BIOL PSYCHIATRY
2000;48:398 – 405
geometric morphometrics specifically aimed at biological psychiatrists.
The purpose of our study was to use these techniques to
examine changes in brain structural shape in schizophrenia, a disease that has only subtle morphological changes
over time. A configuration of landmarks was identified in
brain MRI midsagittal sections of patients and control
subjects participating in a longitudinal study of firstepisode schizophrenia. These landmarks were chosen so as
to span regions of the brain where schizophrenia-associated structural abnormalities have been identified by
previous studies, namely, the corpus callosum, the diencephalon, the superior aspects of the brainstem and the
cerebellum, and the frontal cortex. In particular, the
chosen landmark configuration overlaps with those of
previous geometric morphometric studies (Bookstein
1995; DeQuardo et al 1996) so as to allow testing their
findings.
Given the well-established change in normal brain
anatomy with age and the growing literature on sexual
dimorphism in normal and schizophrenic brain anatomies
(Pearlson and Marsh 1999), the possible effects of gender
and age and their interactions with diagnosis and time
elapsed since diagnosis were also examined.
Table 1. Means and Standard Deviations of Age at Initial
Scan (in Years) and the Length of Follow-Up Period (in
Months)
Methods and Materials
Magnetic Resonance Imaging Scan Procedures
Subjects
Fifty-five patients (35 male and 20 female) and 22 control
subjects (13 male and 9 female) were examined. Details of the
recruitment and diagnostic procedures have previously been
described (DeLisi et al 1997). Patients were recruited from 1988
through 1994 when they presented to Suffolk County in-patient
facilities (Kings Park Psychiatric Hospital or University Hospital, SUNY, Stony Brook, New York) with a first episode of
nonaffective acute psychosis. All consecutive admissions who
gave informed consent for participation in our studies were
included. Control subjects were obtained from the surrounding
community by recruiting individuals who presented to other
out-patient facilities at University Hospital, Stony Brook for
nonpsychiatric disorders. Control subjects in the same age range
and social class as concurrently recruited patients were interviewed using a structured psychiatric interview format (SADS-L,
revised; Spitzer and Endicott 1978). The mean and standard
deviation of age of subjects at time of initial scan are given in
Table 1. Control subjects with a major psychiatric diagnosis,
physical illness, chronically medicated, or having any substance
abuse were eliminated from study. All subjects gave informed
consent for participation in this study.
Thirty-four of the patients were diagnosed with DSM-III-R
schizophreniform disorder, 7 with schizophrenia, subchronic, 9
schizoaffective disorder, 3 chronic schizophrenia, and 2 psychosis (not otherwise specified) at the time of first hospitalization. At
the follow-up time, 5 of these patients were recovered, 33 were
diagnosed with DSM-III-R chronic schizophrenia, 10 with
Mean age (SD)
Control subjects
Patients
Total
Mean length of follow-up
period (SD)
Control subjects
Patients
Total
Female
Male
Total
27.22 (3.96)
30.45 (8.44)
29.45 (7.43)
25.00 (6.79)
24.51 (6.34)
24.65 (6.40)
25.91 (5.80)
26.67 (7.66)
26.45 (7.15)
49.33 (6.87)
47.05 (7.77)
47.76 (7.46)
46.38 (7.73)
45.89 (5.31)
46.02 (5.98)
47.59 (7.37)
46.31 (6.27)
46.68 (6.58)
schizoaffective disorder, 2 with bipolar disorder, 2 with recurrent
major depression, 1 with schizophrenia-spectrum personality
disorders and 2 with psychosis (not otherwise specified). Statistical analyses were performed using, first, the complete sample
of 55 psychiatric patients as the syndromal group and then
limiting it to the subset of 43 patients (30 male and 13 female)
who were diagnosed with either chronic schizophrenia or schizoaffective disorder at the follow-up time. All patients were treated
with conventional neuroleptic medication for at least a portion of
the follow-up period. Some patients were also treated with
antidepressants, and some with lithium.
For each subject, two MRI scans were acquired using identical
protocols on the same scanner. T-1 weighted sagittal sequences
were obtained through the whole brain with a Signa scanner and
1.5 Tesla magnet (5-mm slice thickness with 2-mm gap; matrix,
256 3 256; field of view, 24; repetition time, 66; echo time, 25;
number of excitations, 2). Scans were taken at the time of first
admission (for patients) and at a follow-up evaluation ranging
from 33 to 64 months later. The mean and standard deviation of
the follow-up period length are indicated in Table 1. The slices
closest to true midsagittal were extracted from these scans
(guided mostly by the clarity of the colliculus and the corpus
callosum) and selected for analysis by an observer familiar with
brain anatomy but blind to the classification of each subject.
Landmark Configuration and Shape Analysis
An 11-landmark configuration was used to characterize the brain
morphology in midsagittal MRI images (Figure 1). This configuration largely followed the 13-landmark scheme of DeQuardo et
al (1996) except in that three landmarks—the vertex, tentorium at
dura, and bottom of cerebellum—were eliminated for being
difficult to identify and/or sensitive to imaging artifacts. The
procedures for localizing some of the landmarks were modified
from DeQuardo et al (1996) and one landmark, on the inside of
the splenium, was added (landmark 11). The inside of the
splenium had been associated with the most significant shape
aberration in patients with schizophrenia found by Bookstein
(1995). Landmarks 1 and 9 were located by extending a straight
line that passed through landmarks 2 and 11 to the outside of the
400
W.S. Gharaibeh et al
BIOL PSYCHIATRY
2000;48:398 – 405
each individual, are permuted as blocks; within each block the
first and second rows are permuted independently of the other
blocks. For each permutation the modified Goodall’s statistic is
calculated and saved. The proportion of F values that are equal
to or larger than the original value of the statistic is reported as
the p value of the test. Using the permutation procedure has the
advantage of eliminating dependence on the restrictive assumptions of Goodall’s F test.
Graphic Representation
Figure 1. The locations of the landmarks used in the present
study. 1, outside of splenium; 2, genu; 3, midpoint of corpus
callosum; 4, top of cerebellum; 5, tip of fourth ventricle; 6, top
of pons; 7, bottom of pons; 8, optic chiasm; 9, frontal pole (on the
extension of a line from 1 to 2); 10, superior colliculus; 11, inside
of splenium. (Modified from DeQuardo et al 1996 and Bookstein
1995.)
splenium and to the surface of the frontal cortex, respectively.
Landmark 3 was located at the upper edge of the corpus callosum
exactly midway between landmarks 1 and 2. The positions of
these and the remaining landmarks can be identified from a
number of figures in this article.
Prior to digitization, the MRI image files were coded and their
order permuted to keep the analyst blind to age, gender, diagnosis, and timing (initial or follow-up), and to guard against subtle
temporal alterations in the digitizing procedure. The landmark
configurations were superimposed using generalized leastsquares Procrustes fit (GLS) to remove nonshape variability
(differences in location, orientation, and size) in the landmark
coordinates (Rohlf and Slice 1990). A matrix, W, consisting of
the partial warp scores and the uniform component was then
computed (Bookstein 1991, 1996). This matrix contains all of the
shape information captured in the landmark configurations.
Statistical Analyses
Permutation tests were applied to test the effects of gender,
diagnosis, time, age, and all their possible interactions on brain
shape (as captured by the W matrix). The permutation test is
based on Goodall’s F statistic for comparing the difference in
mean shape between two samples relative to the shape variation
found within the samples (Goodall 1991). Our permutation
procedure uses a tangent space approximation of Goodall’s F
statistic. It is computed using Euclidean distances from the W
matrix instead of Procrustes distances. The paired comparisons
design (each individual’s initial and follow-up scans form a pair)
is not considered when calculating this statistic, but it is taken
into account by the permutation procedure. Pairs of rows of the
design matrix, corresponding to the initial and follow-up scans of
Consensus configurations, displacement vectors, thin plate
splines, and averaged unwarped images were used for visualizing
brain shape changes associated with diagnosis, time, gender, and
age. A consensus configuration is the average of x and y
coordinates of the Procrustes-aligned configurations taken landmark by landmark. Displacement vectors connect the landmarks
of one configuration with the corresponding landmarks of another configuration that has been superimposed on the first. A
thin plate spline (TPS) deformation grid represents the smoothest
interpolation required for exactly mapping the coordinates of one
landmark configuration into another (Bookstein 1991). An average unwarped image is a pixel-by-pixel average of a number of
individual images (here MRI scans) after being aligned to their
consensus configuration using the TPS function (Bookstein
1991). The apparent movements of landmarks and expansion and
contraction of areas in the deformation grids are all relative to the
configuration as a whole and indicate shape (not absolute size)
changes of different regions in the brain.
Results
Overall Test
Table 2 summarizes the results of testing the multivariate
general linear model with 55 psychiatric patients as the
syndromal group. The probabilities were estimated using
the permutation test procedure described above. The shape
differences between male and female patients were larger
than expected due to chance (p , .039). The diagnosisby-time interaction was also larger than expected due to
chance (p , .027), indicating that shape differences
between patients and control subjects are not the same
through the 3–5 year follow-up period, or, equivalently,
that the shape changes between the initial and follow-up
scans were not the same in patients and control subjects.
This interaction precludes a simple interpretation of the
main effects of either diagnosis or time (even though the
effect of time by itself is statistically significant). Instead,
the effects of diagnosis and of time were examined by
retesting each within the categories of the other. The
effects of gender and age on brain shape were also
examined within each diagnostic and time category.
Similar results were obtained when limiting the syndromal group to the patients diagnosed with chronic schizophrenia or schizoaffective disorder at the follow-up time
Morphometrics of First-Episode Schizophrenia
Table 2. Testing the Effects of Gender, Diagnosis, Time, and
Age on Brain Shapea in the Complete Sample (55 Patients and
22 Control Subjects)
Fb
pc
3.7948
2.3154
0.4767
2.5309
1.6544
0.1815
1.7450
0.4922
0.5644
0.0546
0.3257
2.0700
0.1599
0.0629
0.1752
.0388
.2376
.0325
.1822
.5054
.5503
.4512
.0268
.9783
.9870
.1436
.3187
.5965
.9746
.5276
Source of variation
Gender
Diagnosisd
Timee
Agef
Gender 3 diagnosis
Gender 3 time
Gender 3 age
Diagnosis 3 time
Diagnosis 3 age
Time 3 age
Gender 3 diagnosis 3 time
Gender 3 diagnosis 3 age
Gender 3 time 3 age
Diagnosis 3 time 3 age
Gender 3 diagnosis 3 time 3 age
401
BIOL PSYCHIATRY
2000;48:398 – 405
a
As captured by the matrix of partial warp scores and uniform component (W).
b
Modified Goodall’s F.
c
Probabilities are calculated using a permutation test (10,000 iterations).
d
Diagnosis is treated as a simple binary variable.
e
Time is in months since the initial scan (zero in all of the initial scans).
f
Age is in years at time of initial scan.
(n 5 43). The diagnosis-by-time interaction remained
significant (p , .038), but the gender effect lost significance (p , .11), as would be expected from a reduction in
the sample size.
The paired comparison design for the effect of time was
not considered when calculating Goodall’s F values, but it
was taken into account by the permutation testing procedure. This is why F values for terms that include a time
effect may have lower p values than F values of a similar
or even larger magnitude for terms not including a time
effect.
The Diagnosis-by-Time Interaction
We report the nominal p values for the permutation tests
performed within each subset of the data; however, the
Bonferroni method can be used to control for the experimentwise error rate. This is done by comparing the
nominal p values with the desired type I error rate divided
by 4 (for example, use .0025 to obtain an experimentwise
type I error rate of .01).
ANALYSES WITHIN EACH DIAGNOSTIC CATEGORY.
Table 3 shows that the effect of time was statistically
significant (even after a Bonferroni adjustment) in patients
(p , .002) but not in control subjects. The effect of time
remained significant when limiting the test to patients
diagnosed at follow-up time with chronic schizophrenia or
schizoaffective disorder (p , .009).
Figure 2 shows the changes in brain morphology from
the initial to the follow-up scans in patients (A) and in
Table 3. Testing the Effects of Gender, Time, and Age on
Brain Shape within Diagnostic Categories
Patients
(n 5 55)
Source of variation
Gender
Time
Age
Gender 3 time
Gender 3 age
Time 3 age
Gender 3 time 3 age
Control Subjects
(n 5 22)
Fa
pb
F
p
1.8615
0.7761
3.1764
0.1239
1.1323
0.1281
0.0806
.4051
.0017
.0819
.8212
.7776
.7663
.9462
3.0920
0.3719
1.4624
0.3513
2.2919
0.0533
0.2065
.1024
.0896
.5882
.1158
.2682
.9856
.4013
The independent and shape variables are as in Table 2.
Modified Goodall’s F.
Probabilities are calculated using a permutation test (10,000 iterations).
a
b
control subjects (B). For the purpose of graphic representation, time was treated as a binary variable, disregarding
differences in the duration of the follow-up period among
individuals. The superimposition of the consensus configurations of the initial and follow-up scans allows us to see
the magnitude and direction of changes in mean landmark
locations. The TPS deformation grid and displacement
vectors (but not the positions of the landmarks) were
exaggerated by a factor of 10 to make the subtle patterns
of shape change more visible. The distribution of age and
gender in the initial and follow-up sets of scans are
identical and so the graphics here, in parallel to the
statistical procedure, characterize the shape change over
time while holding other variables constant. Only the 43
patients diagnosed with chronic schizophrenia or schizoaffective disorder at follow-up time were used in Figure 2.
Including the 12 remaining patients did not change this
figure in any appreciable way.
ANALYSES AT EACH TIME OF SCAN. Similar statistical procedures were used to test for the effects of gender,
diagnosis, and age on brain morphology within the initial
and follow-up sets of scans (Table 4). The differences
between patients and control subjects were not significant
in the initial scans, and they were even less so in the
follow-up scans. Similar results were obtained when
limiting the tests to patients diagnosed with chronic
schizophrenia or schizoaffective disorder at follow-up
time.
Proper graphic presentation of shape differences between patients and control subjects is complicated by
unbalanced sampling. The proportion of female subjects is
higher in the control group (.41) than in the subset of
patients diagnosed at follow-up with chronic schizophrenia and schizoaffective disorder (.30); thus differences
between the consensus configurations for patients and
control subjects would be partially confounded by shape
402
W.S. Gharaibeh et al
BIOL PSYCHIATRY
2000;48:398 – 405
Table 4. Testing the Effects of Gender, Diagnosis, and Age on
Brain Shape within Initial and Follow-Up Scans
Initial scans
(n 5 77)
Source of variation
Gender
Diagnosis
Age
Gender 3 diagnosis
Gender 3 age
Diagnosis 3 age
Gender 3 diagnosis 3 age
Follow-up scans
(n 5 77)
Fa
pb
F
p
1.9777
1.6178
1.3056
0.8732
0.9593
0.3301
1.1917
.0437
.1105
.2220
.5351
.4511
.9770
.2905
2.0278
1.1335
1.1364
1.1329
0.8366
0.2297
1.0245
.0414
.3206
.3211
.3275
.5632
.9952
.3975
The independent and shape variables are as in Table 2.
a
Modified Goodall’s F.
b
Probabilities are calculated using a permutation test (10,000 iterations).
essentially the same as those constructed using the schizophrenia/schizoaffective disorder subset. The distributions
of age and time were little affected by the elimination of
“excess” male patients, the mean follow-up period length
and age remained very similar between patients and
control subjects. Differences in variance among the different categories should not affect the graphic representation, barring a strong presence of certain kinds of nonlinear relationship between shape, on one hand, and time or
age, on the other. Plotting partial warp scores against time
and age showed little evidence for such nonlinearity.
Gender Effects
Figure 2. The effect of time on brain shape in patients (A) and
control subjects (B). The follow-up consensus configuration
(white circles) is least squares superimposed on the initial
consensus configuration (large dark circles). The average unwarped image of the initial scans is shown in the background.
The displacement vectors and thin plate spline grid show the
deformation of the initial consensus into the follow-up consensus
with their magnitude exaggerated by a factor of 10 (the position
of the landmarks is not exaggerated).
differences due to gender. To make the female-to-male
ratio in patients with schizophrenia equal to that in control
subjects, the 11 male patients with schizophrenia with the
lowest quality MRI images were not used in preparation of
Figure 3. A similar procedure was used with the complete
sample of patients and the resulting visualizations were
Table 2 indicates the presence of sexual dimorphism in
brain morphology. Testing for gender effects within initial
and follow-up scans (Table 4) confirms this finding and
suggests that this difference is stable throughout the
follow-up period. Gender and its interactions were not
significant when tested separately within either diagnostic
category, as would be expected given the reduction in the
sample size. Note, however, that the results are much
closer to significance in control subjects than in patients
despite their smaller sample size (Table 3).
The unbalanced design makes the graphic representation of gender effects difficult. Not only is the ratio of
control subjects to patients higher in women than in men,
but there is also a mean age difference of about 6 years
between the genders in the patient group (Table 1).
Examining the deformation grids within diagnostic categories (Figure 4) avoids confounding the effects of schizophrenia with those of gender, and provides a separate
graphic representation of sexual dimorphism in the 43
patients with chronic schizophrenia or schizoaffective
disorder (for whom the age gap between the genders is
part of the disease phenomenology) for comparison with
that of control subjects.
Morphometrics of First-Episode Schizophrenia
Figure 3. The effect of schizophrenia on brain shape at time of
diagnosis (A) and 3–5 years later (B). The patient consensus
configuration (white circles) is least squares superimposed on the
control consensus configuration (large dark circles). The average
unwarped image of the control scans is shown in the background.
The displacement vectors and thin plate spline grid show the
deformation of the control consensus into the patient consensus
with their magnitude exaggerated by a factor of 10 (the position
of the landmarks is not exaggerated).
Age Effects
Age effects were not statistically significant in any of the
tests performed with the complete patient sample; however, the probability for an age effect was much reduced
when limiting the syndromal group to the chronic schizophrenia and schizoaffective disorder (at follow-up time)
BIOL PSYCHIATRY
2000;48:398 – 405
403
Figure 4. The effect of gender on brain shape in patients (A) and
control subjects (B). The male consensus configuration (white
circles) is least squares superimposed on the female consensus
configuration (large dark circles). The average unwarped image
of the female subjects is shown in the background. The displacement vectors and thin plate spline grid show the deformation of
the female consensus into the male consensus with their magnitude exaggerated 10 times (the position of the landmarks is not
exaggerated).
patients in the overall test (n 5 65, F 5 3.217, p , .08)
and was significant when testing within that subset of
patients (n 5 43, F 5 4.7396, p , .011).
Discussion
Our study investigated midline regional brain shape differences between patients with schizophrenia and control sub-
404
BIOL PSYCHIATRY
2000;48:398 – 405
jects at the time of a first psychotic episode and 3–5 years
later. The significant diagnosis-by-time interaction term and
the significant time main effect within patients suggest that
some aspects of brain morphology do change in patients with
schizophrenia during the 3–5 years following the onset of
symptoms beyond the normal change found in nonpsychiatric individuals. The effect size for time in patients was twice
that in control subjects (F 5 0.776 and F 5 0.372, respectively) suggesting that the higher significance of time effect
in patients is not simply owing to their larger sample size.
The midline shape differences in brain morphology between patients and control subjects were not statistically
significant at the time of diagnosis and seem to become less
significant with the passage of time. Many of the differences
found by graphic inspection, however, are very similar to
those reported in DeQuardo et al (1996) and Bookstein
(1995), and these abnormalities seem to largely persist
throughout the follow-up period. The deformation grids of
these two studies—which shared the syndromal group—
show a relative expansion of the area between the splenium,
the colliculus, and the cerebellum (and a complementary
compression in the area between the colliculus, the pons, and
the chiasm) that is comparable to what we see in Figure 3 at
either time. In fact, these two studies reported an increase in
the relative distance between the superior colliculus and the
splenium (contributing to the said expansion) to be the most
significant schizophrenia abnormality using the 13-landmark
scheme. The relative thinning of the splenium in patients with
schizophrenia that was emphasized in Bookstein (1995) can
be detected in our contrast of patients and control subjects at
both diagnosis and follow-up times. The arching of the
corpus callosum and the relative backward movement of the
genu seen in Figure 3 (both A and B) are also comparable to
what was found in the two earlier studies. The thinning of the
splenium, backward movement of the genu, and arching of
the corpus callosum were better resolved in Bookstein
(1997), in which a variant of landmark analysis was devised
and applied to the data of Bookstein (1995) focusing on the
complete outline of the corpus callosum. DeQuardo et al
(1999) used a landmark configuration that partially overlaps
with those used in Bookstein (1995), DeQuardo et al (1996),
and the study discussed here to examine the neuroanatomy of
first-episode schizophrenia. The increased number of nonoverlapping landmarks between the configuration of DeQuardo et al (1999) and those of the three other studies
complicates the comparison; indeed, their deformation grids
and superimposition plots are less similar to those of the other
three. Still, the plots of DeQuardo et al (1999) clearly
reproduce the relative backward movement of the genu and
the relative expansion of the area between the splenium, the
cerebellum, and the superior colliculus that was found in the
other three studies. In rendering Figure 3, individuals were
chosen so that the gender composition within both patients
W.S. Gharaibeh et al
and control subjects was the same (.41 female, .59 male).
This proportion is intermediate between those used in Bookstein (1995) and DeQuardo et al (1996) (.50:.50, on one hand,
and those in DeQuardo et al [1999], .31:.69, on the other) and
is close enough to either to make cross-study graphic comparisons legitimate. This might be important because although the gender-by-diagnosis and gender-by-diagnosis-bytime interaction terms were not statistically significant, the
differences between patients and control subjects were quite
different when examined separately in men and women
(figures available upon request). The resolution of this
question requires a larger sample size than was available for
this study. Tibbo et al (1998) found no significant differences
in corpus callosum shape between 79 male patients with
schizophrenia and 65 male control subjects using a landmarkbased analysis (but see comment by DeQuardo 1999).
The presence of structural abnormalities at the time of
diagnosis and the significant change with time in patients, but
not in control subjects, are consistent with a neurodegenerative process that is closely associated with the onset of
symptoms. Data are now accumulating from MRI and other
studies that show that brain plasticity is different in patients
with schizophrenia than in normal control subjects (for a
review, see DeLisi 1999). Since it is now known that
neuronal growth and change occurs within the adult brain
(Gould et al 1999), one can speculate that any one aspect of
numerous steps in this process could be abnormal in individuals with schizophrenia, resulting in greater than normal cell
death, less cell growth, or both over time in some areas more
than others. It is worth noting, however, that all patients, and
none of the control subjects, were exposed to psychiatric
drugs during the follow-up period. This allows for an alternative interpretation of our results in which structural abnormalities are established much earlier in brain development
with the medication producing the observed rapid change
during the follow-up period. This interpretation is consistent
with the reduced statistical significance of differences between patients and control subjects at follow-up time. The
possibility of psychotropic drugs having a normalizing affect
on brain morphology has been suggested in the literature
(e.g., Keshavan 1998). A further clarification of the relationship between developmental processes and the establishment
of structural abnormalities could be obtained by longitudinal
studies that track the change in brain morphology with time
in control and individuals at high risk for schizophrenia prior
to the onset of symptoms, such as that of Lawrie et al (1999).
Our results also suggest the presence of a significant
sexual dimorphism in brain morphology that is stable
throughout the follow-up period but which seems to be
somewhat diminished in schizophrenia. The splenium in
women seems to be relatively thicker and more distant from
the cerebellum and the colliculus than in men, who in turn
seem to have a proportionally larger frontal cortical area.
Morphometrics of First-Episode Schizophrenia
Useful comparisons with previous studies (e.g., Hoff et al
1994, which overlaps the study sample of the work discussed
here) are precluded by differences in method of quantification (landmark vs. linear distances, area or volumetric analysis) and whether measurements were controlled for total
brain size.
The effect of age on brain shape was at best marginally
significant in the overall test but significant within the
subset of patients diagnosed with chronic schizophrenia or
schizoaffective disorder at time of follow-up. Still, separate TPS representations of age effects within patients and
control subjects showed similarities that are difficult to
interpret as anything other than a reflection of a common
brain aging process (figures available upon request). These
results suggest that taking age effects into consideration
when examining the dynamics of schizophrenia abnormalities was a warranted precaution.
Although not conclusive in their support of one etiologic
hypothesis over the others, our results could be instructive in
planning further tests of these hypotheses. The methodology
we have presented in this article seems to be especially suited
for application in studies that examine changes in brain
morphology with aging, progressive disease processes, medication, and other variables.
Supported by Ecological and Evolutionary Physiology (IBN-9728160)
program of the National Science Foundation (FJR, DES).
The landmark configurations were digitized using tpsDig, the GLS
superimposition and calculation of W were performed using tpsRelw and
the average unwarped images were produced using tpsSuper, both
written by F. James Rohlf. The figures that combine consensus configurations, TPS splines, displacement vectors, and unwarped average
images were produced in Morpheus et al, written by Dennis E. Slice. The
permutation procedures for multivariate testing were written and performed in MATLAB (MathWorks, Natick, MA) by Waleed Gharaibeh.
These programs are available from the SUNY at Stony Brook morphometrics website: http://life.bio.sunysb.edu/morph.
The authors thank Mary Kritzer for providing technical help with data
collection, Dean Adams for critiquing the morphometric and statistical
methodology, and Kerry Brown for help with the manuscript. The
constructive comments by three anonymous reviewers were also greatly
appreciated.
References
Bookstein FL (1991): Morphometric Tools for Landmark Data:
Geometry and Biology. Cambridge, UK: Cambridge University Press.
Bookstein FL (1995): How to produce a landmark point: The
statistical geometry of incompletely registered images. In:
Metler R, Wu A, Bookstein F, Green W, editors. Vision
Geometry IV. S.P.I.E. Proceedings, Vol 2573. Bellingham,
WA: SPIE Press, 266 –277.
Bookstein FL (1996): Standard formula for the uniform shape
component in landmark data. In: Marcus LF, Corti M, Loy A,
Naylor G, Slice D, editors. Advances in Morphometrics. New
York: Plenum Press, 153–168.
BIOL PSYCHIATRY
2000;48:398 – 405
405
Bookstein FL (1997): Landmark methods for forms without
landmarks: Localizing group differences in outline shape.
Med Image Analysis 1:225–243.
Bookstein FL (1998): Linear methods for nonlinear maps:
Procrustes fits, thin-plate splines, and the biometric analysis
of shape variability. In: Toga A, editor. Brain Warping. San
Diego: Academic Press, 157–181.
DeLisi LE (1999): Defining the course of brain structural change
and plasticity in schizophrenia. Psychiatry Res Neuroimaging
92:1–9.
DeLisi LE, Sakuma M, Kushner M, Finer DL, Huff AL, Crow TJ
(1997): Anomalous cerebral asymmetry and language processing in schizophrenia. Schizophr Bull 23:255–271.
DeQuardo JR (1999): Landmark analysis of corpus callosum
shape in schizophrenia. Biol Psychiatry 46:1712–1713.
DeQuardo JR, Bookstein FL, Green WDK, Brunberg JA, Tandon R
(1996): Spatial relationships of the neuroanatomic landmarks in
schizophrenia. Psychiatry Res Neuroimaging 67:81–95.
DeQuardo JR, Keshavan MS, Bookstein FL, Bagwell WW,
Green WDK, Sweeney JA, et al (1999): Landmark-based
morphometric analysis of first-episode schizophrenia. Biol
Psychiatry 45:1321–1328.
Goodall C (1991): Procrustes methods in the statistical analysis
of shape. J R Stat Soc B 53:285–339.
Gould E, Reeves AJ, Graziano MSA, Gross CG (1999): Neurogenesis in the neocortex of adult primates. Science 286:548 –552.
Hoff AL, Neal C, Kushner M, DeLisi LE (1994): Gender
differences in corpus callosum size in first-episode schizophrenia. Biol Psychiatry 35:913–919.
Keshavan MS, Haas GL, Kahn CE, Auilar E, Scooler NR,
Sweeney JA, et al (1998): Superior temporal gyrus and the
course of early schizophrenia: Progressive, static or reversible? J Psychiatr Res 32:161–167.
Lawrie SM, Whalley H, Kestelman JN, Abukmeil SS, Byrne M,
Hodges A, et al (1999): Magnetic resonance imaging of brain
in people at high risk of developing schizophrenia. Lancet
353:30 –33.
Lieberman JA (1999): Is schizophrenia a neurodegenerative
disorder? A clinical and neurobiological perspective. Biol
Psychiatry 46:729 –739.
Marcus LF, Corti M, Loy A, Naylor G, Slice D, editors (1996):
Advances in Morphometrics. New York: Plenum Press.
Pearlson GD, Marsh L (1999): Structural brain imaging in schizophrenia: A selective review. Biol Psychiatry 46:627– 649.
Rohlf FJ (1999): Shape statistics: Procrustes superimpositions
and tangent spaces. J Classification 16:197–223.
Rohlf FJ, Slice D (1990): Extensions of the Procrustes method
for the optimal superimposition of landmarks. Syst Zool
39:40 –59.
Spitzer RL, Endicott J (1978): Schedule for Affective Disorders
and Schizophrenia (SADS). New York: New York State
Psychiatric Institute, Biometrics Research Department.
Tibbo P, Nopoulos P, Arndt S, Andreasen NC (1998): Corpus
callosum shape and size in male patients with schizophrenia.
Biol Psychiatry 44:405– 412.
Weinberger DR (1987): Implications of normal brain development for the pathogenesis of schizophrenia. Arch Gen Psychiatry 44:660 – 669.
Midline Brain Structural Shape following a First
Episode of Schizophrenia
Waleed S. Gharaibeh, F. James Rohlf, Dennis E. Slice, and Lynn E. DeLisi
Background: Previous reports indicate that brain structural abnormalities may be progressive in some patients
with schizophrenia. Our study was designed to determine
deviations in the shape of midline brain structures at the
time of onset of symptoms of schizophrenia and 3–5 years
later.
Methods: Eleven landmarks were located on the midsagittal magnetic resonance imagery brain scans of 55
patients with schizophrenia and 22 nonpsychiatric control
individuals. Geometric morphometric methods were used
for the extraction of shape variables from landmark
coordinates. Permutation tests were used to test the effects
of gender, diagnosis, time elapsed since illness onset, and
age on brain shape.
Results: The diagnosis-by-time interaction and the effect
of gender were significantly different from zero (p , .027
and p , .039, respectively). The effect of time was
significant in patients (p , .002), but not in control
subjects. Some anatomical abnormalities in mean patient
brain morphology seem to be present both at the time of
diagnosis and at follow-up. These are similar to anomalies
reported by previous geometric morphometrics studies.
Conclusions: Some previously identified brain abnormalities are detectable at the time of first hospitalization. The
rapid change in midline brain morphology in patients with
schizophrenia during the subsequent 3–5 years is consistent with either a neurodegenerative disease process or an
effect of treatment with psychiatric drugs. There is a
sexual dimorphism in brain morphology that might be
reduced by schizophrenia. Biol Psychiatry 2000;48:
398 – 405 © 2000 Society of Biological Psychiatry
Key Words: First-episode schizophrenia, MRI, brain
morphology, corpus callosum, Procrustes, thin-plate spline
From the Departments of Ecology and Evolution (WSG, FJR, DES) and Psychiatry
(LED), SUNY, Stony Brook, New York.
Address reprint requests to F. James Rohlf, SUNY, Department of Ecology and
Evolution, Stony Brook NY 11794-5245.
Received January 27, 2000; revised April 20, 2000; accepted May 2, 2000.
© 2000 Society of Biological Psychiatry
Introduction
O
ver the past 25 years, neuroimaging studies have
associated schizophrenia with a number of anatomic
brain abnormalities with varying degrees of reproducibility (for a review, see Pearlson and Marsh 1999); however,
only a few have investigated the timing of emergence of
these neuroanatomic abnormalities (for a review, see
DeLisi 1999). Establishing the temporal progress of pathology can be of value in testing competing hypotheses
for the etiology of schizophrenia. A model in which the
clinical manifestation of the disease is the result of a much
earlier fixed derailment of normal brain development
(Weinberger 1987) would predict a different temporal
profile for structural anomalies than a model of schizophrenia as a continuing active neurodegenerative process
(DeLisi 1999; Lieberman 1999). Identifying the most
accurate disease model has implications for early detection
and treatment. The corpus callosum, in particular, could be
a focus of the progressive brain change in schizophrenia
since it is known to increase in size during normal
adulthood in a process that might involve changes in the
number and size of fibers connecting the cerebral
hemispheres.
Most studies of the neuroanatomy of schizophrenia
have quantified volumes, areas, or linear dimensions of
different brain structures from computed tomography or
magnetic resonance imaging (MRI) images. Recently, a
number of studies have employed landmark-based approaches to address some of the same questions (e.g.,
Bookstein 1995, 1997; DeQuardo et al 1996, 1999). The
methods are based on the rapidly developing field of
geometric morphometrics and have the advantage of
allowing a geometrically correct partitioning of size and
shape information and preserving all of that information
for use in statistical tests and graphical representation.
An overview of the theory of geometric morphometrics,
its merits as a tool for shape analysis, and examples of
application are discussed in Marcus et al (1996) and
Rohlf (1999). Bookstein (1998) is an introduction to
0006-3223/00/$20.00
PII S0006-3223(00)00916-1
Morphometrics of First-Episode Schizophrenia
399
BIOL PSYCHIATRY
2000;48:398 – 405
geometric morphometrics specifically aimed at biological psychiatrists.
The purpose of our study was to use these techniques to
examine changes in brain structural shape in schizophrenia, a disease that has only subtle morphological changes
over time. A configuration of landmarks was identified in
brain MRI midsagittal sections of patients and control
subjects participating in a longitudinal study of firstepisode schizophrenia. These landmarks were chosen so as
to span regions of the brain where schizophrenia-associated structural abnormalities have been identified by
previous studies, namely, the corpus callosum, the diencephalon, the superior aspects of the brainstem and the
cerebellum, and the frontal cortex. In particular, the
chosen landmark configuration overlaps with those of
previous geometric morphometric studies (Bookstein
1995; DeQuardo et al 1996) so as to allow testing their
findings.
Given the well-established change in normal brain
anatomy with age and the growing literature on sexual
dimorphism in normal and schizophrenic brain anatomies
(Pearlson and Marsh 1999), the possible effects of gender
and age and their interactions with diagnosis and time
elapsed since diagnosis were also examined.
Table 1. Means and Standard Deviations of Age at Initial
Scan (in Years) and the Length of Follow-Up Period (in
Months)
Methods and Materials
Magnetic Resonance Imaging Scan Procedures
Subjects
Fifty-five patients (35 male and 20 female) and 22 control
subjects (13 male and 9 female) were examined. Details of the
recruitment and diagnostic procedures have previously been
described (DeLisi et al 1997). Patients were recruited from 1988
through 1994 when they presented to Suffolk County in-patient
facilities (Kings Park Psychiatric Hospital or University Hospital, SUNY, Stony Brook, New York) with a first episode of
nonaffective acute psychosis. All consecutive admissions who
gave informed consent for participation in our studies were
included. Control subjects were obtained from the surrounding
community by recruiting individuals who presented to other
out-patient facilities at University Hospital, Stony Brook for
nonpsychiatric disorders. Control subjects in the same age range
and social class as concurrently recruited patients were interviewed using a structured psychiatric interview format (SADS-L,
revised; Spitzer and Endicott 1978). The mean and standard
deviation of age of subjects at time of initial scan are given in
Table 1. Control subjects with a major psychiatric diagnosis,
physical illness, chronically medicated, or having any substance
abuse were eliminated from study. All subjects gave informed
consent for participation in this study.
Thirty-four of the patients were diagnosed with DSM-III-R
schizophreniform disorder, 7 with schizophrenia, subchronic, 9
schizoaffective disorder, 3 chronic schizophrenia, and 2 psychosis (not otherwise specified) at the time of first hospitalization. At
the follow-up time, 5 of these patients were recovered, 33 were
diagnosed with DSM-III-R chronic schizophrenia, 10 with
Mean age (SD)
Control subjects
Patients
Total
Mean length of follow-up
period (SD)
Control subjects
Patients
Total
Female
Male
Total
27.22 (3.96)
30.45 (8.44)
29.45 (7.43)
25.00 (6.79)
24.51 (6.34)
24.65 (6.40)
25.91 (5.80)
26.67 (7.66)
26.45 (7.15)
49.33 (6.87)
47.05 (7.77)
47.76 (7.46)
46.38 (7.73)
45.89 (5.31)
46.02 (5.98)
47.59 (7.37)
46.31 (6.27)
46.68 (6.58)
schizoaffective disorder, 2 with bipolar disorder, 2 with recurrent
major depression, 1 with schizophrenia-spectrum personality
disorders and 2 with psychosis (not otherwise specified). Statistical analyses were performed using, first, the complete sample
of 55 psychiatric patients as the syndromal group and then
limiting it to the subset of 43 patients (30 male and 13 female)
who were diagnosed with either chronic schizophrenia or schizoaffective disorder at the follow-up time. All patients were treated
with conventional neuroleptic medication for at least a portion of
the follow-up period. Some patients were also treated with
antidepressants, and some with lithium.
For each subject, two MRI scans were acquired using identical
protocols on the same scanner. T-1 weighted sagittal sequences
were obtained through the whole brain with a Signa scanner and
1.5 Tesla magnet (5-mm slice thickness with 2-mm gap; matrix,
256 3 256; field of view, 24; repetition time, 66; echo time, 25;
number of excitations, 2). Scans were taken at the time of first
admission (for patients) and at a follow-up evaluation ranging
from 33 to 64 months later. The mean and standard deviation of
the follow-up period length are indicated in Table 1. The slices
closest to true midsagittal were extracted from these scans
(guided mostly by the clarity of the colliculus and the corpus
callosum) and selected for analysis by an observer familiar with
brain anatomy but blind to the classification of each subject.
Landmark Configuration and Shape Analysis
An 11-landmark configuration was used to characterize the brain
morphology in midsagittal MRI images (Figure 1). This configuration largely followed the 13-landmark scheme of DeQuardo et
al (1996) except in that three landmarks—the vertex, tentorium at
dura, and bottom of cerebellum—were eliminated for being
difficult to identify and/or sensitive to imaging artifacts. The
procedures for localizing some of the landmarks were modified
from DeQuardo et al (1996) and one landmark, on the inside of
the splenium, was added (landmark 11). The inside of the
splenium had been associated with the most significant shape
aberration in patients with schizophrenia found by Bookstein
(1995). Landmarks 1 and 9 were located by extending a straight
line that passed through landmarks 2 and 11 to the outside of the
400
W.S. Gharaibeh et al
BIOL PSYCHIATRY
2000;48:398 – 405
each individual, are permuted as blocks; within each block the
first and second rows are permuted independently of the other
blocks. For each permutation the modified Goodall’s statistic is
calculated and saved. The proportion of F values that are equal
to or larger than the original value of the statistic is reported as
the p value of the test. Using the permutation procedure has the
advantage of eliminating dependence on the restrictive assumptions of Goodall’s F test.
Graphic Representation
Figure 1. The locations of the landmarks used in the present
study. 1, outside of splenium; 2, genu; 3, midpoint of corpus
callosum; 4, top of cerebellum; 5, tip of fourth ventricle; 6, top
of pons; 7, bottom of pons; 8, optic chiasm; 9, frontal pole (on the
extension of a line from 1 to 2); 10, superior colliculus; 11, inside
of splenium. (Modified from DeQuardo et al 1996 and Bookstein
1995.)
splenium and to the surface of the frontal cortex, respectively.
Landmark 3 was located at the upper edge of the corpus callosum
exactly midway between landmarks 1 and 2. The positions of
these and the remaining landmarks can be identified from a
number of figures in this article.
Prior to digitization, the MRI image files were coded and their
order permuted to keep the analyst blind to age, gender, diagnosis, and timing (initial or follow-up), and to guard against subtle
temporal alterations in the digitizing procedure. The landmark
configurations were superimposed using generalized leastsquares Procrustes fit (GLS) to remove nonshape variability
(differences in location, orientation, and size) in the landmark
coordinates (Rohlf and Slice 1990). A matrix, W, consisting of
the partial warp scores and the uniform component was then
computed (Bookstein 1991, 1996). This matrix contains all of the
shape information captured in the landmark configurations.
Statistical Analyses
Permutation tests were applied to test the effects of gender,
diagnosis, time, age, and all their possible interactions on brain
shape (as captured by the W matrix). The permutation test is
based on Goodall’s F statistic for comparing the difference in
mean shape between two samples relative to the shape variation
found within the samples (Goodall 1991). Our permutation
procedure uses a tangent space approximation of Goodall’s F
statistic. It is computed using Euclidean distances from the W
matrix instead of Procrustes distances. The paired comparisons
design (each individual’s initial and follow-up scans form a pair)
is not considered when calculating this statistic, but it is taken
into account by the permutation procedure. Pairs of rows of the
design matrix, corresponding to the initial and follow-up scans of
Consensus configurations, displacement vectors, thin plate
splines, and averaged unwarped images were used for visualizing
brain shape changes associated with diagnosis, time, gender, and
age. A consensus configuration is the average of x and y
coordinates of the Procrustes-aligned configurations taken landmark by landmark. Displacement vectors connect the landmarks
of one configuration with the corresponding landmarks of another configuration that has been superimposed on the first. A
thin plate spline (TPS) deformation grid represents the smoothest
interpolation required for exactly mapping the coordinates of one
landmark configuration into another (Bookstein 1991). An average unwarped image is a pixel-by-pixel average of a number of
individual images (here MRI scans) after being aligned to their
consensus configuration using the TPS function (Bookstein
1991). The apparent movements of landmarks and expansion and
contraction of areas in the deformation grids are all relative to the
configuration as a whole and indicate shape (not absolute size)
changes of different regions in the brain.
Results
Overall Test
Table 2 summarizes the results of testing the multivariate
general linear model with 55 psychiatric patients as the
syndromal group. The probabilities were estimated using
the permutation test procedure described above. The shape
differences between male and female patients were larger
than expected due to chance (p , .039). The diagnosisby-time interaction was also larger than expected due to
chance (p , .027), indicating that shape differences
between patients and control subjects are not the same
through the 3–5 year follow-up period, or, equivalently,
that the shape changes between the initial and follow-up
scans were not the same in patients and control subjects.
This interaction precludes a simple interpretation of the
main effects of either diagnosis or time (even though the
effect of time by itself is statistically significant). Instead,
the effects of diagnosis and of time were examined by
retesting each within the categories of the other. The
effects of gender and age on brain shape were also
examined within each diagnostic and time category.
Similar results were obtained when limiting the syndromal group to the patients diagnosed with chronic schizophrenia or schizoaffective disorder at the follow-up time
Morphometrics of First-Episode Schizophrenia
Table 2. Testing the Effects of Gender, Diagnosis, Time, and
Age on Brain Shapea in the Complete Sample (55 Patients and
22 Control Subjects)
Fb
pc
3.7948
2.3154
0.4767
2.5309
1.6544
0.1815
1.7450
0.4922
0.5644
0.0546
0.3257
2.0700
0.1599
0.0629
0.1752
.0388
.2376
.0325
.1822
.5054
.5503
.4512
.0268
.9783
.9870
.1436
.3187
.5965
.9746
.5276
Source of variation
Gender
Diagnosisd
Timee
Agef
Gender 3 diagnosis
Gender 3 time
Gender 3 age
Diagnosis 3 time
Diagnosis 3 age
Time 3 age
Gender 3 diagnosis 3 time
Gender 3 diagnosis 3 age
Gender 3 time 3 age
Diagnosis 3 time 3 age
Gender 3 diagnosis 3 time 3 age
401
BIOL PSYCHIATRY
2000;48:398 – 405
a
As captured by the matrix of partial warp scores and uniform component (W).
b
Modified Goodall’s F.
c
Probabilities are calculated using a permutation test (10,000 iterations).
d
Diagnosis is treated as a simple binary variable.
e
Time is in months since the initial scan (zero in all of the initial scans).
f
Age is in years at time of initial scan.
(n 5 43). The diagnosis-by-time interaction remained
significant (p , .038), but the gender effect lost significance (p , .11), as would be expected from a reduction in
the sample size.
The paired comparison design for the effect of time was
not considered when calculating Goodall’s F values, but it
was taken into account by the permutation testing procedure. This is why F values for terms that include a time
effect may have lower p values than F values of a similar
or even larger magnitude for terms not including a time
effect.
The Diagnosis-by-Time Interaction
We report the nominal p values for the permutation tests
performed within each subset of the data; however, the
Bonferroni method can be used to control for the experimentwise error rate. This is done by comparing the
nominal p values with the desired type I error rate divided
by 4 (for example, use .0025 to obtain an experimentwise
type I error rate of .01).
ANALYSES WITHIN EACH DIAGNOSTIC CATEGORY.
Table 3 shows that the effect of time was statistically
significant (even after a Bonferroni adjustment) in patients
(p , .002) but not in control subjects. The effect of time
remained significant when limiting the test to patients
diagnosed at follow-up time with chronic schizophrenia or
schizoaffective disorder (p , .009).
Figure 2 shows the changes in brain morphology from
the initial to the follow-up scans in patients (A) and in
Table 3. Testing the Effects of Gender, Time, and Age on
Brain Shape within Diagnostic Categories
Patients
(n 5 55)
Source of variation
Gender
Time
Age
Gender 3 time
Gender 3 age
Time 3 age
Gender 3 time 3 age
Control Subjects
(n 5 22)
Fa
pb
F
p
1.8615
0.7761
3.1764
0.1239
1.1323
0.1281
0.0806
.4051
.0017
.0819
.8212
.7776
.7663
.9462
3.0920
0.3719
1.4624
0.3513
2.2919
0.0533
0.2065
.1024
.0896
.5882
.1158
.2682
.9856
.4013
The independent and shape variables are as in Table 2.
Modified Goodall’s F.
Probabilities are calculated using a permutation test (10,000 iterations).
a
b
control subjects (B). For the purpose of graphic representation, time was treated as a binary variable, disregarding
differences in the duration of the follow-up period among
individuals. The superimposition of the consensus configurations of the initial and follow-up scans allows us to see
the magnitude and direction of changes in mean landmark
locations. The TPS deformation grid and displacement
vectors (but not the positions of the landmarks) were
exaggerated by a factor of 10 to make the subtle patterns
of shape change more visible. The distribution of age and
gender in the initial and follow-up sets of scans are
identical and so the graphics here, in parallel to the
statistical procedure, characterize the shape change over
time while holding other variables constant. Only the 43
patients diagnosed with chronic schizophrenia or schizoaffective disorder at follow-up time were used in Figure 2.
Including the 12 remaining patients did not change this
figure in any appreciable way.
ANALYSES AT EACH TIME OF SCAN. Similar statistical procedures were used to test for the effects of gender,
diagnosis, and age on brain morphology within the initial
and follow-up sets of scans (Table 4). The differences
between patients and control subjects were not significant
in the initial scans, and they were even less so in the
follow-up scans. Similar results were obtained when
limiting the tests to patients diagnosed with chronic
schizophrenia or schizoaffective disorder at follow-up
time.
Proper graphic presentation of shape differences between patients and control subjects is complicated by
unbalanced sampling. The proportion of female subjects is
higher in the control group (.41) than in the subset of
patients diagnosed at follow-up with chronic schizophrenia and schizoaffective disorder (.30); thus differences
between the consensus configurations for patients and
control subjects would be partially confounded by shape
402
W.S. Gharaibeh et al
BIOL PSYCHIATRY
2000;48:398 – 405
Table 4. Testing the Effects of Gender, Diagnosis, and Age on
Brain Shape within Initial and Follow-Up Scans
Initial scans
(n 5 77)
Source of variation
Gender
Diagnosis
Age
Gender 3 diagnosis
Gender 3 age
Diagnosis 3 age
Gender 3 diagnosis 3 age
Follow-up scans
(n 5 77)
Fa
pb
F
p
1.9777
1.6178
1.3056
0.8732
0.9593
0.3301
1.1917
.0437
.1105
.2220
.5351
.4511
.9770
.2905
2.0278
1.1335
1.1364
1.1329
0.8366
0.2297
1.0245
.0414
.3206
.3211
.3275
.5632
.9952
.3975
The independent and shape variables are as in Table 2.
a
Modified Goodall’s F.
b
Probabilities are calculated using a permutation test (10,000 iterations).
essentially the same as those constructed using the schizophrenia/schizoaffective disorder subset. The distributions
of age and time were little affected by the elimination of
“excess” male patients, the mean follow-up period length
and age remained very similar between patients and
control subjects. Differences in variance among the different categories should not affect the graphic representation, barring a strong presence of certain kinds of nonlinear relationship between shape, on one hand, and time or
age, on the other. Plotting partial warp scores against time
and age showed little evidence for such nonlinearity.
Gender Effects
Figure 2. The effect of time on brain shape in patients (A) and
control subjects (B). The follow-up consensus configuration
(white circles) is least squares superimposed on the initial
consensus configuration (large dark circles). The average unwarped image of the initial scans is shown in the background.
The displacement vectors and thin plate spline grid show the
deformation of the initial consensus into the follow-up consensus
with their magnitude exaggerated by a factor of 10 (the position
of the landmarks is not exaggerated).
differences due to gender. To make the female-to-male
ratio in patients with schizophrenia equal to that in control
subjects, the 11 male patients with schizophrenia with the
lowest quality MRI images were not used in preparation of
Figure 3. A similar procedure was used with the complete
sample of patients and the resulting visualizations were
Table 2 indicates the presence of sexual dimorphism in
brain morphology. Testing for gender effects within initial
and follow-up scans (Table 4) confirms this finding and
suggests that this difference is stable throughout the
follow-up period. Gender and its interactions were not
significant when tested separately within either diagnostic
category, as would be expected given the reduction in the
sample size. Note, however, that the results are much
closer to significance in control subjects than in patients
despite their smaller sample size (Table 3).
The unbalanced design makes the graphic representation of gender effects difficult. Not only is the ratio of
control subjects to patients higher in women than in men,
but there is also a mean age difference of about 6 years
between the genders in the patient group (Table 1).
Examining the deformation grids within diagnostic categories (Figure 4) avoids confounding the effects of schizophrenia with those of gender, and provides a separate
graphic representation of sexual dimorphism in the 43
patients with chronic schizophrenia or schizoaffective
disorder (for whom the age gap between the genders is
part of the disease phenomenology) for comparison with
that of control subjects.
Morphometrics of First-Episode Schizophrenia
Figure 3. The effect of schizophrenia on brain shape at time of
diagnosis (A) and 3–5 years later (B). The patient consensus
configuration (white circles) is least squares superimposed on the
control consensus configuration (large dark circles). The average
unwarped image of the control scans is shown in the background.
The displacement vectors and thin plate spline grid show the
deformation of the control consensus into the patient consensus
with their magnitude exaggerated by a factor of 10 (the position
of the landmarks is not exaggerated).
Age Effects
Age effects were not statistically significant in any of the
tests performed with the complete patient sample; however, the probability for an age effect was much reduced
when limiting the syndromal group to the chronic schizophrenia and schizoaffective disorder (at follow-up time)
BIOL PSYCHIATRY
2000;48:398 – 405
403
Figure 4. The effect of gender on brain shape in patients (A) and
control subjects (B). The male consensus configuration (white
circles) is least squares superimposed on the female consensus
configuration (large dark circles). The average unwarped image
of the female subjects is shown in the background. The displacement vectors and thin plate spline grid show the deformation of
the female consensus into the male consensus with their magnitude exaggerated 10 times (the position of the landmarks is not
exaggerated).
patients in the overall test (n 5 65, F 5 3.217, p , .08)
and was significant when testing within that subset of
patients (n 5 43, F 5 4.7396, p , .011).
Discussion
Our study investigated midline regional brain shape differences between patients with schizophrenia and control sub-
404
BIOL PSYCHIATRY
2000;48:398 – 405
jects at the time of a first psychotic episode and 3–5 years
later. The significant diagnosis-by-time interaction term and
the significant time main effect within patients suggest that
some aspects of brain morphology do change in patients with
schizophrenia during the 3–5 years following the onset of
symptoms beyond the normal change found in nonpsychiatric individuals. The effect size for time in patients was twice
that in control subjects (F 5 0.776 and F 5 0.372, respectively) suggesting that the higher significance of time effect
in patients is not simply owing to their larger sample size.
The midline shape differences in brain morphology between patients and control subjects were not statistically
significant at the time of diagnosis and seem to become less
significant with the passage of time. Many of the differences
found by graphic inspection, however, are very similar to
those reported in DeQuardo et al (1996) and Bookstein
(1995), and these abnormalities seem to largely persist
throughout the follow-up period. The deformation grids of
these two studies—which shared the syndromal group—
show a relative expansion of the area between the splenium,
the colliculus, and the cerebellum (and a complementary
compression in the area between the colliculus, the pons, and
the chiasm) that is comparable to what we see in Figure 3 at
either time. In fact, these two studies reported an increase in
the relative distance between the superior colliculus and the
splenium (contributing to the said expansion) to be the most
significant schizophrenia abnormality using the 13-landmark
scheme. The relative thinning of the splenium in patients with
schizophrenia that was emphasized in Bookstein (1995) can
be detected in our contrast of patients and control subjects at
both diagnosis and follow-up times. The arching of the
corpus callosum and the relative backward movement of the
genu seen in Figure 3 (both A and B) are also comparable to
what was found in the two earlier studies. The thinning of the
splenium, backward movement of the genu, and arching of
the corpus callosum were better resolved in Bookstein
(1997), in which a variant of landmark analysis was devised
and applied to the data of Bookstein (1995) focusing on the
complete outline of the corpus callosum. DeQuardo et al
(1999) used a landmark configuration that partially overlaps
with those used in Bookstein (1995), DeQuardo et al (1996),
and the study discussed here to examine the neuroanatomy of
first-episode schizophrenia. The increased number of nonoverlapping landmarks between the configuration of DeQuardo et al (1999) and those of the three other studies
complicates the comparison; indeed, their deformation grids
and superimposition plots are less similar to those of the other
three. Still, the plots of DeQuardo et al (1999) clearly
reproduce the relative backward movement of the genu and
the relative expansion of the area between the splenium, the
cerebellum, and the superior colliculus that was found in the
other three studies. In rendering Figure 3, individuals were
chosen so that the gender composition within both patients
W.S. Gharaibeh et al
and control subjects was the same (.41 female, .59 male).
This proportion is intermediate between those used in Bookstein (1995) and DeQuardo et al (1996) (.50:.50, on one hand,
and those in DeQuardo et al [1999], .31:.69, on the other) and
is close enough to either to make cross-study graphic comparisons legitimate. This might be important because although the gender-by-diagnosis and gender-by-diagnosis-bytime interaction terms were not statistically significant, the
differences between patients and control subjects were quite
different when examined separately in men and women
(figures available upon request). The resolution of this
question requires a larger sample size than was available for
this study. Tibbo et al (1998) found no significant differences
in corpus callosum shape between 79 male patients with
schizophrenia and 65 male control subjects using a landmarkbased analysis (but see comment by DeQuardo 1999).
The presence of structural abnormalities at the time of
diagnosis and the significant change with time in patients, but
not in control subjects, are consistent with a neurodegenerative process that is closely associated with the onset of
symptoms. Data are now accumulating from MRI and other
studies that show that brain plasticity is different in patients
with schizophrenia than in normal control subjects (for a
review, see DeLisi 1999). Since it is now known that
neuronal growth and change occurs within the adult brain
(Gould et al 1999), one can speculate that any one aspect of
numerous steps in this process could be abnormal in individuals with schizophrenia, resulting in greater than normal cell
death, less cell growth, or both over time in some areas more
than others. It is worth noting, however, that all patients, and
none of the control subjects, were exposed to psychiatric
drugs during the follow-up period. This allows for an alternative interpretation of our results in which structural abnormalities are established much earlier in brain development
with the medication producing the observed rapid change
during the follow-up period. This interpretation is consistent
with the reduced statistical significance of differences between patients and control subjects at follow-up time. The
possibility of psychotropic drugs having a normalizing affect
on brain morphology has been suggested in the literature
(e.g., Keshavan 1998). A further clarification of the relationship between developmental processes and the establishment
of structural abnormalities could be obtained by longitudinal
studies that track the change in brain morphology with time
in control and individuals at high risk for schizophrenia prior
to the onset of symptoms, such as that of Lawrie et al (1999).
Our results also suggest the presence of a significant
sexual dimorphism in brain morphology that is stable
throughout the follow-up period but which seems to be
somewhat diminished in schizophrenia. The splenium in
women seems to be relatively thicker and more distant from
the cerebellum and the colliculus than in men, who in turn
seem to have a proportionally larger frontal cortical area.
Morphometrics of First-Episode Schizophrenia
Useful comparisons with previous studies (e.g., Hoff et al
1994, which overlaps the study sample of the work discussed
here) are precluded by differences in method of quantification (landmark vs. linear distances, area or volumetric analysis) and whether measurements were controlled for total
brain size.
The effect of age on brain shape was at best marginally
significant in the overall test but significant within the
subset of patients diagnosed with chronic schizophrenia or
schizoaffective disorder at time of follow-up. Still, separate TPS representations of age effects within patients and
control subjects showed similarities that are difficult to
interpret as anything other than a reflection of a common
brain aging process (figures available upon request). These
results suggest that taking age effects into consideration
when examining the dynamics of schizophrenia abnormalities was a warranted precaution.
Although not conclusive in their support of one etiologic
hypothesis over the others, our results could be instructive in
planning further tests of these hypotheses. The methodology
we have presented in this article seems to be especially suited
for application in studies that examine changes in brain
morphology with aging, progressive disease processes, medication, and other variables.
Supported by Ecological and Evolutionary Physiology (IBN-9728160)
program of the National Science Foundation (FJR, DES).
The landmark configurations were digitized using tpsDig, the GLS
superimposition and calculation of W were performed using tpsRelw and
the average unwarped images were produced using tpsSuper, both
written by F. James Rohlf. The figures that combine consensus configurations, TPS splines, displacement vectors, and unwarped average
images were produced in Morpheus et al, written by Dennis E. Slice. The
permutation procedures for multivariate testing were written and performed in MATLAB (MathWorks, Natick, MA) by Waleed Gharaibeh.
These programs are available from the SUNY at Stony Brook morphometrics website: http://life.bio.sunysb.edu/morph.
The authors thank Mary Kritzer for providing technical help with data
collection, Dean Adams for critiquing the morphometric and statistical
methodology, and Kerry Brown for help with the manuscript. The
constructive comments by three anonymous reviewers were also greatly
appreciated.
References
Bookstein FL (1991): Morphometric Tools for Landmark Data:
Geometry and Biology. Cambridge, UK: Cambridge University Press.
Bookstein FL (1995): How to produce a landmark point: The
statistical geometry of incompletely registered images. In:
Metler R, Wu A, Bookstein F, Green W, editors. Vision
Geometry IV. S.P.I.E. Proceedings, Vol 2573. Bellingham,
WA: SPIE Press, 266 –277.
Bookstein FL (1996): Standard formula for the uniform shape
component in landmark data. In: Marcus LF, Corti M, Loy A,
Naylor G, Slice D, editors. Advances in Morphometrics. New
York: Plenum Press, 153–168.
BIOL PSYCHIATRY
2000;48:398 – 405
405
Bookstein FL (1997): Landmark methods for forms without
landmarks: Localizing group differences in outline shape.
Med Image Analysis 1:225–243.
Bookstein FL (1998): Linear methods for nonlinear maps:
Procrustes fits, thin-plate splines, and the biometric analysis
of shape variability. In: Toga A, editor. Brain Warping. San
Diego: Academic Press, 157–181.
DeLisi LE (1999): Defining the course of brain structural change
and plasticity in schizophrenia. Psychiatry Res Neuroimaging
92:1–9.
DeLisi LE, Sakuma M, Kushner M, Finer DL, Huff AL, Crow TJ
(1997): Anomalous cerebral asymmetry and language processing in schizophrenia. Schizophr Bull 23:255–271.
DeQuardo JR (1999): Landmark analysis of corpus callosum
shape in schizophrenia. Biol Psychiatry 46:1712–1713.
DeQuardo JR, Bookstein FL, Green WDK, Brunberg JA, Tandon R
(1996): Spatial relationships of the neuroanatomic landmarks in
schizophrenia. Psychiatry Res Neuroimaging 67:81–95.
DeQuardo JR, Keshavan MS, Bookstein FL, Bagwell WW,
Green WDK, Sweeney JA, et al (1999): Landmark-based
morphometric analysis of first-episode schizophrenia. Biol
Psychiatry 45:1321–1328.
Goodall C (1991): Procrustes methods in the statistical analysis
of shape. J R Stat Soc B 53:285–339.
Gould E, Reeves AJ, Graziano MSA, Gross CG (1999): Neurogenesis in the neocortex of adult primates. Science 286:548 –552.
Hoff AL, Neal C, Kushner M, DeLisi LE (1994): Gender
differences in corpus callosum size in first-episode schizophrenia. Biol Psychiatry 35:913–919.
Keshavan MS, Haas GL, Kahn CE, Auilar E, Scooler NR,
Sweeney JA, et al (1998): Superior temporal gyrus and the
course of early schizophrenia: Progressive, static or reversible? J Psychiatr Res 32:161–167.
Lawrie SM, Whalley H, Kestelman JN, Abukmeil SS, Byrne M,
Hodges A, et al (1999): Magnetic resonance imaging of brain
in people at high risk of developing schizophrenia. Lancet
353:30 –33.
Lieberman JA (1999): Is schizophrenia a neurodegenerative
disorder? A clinical and neurobiological perspective. Biol
Psychiatry 46:729 –739.
Marcus LF, Corti M, Loy A, Naylor G, Slice D, editors (1996):
Advances in Morphometrics. New York: Plenum Press.
Pearlson GD, Marsh L (1999): Structural brain imaging in schizophrenia: A selective review. Biol Psychiatry 46:627– 649.
Rohlf FJ (1999): Shape statistics: Procrustes superimpositions
and tangent spaces. J Classification 16:197–223.
Rohlf FJ, Slice D (1990): Extensions of the Procrustes method
for the optimal superimposition of landmarks. Syst Zool
39:40 –59.
Spitzer RL, Endicott J (1978): Schedule for Affective Disorders
and Schizophrenia (SADS). New York: New York State
Psychiatric Institute, Biometrics Research Department.
Tibbo P, Nopoulos P, Arndt S, Andreasen NC (1998): Corpus
callosum shape and size in male patients with schizophrenia.
Biol Psychiatry 44:405– 412.
Weinberger DR (1987): Implications of normal brain development for the pathogenesis of schizophrenia. Arch Gen Psychiatry 44:660 – 669.