Widespread Cerebral Gray Matter Volume Deficits in Schizophrenia Robert B.
Zipursky, MD;
Kelvin O. Lim, MD; Edith V.
Sullivan, PhD; Byron
Magnetic resonance imaging was used to investigate whether the structural brain differences commonly observed in patients with schizophrenia as compared with normal control subjects are specific to gray or white matter, and furthermore whether such abnormalities are localizable to circumscribed cortical regions. Accordingly, 22 patients meeting DSM-III-R criteria for schizophrenia and 20 healthy community volunteers, all 23 to 45 years old, received magnetic resonance imaging scans. Seven axial magnetic resonance imaging sections of 5-mm thickness were segmented into cerebrospinal fluid, gray matter, and white matter compartments and used for volumetric quantification. For the healthy control subjects, age correlated significantly with the percentage of all magnetic resonance imaging sections taken up by gray matter but not white matter. After correcting for the normal effect of age, the schizophrenic group was found to have significantly less gray matter than the control group but no difference in white matter; ventricular volume was 34% greater in the schizophrenic group. The schizophrenic group had less gray matter in all six cortical subregions analyzed; these differences attained statistical significance for all but the parietal measure. These findings have implications for studies of localized gray matter abnormalities and suggest that regional brain volume measurements need to be expressed in the context of possible widespread gray matter volume deficits \s=b\
schizophrenia.
in
(Arch Gen Psychiatry. 1992;49:195-205)
Kraepelin, reviewing the morbid anatomy of de¬ mentia praecox, stated that Emil has been shown that and disease the have in
"it
in
cortex
we
.
.
severe
widespread
of the nerve-tissue."1 Although he expressed this view almost a century ago, only recently have systematic quantitative in vivo and postmortem studies yielded con.
Accepted
for publication May 6, 1991. From the Departments of Psychiatry and Behavioral Sciences (Drs Zipursky, Lim, Sullivan, and Pfefferbaum) and Health Research and Policy (Dr Brown), Stanford (Calif) University School of Medicine; and the Psychiatry Service, Department of Veterans Affairs Medical Center, Palo Alto, Calif (Drs Zipursky, Lim, Sullivan, and Pfefferbaum). Dr Zipursky is now with the Clark Institute of Psychiatry, Department of Psychiatry, University of Toronto, Ontario. Reprint requests to Psychiatry Service (116A3), Department of Veterans Affairs Medical Center, 3801 Miranda Ave, Palo Alto, CA 94304 (Dr Pfefferbaum).
W.
Brown, PhD; Adolf Pfefferbaum, MD
if not consistent, evidence for neuroanatomical abnormalities in schizophrenia.2 The advent of modern structural brain imaging technologies, particularly com¬ puted tomography (CT) and magnetic resonance imaging (MRI), has now provided tools for in vivo investigation and renewed interest in identifying widespread as well as localized structural brain abnormalities in schizophrenia. The first CT report of increased ventricular volume in schizophrenics3 was soon confirmed by many others,4,5 reinforcing the significance of earlier neuropathological and pneumoencephalographic observations. It is now widely accepted that, as a group, patients with schizo¬ phrenia have larger lateral ventricles and cortical sulci than normal subjects of comparable age.4,6,7 There is also accumulating evidence that schizophrenic patients as a group may have smaller brain volume than control groups from similar cohorts.8"10 Because this effect is small, albeit significant, it may not be observable in every patient sample measured.11,12 Evidence from studies of recentonset cases of schizophrenia,13"17 longitudinal studies,18,19 and cross-sectional studies20 suggests that the structural brain abnormalities are present early in the course of schizophrenia and are not progressive.21 Magnetic resonance imaging is now being used to ex¬ tend the earlier CT findings and to pursue, in vivo, hypotheses of localized brain abnormalities generated by postmortem neuropathological studies. Regions com¬ monly studied include the temporal lobes,22,23 frontal lobes,10,12 limbic system,17 and corpus callosum.24 Recent postmortem studies of brains of schizophrenic patients25"27 have described the hippocampus, amygdala, and parahippocampal gyrus as being smaller in schizophrenic pa¬ tients relative to controls. Using MRI, Suddath et al23 found diminished temporal lobe gray matter in patients with schizophrenia, which appeared to be statistically significant only in the central gray matter of the temporal lobe, corresponding anatomically with the region con¬ taining the anterior hippocampus and amygdala. Some investigators have also reported greater involvement of left than right temporal lobe structures.28 Despite the growing number of neuroimaging studies using improved acquisition and quantification tech¬ niques, most of these studies have focused on specific re¬ gions and structures rather than the brain as a whole. If widespread diffuse changes exist, specific brain struc¬ tures should be studied within the context of globally al-
verging,
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
tered brain structure. Some CT studies of cortical sulcal volume support the notion that brain abnormalities in schizophrenia are in fact widespread.6 In a meta-analysis, Raz and Raz4 concluded that sulci are equally enlarged in both hemispheres as well as in both the anterior and pos¬ terior regions of the brain, implying broadly distributed tissue loss. Other CT studies, however, either have not reported on sulcal changes or have found neither regional nor global sulcal enlargement. Such negative findings may, however, be due to the difficulty in measuring sul¬ cal volume without using computerized quantification
techniques.4,29
An additional consideration is whether brain changes are specific to particular types of brain tissue (ie, gray matter or white matter). A recent postmortem study30 de¬ scribed cortical and subcortical gray matter reductions in schizophrenics relative to age- and sex-matched controls and the lack of cerebral white matter involvement. In vivo studies have not yet examined gray matter changes throughout the brain but rather have focused on specific regions and structures of interest. In normal aging, there are increases in ventricular and sulcal size,31 which recent MRI studies have revealed to be primarily at the expense of gray matter, with relative sparing of white matter.32,33 Gray matter reduction has begun by early adolescence34 and appears to continue throughout the adult age span. An in vivo investigation of gray matter status in schizophrenia therefore needs to take normal age effects into account. Age regression models have proved useful for this purpose in CT stud¬ ies.6 We now report results of an MRI study in which we used an age regression model to investigate whether the cerebral tissue reduction in schizophrenia is localized to specific types of brain tissue, ie, gray matter vs white matter, and to particular broadly defined cortical regions. Despite intensive investigation, it has been difficult to establish correlates of the structural brain abnormalities found in schizophrenia.4,21 This failure may reflect both the underlying complexity of the clinical syndrome of schizophrenia as well as the difficulties in resolving the many critical methodological problems involved in neuroimaging studies.29 Although our current study was de¬ signed primarily to document brain structural abnormal¬ ities in a small sample of patients with schizophrenia, we did test hypotheses about the relationship between mea¬ sures of the prefrontal cortex and temporal lobes and negative and positive symptoms, respectively. Two pre¬ viously established clusters of symptoms derived from the Brief Psychiatric Rating Scale (BPRS) were used35 to measure positive and negative symptoms: the Thinking halluci¬ Disturbance factor (conceptual natory behavior, and unusual thought content), and the Withdrawal-Retardation factor (emotional withdrawal, motor retardation, and blunted affect). We predicted that higher scores for the Thinking Disturbance factor would be associated with greater abnormalities in the temporal but not frontal lobe measures and that higher scores on the Withdrawal-Retardation factor would be associated with greater abnormalities in the prefrontal but not tem¬
disorganization,
poral measures.
PATIENTS AND METHODS Patients included 22 right-handed male veterans of the United States armed services, Patients With
Schizophrenia.—The patients
Age,
y Mean ± SD
Range
Height,
Schizophrenics
Controls
34.1 ±5.5 23-45
36.2 ±7.0 23-45
178.4 ±8.4 160-193
177.1 ±6.0 168-188
cm
Mean ± SD
Range Weight, kg
Mean ± SD
75.3±10.1t 57-100
Range Race W
83.7±13.9 69-118
19 3 0
Other
17 2 1
Handedness
Questionnaire
score
Mean ± SD
Range
18.7±5.6 14-34
20.9 + 5.3 14-34
64.8±64.8 0-248
48.5 ±50.4 3-196
12.5±2.2§
15.8±2.3 12-21
Lifetime alcohol
consumption,
kg ethanol
Mean ± SD
Range Education, y
Mean ± SD
9-18
Range ART IQ Mean ± SD
106.7±9.1 89-121
Range
Vocabulary age-scaled score
107.4±7.5|| 90-121
WAIS-R
Mean ± SD
9.4±2.5t 6-15
Range
11.4±2.9 7-18
*NART indicates National Adult Reading Test; WAIS-R, Wechsler Adult Intelligence Test-Revised. tP=s.05 vs control group. tBased on data from 16 of 20 controls. §P=s.001 vs control group. ¡Based on data from 19 of 20 controls.
23 to 45 years of age, meeting DSM-III-R criteria for the diagno¬ sis of they were recruited from the inpatient psy¬ chiatric service of the Department of Veterans Affairs Medical
schizophrenia;
Center, Palo Alto, Calif (Table 1). All but one were inpatients on
the unlocked voluntary ward of the Mental Health Clinical Re¬ search Center (MHCRC); the remaining patient was housed on a locked ward at the time of the study but had previously been a patient on the MHCRC ward. Informed consent for participa¬ tion in this study was obtained for all subjects. Patients meeting criteria for current DSM-III-R Alcohol Abuse or having ever met criteria for DSM-III-R Alcohol Dependence were excluded from the study. Other exclusion factors were a history of significant medical illness, head injury resulting in loss of consciousness for greater than 30 minutes, or current DSM-
drug dependence. diagnoses were determined by clinical inter¬ view by a psychiatrist, clinical psychologist, or psychiatric research fellow. The diagnosis for all but one of the subjects was reached by consensus of two of the above clinicians; the remain¬ ing patient's diagnosis was reached using an earlier system for establishing diagnoses at our center in which the DSM-III-R di¬ agnosis was determined by a single clinician. All patients met DSM-III-R criteria for chronic schizophrenia and were subtyped as follows: paranoid (n 5), disorganized (n 4), undifferenriated (n 10), and residual (n 3). In addition to receiving DSMIII-R diagnoses, all patients received diagnoses using Research III-R
The DSM-III-R
=
=
=
=
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Diagnostic Criteria (RDC).3* The Schedule for Affective Disor¬ ders and Schizophrenia (SADS)37 was administered to each pa¬ tient by a trained research assistant or research nurse. The RDC diagnoses were established by a process of consensus involving the same clinicians involved in the DSM-III-R diagnosis along with the SADS interviewer. Of the 22 patients in this study, 19 met RDC criteria for chronic schizophrenia of the following sub¬ types: paranoid (n 10), disorganized (n 4), undifferentiated (n 2), and residual (n 3). The three remaining patients met RDC criteria for schizoaffective disorder with the following sub¬ types: chronic, depressed type, mainly schizophrenic; chronic, depressed type, other; and subchronic, manic type, mainly =
=
=
=
schizophrenic. mean
The (±SD) duration of illness for the schizophrenic pa¬ tients was 11.3±6.1 years (range, 1 to 25 years), with a mean age at onset of 22.8±4.3 years (range, 17 to 32 years). All patients had been treated in the past with antipsychotic medications, except for one patient, who received his first dose the day before his MRI scan. One patient had been a subject in an earlier CT study of schizophrenia conducted at this laboratory.6 The clinical con¬
dition of each patient was evaluated when the patient was con¬ sidered to be clinically stable using the BPRS38 at a mean interval of 8 ±10 days (range, 1 to 47 days) from the time of MRI scanning (Table 1). The BPRS was administered by two raters with estab¬ lished reliability, and the average score was used for analysis. The BPRS yielded three scores: the total score for all 18 items combined (possible scores of 18 to 126), the score for the Withdrawal-Retardation factor (possible scores of 3 to 21), and the score for the Thinking Disturbance factor (possible scores of 3 to 21). In all cases, higher scores reflected greater symptom se¬ verity. The mean total BPRS score was 41.4±9.6 (range, 25.5 to 67.6), while the means for the Thinking Disturbance and Withdrawal-Retardation factors were 8.2±3.1 (range, 3 to 12) and 8.2±2.5 (range, 3.5 to 11.5), respectively. Healthy Control Subjects.—Healthy control subjects in¬ cluded 20 right-handed male veterans, also between 23 and 45 years of age, recruited from the local community. The exclusion criteria used for patients were also applied to con¬ trols. Of the 20 subjects, 11 were recruited specifically to serve as controls for MRI studies of schizophrenia and alco¬ holism being conducted at this laboratory. These subjects were interviewed by a psychiatrist or psychiatric research fellow using the SADS-Lifetime version as a guideline. An additional nine subjects were recruited as reference subjects for a separate protocol that included a lumbar puncture, sleep studies, cognitive testing, a CT scan, and an MRI scan. These nine subjects were given the SADS by a trained research assistant and also underwent a clinical interview by a psychi¬ atrist, psychologist, or psychiatric research fellow before en¬ tering the study. None of the controls had participated in our previously reported CT study of schizophrenia.6 Demographic Data for Patient and Control Groups.—The two groups were matched for age in mean and range. All subjects were self-declared right-handers and were matched on quantified handedness.39 The National Adult Reading Test (NART)40 and the age-scaled Vocabulary subtest of the Wechsler Adult Intelligence Scale-Revised (WAIS-R)41 were administered to each patient at a time when they were clini¬ cally stable (Table 1). The NART is considered to reflect premorbid IQ while the Vocabulary subtest provides an estimate of current verbal IQ and is commonly related to educational experience. Although the schizophrenic group had signifi¬ cantly fewer years of education than the control group, NART scores were equivalent, a match we considered more meaningful than years of formal education because the latter is frequently affected by premature dropout from school brought on by prodromal symptoms of schizophrenia. The patients had lower age-scaled Vocabulary scores than con¬ trols, which could reflect the difference in educational experi¬ ence or, alternatively, an illness-related deficit. Lifetime alcohol consumption, quantified for each subject us¬ ing a modification of Skinner's procedure,42 did not differ signif-
icantly between the two groups (Table 1). Although the groups were of similar height, the patient group weighed significantly less than the control group
(Table 1).
Procedure MRI Scan Acquisition.—All patients underwent scanning using a 1.5-T MRI scanner (General Electric Signa, Milwaukee, Wis). Scanning values and procedures used in this study have been previously described in detail by Lim and Pfefferbaum.43 Axial MRI images were acquired using a spin-echo sequence with a field of view of 24 cm. Acquisition was gated to every other cardiac cycle for an effective repetition time of 1600 to 2200 mil¬ liseconds, with one excitation for each of 256 phase encodes. Early and late echoes were obtained at 20 and 80 milliseconds, respectively. All axial images were oriented in an oblique plane, perpendicular to the sagittal plane, and passed through the an¬ terior and posterior commissures, which were identified from a midsagittal image. Beginning interiorly at the base of the pons, 17 to 20 sections 5 mm thick were collected with a 2.5-mm interslice skip to avoid "aliasing" errors. The MRI scans of all patients and control subjects were inde¬ pendently evaluated by a clinical neuroradiologist. No subject was excluded on the basis of focal lesions or structural deformity. MRI Section Selection Criteria and Quantification.—All im¬ ages were stored on magnetic tape, transferred to a laboratory minicomputer, and coded to allow processing to be performed "blind" to the subject's identity, age, diagnosis, and neuroradiologist's report. For each scan, the most inferior section above the level of the orbits, where the anterior horns of the lateral ventricles could be seen, was identified as an index section. We chose this level as a standardized anchor point across all subjects because the orbits and frontal horns provide reliably identifiable landmarks despite varying degrees of ventricular and sulcal en¬ largement found in different subjects. Eight consecutive sec¬ tions, beginning at the section immediately inferior to the index section, were analyzed for each subject. The lowest section was used only for quantification of the third ventricle. The superior seven sections were used in volumetric quantification. Sections superior to this were excluded because, in many subjects, they contained only fluid and partially volumed cortical tissue, mak¬ ing quantification unreliable. Sections inferior to the index sec¬ tion were also excluded from all but the third-ventricle analysis because their irregularly shaped skull boundaries precluded stripping of skull and soft-tissue pixels from the images, a pre¬ processing requirement for automated tissue segmentation. Three-Compartment Image Segmentation.—Each of the MRI sections was segmented into cerebrospinal fluid (CSF), gray matter, and white matter compartments using a semiautomated image analysis technique developed in this laboratory.43 The technique consisted of the following steps. First, skull margins were identified, and skull and all pixels peripheral to it were stripped from each image. Then, to enhance CSF-tissue contrast, a composite image was created by subtracting late from early echo images. This image was filtered using a homomorphic dig¬ ital filter to reduce the effects of radiofrequency inhomogeneity (which results in spatial variations in baseline pixel intensity).43 Trained research assistants blind to subject identity, age, and diagnosis identified the image intensity value above which all pixels could be considered tissue and below which all pixels could be considered CSF. Interrater reliability for the different measures was assessed by having one of us (R.B.Z.) and each of four research assistants analyze a set of 15 MRI scans involving a total of 120 sections. Measures were then entered into a data set and randomly sorted across the five raters to ensure that the correlations measured agreement rather than simple association.44 A correlation matrix was calculated for each measure using the Spearman Rank Cor¬ relation Coefficient, and the mean value from the correlation matrix was calculated. Interrater reliability for the CSF-tissue "thresholding" was .98. Raters also delineated the third ven¬ tricle on the section in which it appeared largest (Table 2) with =
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Regions-of-lnterest Intracranial volume index (IVI) Lateral ventricles
Third ventricle
Subcortex
Total cortex
Definition Total volume of all fluid and tissue voxels in sections 2 through 8 Sum of all fluid voxels in inner 55% of all sections on which the lateral ventricles could be identified beginning with the index section (section 2 in Fig 1), expressed as a percentage of IVI Identified on the one section on which it appeared largest; an irregularly shaped region of interest was traced around the third ventricle and some of the adjacent thalamus but was limited anteriorly by the anterior horns of the lateral ventricles and posteriorly by the pineal gland; the third ventricle volume was defined as the sum of ail fluid voxels in this region of interest and expressed as a percentage of IVI Fluid and tissue voxels in inner 55% of the index section; gray matter and white matter volumes are expressed as percentages of total voxels in this volume Fluid and tissue voxels in outer 45% of the most superior 7 sections (sections 2 through 8 in Fig 1); cerebrospinal fluid; gray matter and white matter volumes are expressed as percentages of total voxels in this volume
reliability of p= .95. Midlines were drawn to sepa¬ right hemispheres of the brain for all brain sec¬ tions using a series of 17 points for all sections inferior to the centrum semiovale and nine points for all sections located more superiorly. Interrater reliability of the left-right hemisphere ra¬ an
interrater
rate the left and
tios for each section, based on these midlines, was .99. For each scan, the most anterior and posterior points of the corpus callosum were also identified. To enhance gray-white contrast, another set of composite im¬ ages was created by adding together pixel intensity values for the early and late echo images. These composite images were also filtered to reduce radiofrequency inhomogeneity. Gray-white segmentation was achieved using an automated procedure32 based on a nonparametric histogram analysis technique.45 All pixels identified as tissue during manual thresholding were subjected to histogram analysis, which determined the thresh¬ old value separating gray from white matter (Fig 1). This auto¬ matic gray-white thresholding technique was validated against our previously reported operator-driven thresholding method.43 Interrater reliability for the manual gray-white separation tech¬ nique was =.73, while comparing the automated technique with this manual technique yielded a value of p= .79. Regional Divisions of Segmented Images. —The images were divided according to anatomical landmarks and a priori geomet¬ ric rules in an effort to achieve standardized regional divisions of the brain images. To this end, each segmented brain section was first divided into an inner 55% region (to facilitate quanti¬ fication of central CSF including the ventricles) and an outer 45% (to facilitate quantification of the cortex). These proportions were empirically determined to maximize differentiation of ventricu¬ lar from cortical sulcal fluid.31 Using these divisions, estimates of the volume of CSF or white matter and gray matter were calcu-
lated for the intracranial volume, the lateral ventricles, the cere¬ bral cortex, and the subcortical region (Fig 1, Table 2). The total volume of all seven sections summed is referred to as the intra¬ cranial volume index (IVI). Each section was further divided into four regions by three coronal planes passing through (1) the most anterior extreme of the genu of the corpus callosum, (2) the most posterior extreme of the splenium of the corpus callosum, and (3) midway between them. The first of these planes was chosen to establish a bound¬ ary for the prefrontal region. The latter planes, though arbitrary, provided a more reliable basis for dividing cortical sections than specific cortical sulcal landmarks, which were difficult to estab¬ lish reliably on these images. These planes were projected through each section, perpendicular to the orientation of the ax¬ ial sections and the midline (Fig 1). The "total cortical" measure, which comprised the outer 45% of the seven MRI sections, was then subdivided empirically into the following anatomical mea¬ sures, by combining the areas labeled on Fig 1 across sections: prefrontal (a), frontotemporal (b), temporoparietal (c), frontal (d), parietal (e), and parieto-occipital (f). Within each of these regions, CSF, gray matter, and white matter voxels were summed bilaterally and then divided by the total volume of the region so that the measures could be expressed as the percent¬ age CSF, gray matter, and white matter in each region. This procedure was undertaken in order to minimize the effect of in¬ dividual differences in head size.
Statistical An age
age
on
Analysis
regression model was used to control for the effect of
all CSF, gray matter, and white matter measures.29 This
approach enables direct comparison of groups of different ages with the effect of age removed. Although the groups were of similar ages in this study, an age regression model reduces within-group variance and so enhances the ability to detect between-group differences. For each of the percentage CSF, gray matter, and white matter measures in all regions-of-interest, the values for controls were regressed against age. Values for the schizophrenics were then compared with the control regression and expressed as scores (the difference between the schizo¬ phrenic value and the control value predicted from the control regression, divided by the control SE of the regression). Patient scores were evaluated for statistical significance using a f sta¬ tistic of the form: [(mean y'-mean y)-i>(mean x'-mean x)] t=-
=
V
(mean a:'-mean
x)2}
-s^-)
where
mean y and mean y' are the mean percentage CSF, per¬ centage gTay matter, or percentage white matter measure for patients and controls, respectively; b, the slope of age vs percentage CSF, percentage gray matter, or percentage white matter for controls; Sp, the pooled SE of the estimate for the pa¬ tient and control regressions; Sxx, the sum of squared differences between the control ages and the mean age for controls; and n' and are the sample sizes for patients and controls, respectively.
Tests of differences in CSF measures were tested one-tailed while differences in gray matter and white matter measures were tested two-tailed, with =.05. The age regression model was compared with a more conventional statistical approach, analy¬ sis of covariance (ANCOVA), to control for age effects on these data. Correlations of global brain measures with demographic vari¬ ables were calculated with Pearson Product-Moment tests. Cor¬ relations of specific brain measures with clinical measures of the BPRS were assessed using the Spearman Rank Correlation test. To test whether the schizophrenics differed horn controls in in¬ tracranial volume, differences in IVI volume were assessed us¬ ing ANCOVA with height as a covariate. Because the multiplicity of statistical analyses used could re¬ sult in Type I errors, we have therefore explicitly described all
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Fig 1.—Final processed and segmented images for eight consecutive axial brain sections used for quantitative analysis (see "Patients and Methods" section). These brain images have been segmented into cerebrospinal fluid, gray matter, and white matter pixels. Definitions of regions-of-interest appear in Table 2. Section I or 2 was used to measure the third ventricle. Combinations from sections 2 to 8 were used for volumetric quantification of other regions-of-interest. Each segmented brain section was first divided into an inner 55% region, which included the ventricles and subcortical regions-of-interest, and an outer 45%, which included the cortical regions-of-interest. The cortical region was subdivided as follows: prefrontal (a), frontotemporal (b), temporoparietal (c), frontal (d), parietal (e), and parieto-occipital (f).
specifying multiple levels of statistical signifi¬ and have been conservative in interpreting the differences, taking this multiplicity into account. tests carried out
cance
mean z=
RESULTS Brain Volume Estimates of Normal
Control
ume was
Subjects
The percentage CSF, gray matter, and white matter for the IVI for controls is shown in Fig 2. The percentage gray matter was significantly negatively correlated with age (r= —.51, Ps.05), while the correlation of percentage CSF with age failed to reach statistical significance (r= .31, P^.10) and the percentage white matter was unrelated to age (r— .05). Third-ventricular volume was significantly correlated with age (r= .39, P=s.05), while lat¬ eral ventricular volume was not (r= .14). The percentage CSF, gray matter, and white matter for each of the brain regions stud¬ ied are presented in Table 3. —
Brain Volume Estimates of Schizophrenic Patients Within the schizophrenic group, the same pattern of agerelated changes in CSF, gray matter, and white matter were found; significant correlations were found between age and both CSF (r .42, P« .05) and IVI gray matter (r .48, P« .05) but not between age and IVI white matter (r .22). Neither lateral ventricular volume nor third-ventricular volume was signifi¬ cantly correlated with age in the schizophrenic group (r .07 and r— .26, respectively). The significant effect of age in the normal control group necessitated correcting the patients' MRI mea¬ sures for age. Accordingly, the raw percentage measures of CSF, gray matter, and white matter of the schizophrenic patients were converted to age-corrected scores derived from the control data (Fig 3). Mean percentage values and mean scores for schizo¬ phrenics for all CSF, gray matter, and white matter measures are listed in Tables 3 and 4, respectively. Mean scores for the con¬ trols are 0 for each measure. Values for schizophrenic patients that differed significantly from 0 are indicated in Table 4. For the brain as a whole (ie, IVI), the schizophrenic patients showed less gray matter (mean z= —1.53, P=£.001) but no significant differ=
=
in white matter (mean 0.55) or CSF (mean 0.48, P=s.25). Gray matter differences were apparent for both the left and right hemispheres of the TVI (mean z= —1.22, P=s.01, and enees
-
=
—
=
=
=
-1.62, P=£.001, respectively). Lateral ventricular vol¬ significantly larger in the schizophrenic group than in
the controls, while the differences in third-ventricular volume and total cortical CSF volume did not reach statistical significance. Cortical subregions were then analyzed to determine whether the global differences in gray matter could be localized. Gray matter volume scores were lower than expected from normal values in all six cortical subregions, with differences attaining statistical significance in all but the parietal region (area e). Re¬ gional cortical gray matter scores were significantly intercorrelated in 13 of 15 comparisons within the schizophrenic group and eight of 15 within the control group (Table 5). Percentage white matter of the schizophrenic group was significantly increased in the parieto-occipital region but did not differ from normal in any of the other brain areas. Cerebrospinal fluid volume was signif¬ icantly greater in the frontotemporal and temporoparietal re¬ gions (areas b and c) of the schizophrenic group. The ANCOVA, with age as a covariate, yielded results similar to those obtained from the age-regression analysis (Table 3). Diagnosis-by-age interactions were significant only for total cor¬ tical white matter and parieto-occipital CSF, thus precluding co¬ variate analysis. The schizophrenic group had significantly less gray matter over all seven sections (F 19.38, P=£.001), in the cortical region as a whole (F 10.62, P«s.01), in the subcortical region (F 6.25, P«£.05), and in each regional cortical measure except for the parietal region (Table 3). Ventricular volume was also significantly greater in the schizophrenic group than in the control group (F 5.30, P=£.05). White matter measures did not differ significantly between groups except in the parieto-occipital region, which was significantly greater in the schizophrenic as compared with the control group (F 5.28, P=s.05). As in the age regression analysis, cortical sulcal CSF volumes were signifi¬ cantly greater in the schizophrenic group only in the frontotem¬ poral and temporoparietal regions. When estimating intracranial volume (ie, IVI, the sum of all =
=
=
=
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
=
Intracranial volume index
Total cortical Subcortical
45 40
30 20
— -
25
30
40
35
45
44.6±3.5|| 47.3 ±3.4
44.7 ±5.0 43.4 ±3.9
47.2 + 5.7 50.7±4.0
29.2 + 5.4 27.7±3.5
18.4±10.7|| 50.6±6.6
31.1 ±9.1 31.8±9.0
...
...
33.2±2.9
17.8±9.0 16.3 ±6.5
46.5±5.0 49.5 ±5.3
35.7±5.2 34.2 ±3.5
Parietal
S C
14.7±6.4 14.7±5.8
46.8±4.4 47.6±5.9
38.4±3.5 37.7±3.5
Parieto-occipital
S C
17.3±8.2§
48.0±4.6 50.1 ±3.7
34.7±4.7||
S C
4.3±1.8||
S C
0.074±0.046 0.063±0.027
Third ventricle
50
35.0±3.7§
51.8±4.9* 37.0±5.5 55.2±3.9 37.1 ±4.3
Lateral ventricles
35
48.2±4.3 50.9±3.7
11.2±6.8||
:
sí-
16.8±6.9 15.9±4.7
S C S C
Frontal
55
42.4±2.6t 44.4±3.9 44.9 ±2.1 43.0±2.4
% White
55.2±5.8
Temporoparietal
(j 50
13.3±5.0 12.1 ±3.2
Gray
13.0±5.7
65-
>~
S C S C S C
%
23.6 ±9.7 21.6±6.5
Frontotemporal
60
% CSF
S C S C
Prefrontal
70
Group
7.8±3.6
19.6±8.0
30.3±6.4
3.2±1.0
*CSF indicates cerebrospinal fluid; S, schizophrenic; and C, control. Values are mean ±SD. tPss.001 vs control group. ^.01 vs control group. §Analysis of covariance was not calculated owing to differences in slopes between the two groups. || «.05 vs control group.
70 65 60 55 50-
not significantly correlated with duration of illness or age at onset of (r= .12 and r= .01, respectively). correlated Gray matter scores for the IVI were were
.-·.· I
45
schizophrenia
40 35
-
-
30 20
——
25
30
——
35
40
45
=
50
Age, y
Fig 2.—Age vs percentage of intracranial volume index made up of cerebrospinal fluid (CSF), gray matter, and white matter for 20 right-handed male community volunteers (control group). Each graph displays the linear regression line (solid line) as well as lines 1 and 2 SEs of the regression above and below the mean (dashed lines).
CSF, gray matter, and white matter pixels in the seven sections), the schizophrenic patients did not differ significantly from trols (542 vs 523 cm3, respectively; F 1.964).
con¬
=
Correlations of MRI Scores With Demographic and Behavioral Data
=
-
—
=
—
—
-
Thinking
=
The schizophrenic patients did not differ significantly from controls on estimates of total ufetime alcohol consumption (Ta¬ ble 1). In the schizophrenic group, IVI gray matter and white scores were not significantly correlated with the matter estimates of lifetime alcohol consumption (r= .07 and r= .06, respectively). The IVI gray matter scores for the schizophren¬ ics were not significantly correlated with years of education, NART IQ, or the WAIS-R Vocabulary age-scaled score (r= .16, r .05, r .02, respectively). Gray matter scores for the IVI =
significantly
-
with the BPRS Withdrawal-Retardation factor (p= .47, P«.05) but not with the total BPRS score (p= .21) or the BPRS Think¬ ing Disturbance factor (p .05). White matter scores for the IVI were not significantly correlated with any of the three BPRS measures: total score (p=.01), Withdrawal-Retardation factor (p= .16), or Thinking Disturbance factor (p= .26). The predicted association between scores on the Withdrawal-Retardation fac¬ tor and the prefrontal (area a) gray matter scores failed to reach the a priori significance level (p= .35, P= .053). The scores on the Disturbance factor were not significantly correlated with either the frontotemporal (area b) or temporoparietal (area c) gray matter scores (p=-.02 and -.22, respectively). Hypothesized brain-behavior relationships were further pur¬ sued by creating composite scores for the two frontal measures (areas a and d) and the two temporal measures (areas b and c). The association between the scores on the WithdrawalRetardation factor and the composite frontal measure was in the predicted direction (p .36, .051), while no association was apparent with the composite temporal measure (p= .11). The Disturbance factor was not significantly correlated with either the composite frontal or the temporal measure (p .17and p= —.14, respectively).
Thinking
-
=
COMMENT In this MRI study, quantitative estimates of gray mat¬ ter volume but not white matter volume were lower in all regions examined in patients with schizophrenia as com¬ pared with healthy control subjects. Thus, the diminished
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Regions-of-lnterest
CSF
Intracranial volume index
0.48 ±1.52
-1.53 ±1.24+ 0.55 ±1.56
Total cortical
0.31 ±1.41
-1.15±1.26
Gray
White
0.68±1.36
Subcortical Prefrontal
0.42 ±1.48
-1.12±1.47
0.41 ±1.51
Frontotemporal Temporoparietal
0.89±1.85§ -0.96±1.19 0.99±1.85§ -1.30±1.51
0.01 ±1.06
Frontal
0.41 ±1.39
-1.00+1.11* 0.46 ±1.47
0.09 ±1.04
-0.38 ±0.74
Parietal
Parieto-occipital
...
-0.21 ±0.90
-0.78±1.00§ 0.31 ±1.27
0.12 ±1.43 0.43 ±1.33
-0.74±1.12§ 0.65±0.68§
Lateral ventricles 1.02±1.78§ 0.55 ±1.74 Third ventricle *CSF indicates cerebrospinal fluid. Values are mean ± SD. Cerebrospinal fluid probability values are based on one-tailed tests; gray and white matter values are based on two-tailed tests. tPss.001 vs control group. «.01 vs control group. §P=s.05 vs control group.
Suddath et al23 and the postmortem study by Pakkenberg.30 In the Suddath et al study, the schizophrenic pa¬ tients had 67% larger ventricles and 19% smaller tempo¬ ral lobe gray matter volume than did controls. In the present study, the schizophrenic group's ventricular vol¬ ume was 34% larger than that of the control group. The schizophrenic group's gray matter volume was consis¬ tently smaller across the cortical subregions: 9% less for frontotemporal, 7% less for temporoparietal, 8% less for prefrontal, 8% less for frontal, 4% less for parietal, and 6% less for parieto-occipital. The percentage difference in gray matter volume for the cortical region as a whole was 7% less, which compares with the 12% mean reduction in cortical gray matter found in the postmortem study by
Pakkenberg.30
Fig 3.—Age vs percentage of intracranial volume index made up of cerebrospinal fluid (CSF), gray matter, and white matter for 22 right-handed male schizophrenic patients. The points for the schizophrenic patients are superimposed on the linear regression lines and the SEs of the regression lines determined for the control group (see Fig 2). The schizophrenic group differed significantly from the control group
not on the
3 and 4).
on
the percentage gray matter (P
Zipursky, MD;
Kelvin O. Lim, MD; Edith V.
Sullivan, PhD; Byron
Magnetic resonance imaging was used to investigate whether the structural brain differences commonly observed in patients with schizophrenia as compared with normal control subjects are specific to gray or white matter, and furthermore whether such abnormalities are localizable to circumscribed cortical regions. Accordingly, 22 patients meeting DSM-III-R criteria for schizophrenia and 20 healthy community volunteers, all 23 to 45 years old, received magnetic resonance imaging scans. Seven axial magnetic resonance imaging sections of 5-mm thickness were segmented into cerebrospinal fluid, gray matter, and white matter compartments and used for volumetric quantification. For the healthy control subjects, age correlated significantly with the percentage of all magnetic resonance imaging sections taken up by gray matter but not white matter. After correcting for the normal effect of age, the schizophrenic group was found to have significantly less gray matter than the control group but no difference in white matter; ventricular volume was 34% greater in the schizophrenic group. The schizophrenic group had less gray matter in all six cortical subregions analyzed; these differences attained statistical significance for all but the parietal measure. These findings have implications for studies of localized gray matter abnormalities and suggest that regional brain volume measurements need to be expressed in the context of possible widespread gray matter volume deficits \s=b\
schizophrenia.
in
(Arch Gen Psychiatry. 1992;49:195-205)
Kraepelin, reviewing the morbid anatomy of de¬ mentia praecox, stated that Emil has been shown that and disease the have in
"it
in
cortex
we
.
.
severe
widespread
of the nerve-tissue."1 Although he expressed this view almost a century ago, only recently have systematic quantitative in vivo and postmortem studies yielded con.
Accepted
for publication May 6, 1991. From the Departments of Psychiatry and Behavioral Sciences (Drs Zipursky, Lim, Sullivan, and Pfefferbaum) and Health Research and Policy (Dr Brown), Stanford (Calif) University School of Medicine; and the Psychiatry Service, Department of Veterans Affairs Medical Center, Palo Alto, Calif (Drs Zipursky, Lim, Sullivan, and Pfefferbaum). Dr Zipursky is now with the Clark Institute of Psychiatry, Department of Psychiatry, University of Toronto, Ontario. Reprint requests to Psychiatry Service (116A3), Department of Veterans Affairs Medical Center, 3801 Miranda Ave, Palo Alto, CA 94304 (Dr Pfefferbaum).
W.
Brown, PhD; Adolf Pfefferbaum, MD
if not consistent, evidence for neuroanatomical abnormalities in schizophrenia.2 The advent of modern structural brain imaging technologies, particularly com¬ puted tomography (CT) and magnetic resonance imaging (MRI), has now provided tools for in vivo investigation and renewed interest in identifying widespread as well as localized structural brain abnormalities in schizophrenia. The first CT report of increased ventricular volume in schizophrenics3 was soon confirmed by many others,4,5 reinforcing the significance of earlier neuropathological and pneumoencephalographic observations. It is now widely accepted that, as a group, patients with schizo¬ phrenia have larger lateral ventricles and cortical sulci than normal subjects of comparable age.4,6,7 There is also accumulating evidence that schizophrenic patients as a group may have smaller brain volume than control groups from similar cohorts.8"10 Because this effect is small, albeit significant, it may not be observable in every patient sample measured.11,12 Evidence from studies of recentonset cases of schizophrenia,13"17 longitudinal studies,18,19 and cross-sectional studies20 suggests that the structural brain abnormalities are present early in the course of schizophrenia and are not progressive.21 Magnetic resonance imaging is now being used to ex¬ tend the earlier CT findings and to pursue, in vivo, hypotheses of localized brain abnormalities generated by postmortem neuropathological studies. Regions com¬ monly studied include the temporal lobes,22,23 frontal lobes,10,12 limbic system,17 and corpus callosum.24 Recent postmortem studies of brains of schizophrenic patients25"27 have described the hippocampus, amygdala, and parahippocampal gyrus as being smaller in schizophrenic pa¬ tients relative to controls. Using MRI, Suddath et al23 found diminished temporal lobe gray matter in patients with schizophrenia, which appeared to be statistically significant only in the central gray matter of the temporal lobe, corresponding anatomically with the region con¬ taining the anterior hippocampus and amygdala. Some investigators have also reported greater involvement of left than right temporal lobe structures.28 Despite the growing number of neuroimaging studies using improved acquisition and quantification tech¬ niques, most of these studies have focused on specific re¬ gions and structures rather than the brain as a whole. If widespread diffuse changes exist, specific brain struc¬ tures should be studied within the context of globally al-
verging,
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
tered brain structure. Some CT studies of cortical sulcal volume support the notion that brain abnormalities in schizophrenia are in fact widespread.6 In a meta-analysis, Raz and Raz4 concluded that sulci are equally enlarged in both hemispheres as well as in both the anterior and pos¬ terior regions of the brain, implying broadly distributed tissue loss. Other CT studies, however, either have not reported on sulcal changes or have found neither regional nor global sulcal enlargement. Such negative findings may, however, be due to the difficulty in measuring sul¬ cal volume without using computerized quantification
techniques.4,29
An additional consideration is whether brain changes are specific to particular types of brain tissue (ie, gray matter or white matter). A recent postmortem study30 de¬ scribed cortical and subcortical gray matter reductions in schizophrenics relative to age- and sex-matched controls and the lack of cerebral white matter involvement. In vivo studies have not yet examined gray matter changes throughout the brain but rather have focused on specific regions and structures of interest. In normal aging, there are increases in ventricular and sulcal size,31 which recent MRI studies have revealed to be primarily at the expense of gray matter, with relative sparing of white matter.32,33 Gray matter reduction has begun by early adolescence34 and appears to continue throughout the adult age span. An in vivo investigation of gray matter status in schizophrenia therefore needs to take normal age effects into account. Age regression models have proved useful for this purpose in CT stud¬ ies.6 We now report results of an MRI study in which we used an age regression model to investigate whether the cerebral tissue reduction in schizophrenia is localized to specific types of brain tissue, ie, gray matter vs white matter, and to particular broadly defined cortical regions. Despite intensive investigation, it has been difficult to establish correlates of the structural brain abnormalities found in schizophrenia.4,21 This failure may reflect both the underlying complexity of the clinical syndrome of schizophrenia as well as the difficulties in resolving the many critical methodological problems involved in neuroimaging studies.29 Although our current study was de¬ signed primarily to document brain structural abnormal¬ ities in a small sample of patients with schizophrenia, we did test hypotheses about the relationship between mea¬ sures of the prefrontal cortex and temporal lobes and negative and positive symptoms, respectively. Two pre¬ viously established clusters of symptoms derived from the Brief Psychiatric Rating Scale (BPRS) were used35 to measure positive and negative symptoms: the Thinking halluci¬ Disturbance factor (conceptual natory behavior, and unusual thought content), and the Withdrawal-Retardation factor (emotional withdrawal, motor retardation, and blunted affect). We predicted that higher scores for the Thinking Disturbance factor would be associated with greater abnormalities in the temporal but not frontal lobe measures and that higher scores on the Withdrawal-Retardation factor would be associated with greater abnormalities in the prefrontal but not tem¬
disorganization,
poral measures.
PATIENTS AND METHODS Patients included 22 right-handed male veterans of the United States armed services, Patients With
Schizophrenia.—The patients
Age,
y Mean ± SD
Range
Height,
Schizophrenics
Controls
34.1 ±5.5 23-45
36.2 ±7.0 23-45
178.4 ±8.4 160-193
177.1 ±6.0 168-188
cm
Mean ± SD
Range Weight, kg
Mean ± SD
75.3±10.1t 57-100
Range Race W
83.7±13.9 69-118
19 3 0
Other
17 2 1
Handedness
Questionnaire
score
Mean ± SD
Range
18.7±5.6 14-34
20.9 + 5.3 14-34
64.8±64.8 0-248
48.5 ±50.4 3-196
12.5±2.2§
15.8±2.3 12-21
Lifetime alcohol
consumption,
kg ethanol
Mean ± SD
Range Education, y
Mean ± SD
9-18
Range ART IQ Mean ± SD
106.7±9.1 89-121
Range
Vocabulary age-scaled score
107.4±7.5|| 90-121
WAIS-R
Mean ± SD
9.4±2.5t 6-15
Range
11.4±2.9 7-18
*NART indicates National Adult Reading Test; WAIS-R, Wechsler Adult Intelligence Test-Revised. tP=s.05 vs control group. tBased on data from 16 of 20 controls. §P=s.001 vs control group. ¡Based on data from 19 of 20 controls.
23 to 45 years of age, meeting DSM-III-R criteria for the diagno¬ sis of they were recruited from the inpatient psy¬ chiatric service of the Department of Veterans Affairs Medical
schizophrenia;
Center, Palo Alto, Calif (Table 1). All but one were inpatients on
the unlocked voluntary ward of the Mental Health Clinical Re¬ search Center (MHCRC); the remaining patient was housed on a locked ward at the time of the study but had previously been a patient on the MHCRC ward. Informed consent for participa¬ tion in this study was obtained for all subjects. Patients meeting criteria for current DSM-III-R Alcohol Abuse or having ever met criteria for DSM-III-R Alcohol Dependence were excluded from the study. Other exclusion factors were a history of significant medical illness, head injury resulting in loss of consciousness for greater than 30 minutes, or current DSM-
drug dependence. diagnoses were determined by clinical inter¬ view by a psychiatrist, clinical psychologist, or psychiatric research fellow. The diagnosis for all but one of the subjects was reached by consensus of two of the above clinicians; the remain¬ ing patient's diagnosis was reached using an earlier system for establishing diagnoses at our center in which the DSM-III-R di¬ agnosis was determined by a single clinician. All patients met DSM-III-R criteria for chronic schizophrenia and were subtyped as follows: paranoid (n 5), disorganized (n 4), undifferenriated (n 10), and residual (n 3). In addition to receiving DSMIII-R diagnoses, all patients received diagnoses using Research III-R
The DSM-III-R
=
=
=
=
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Diagnostic Criteria (RDC).3* The Schedule for Affective Disor¬ ders and Schizophrenia (SADS)37 was administered to each pa¬ tient by a trained research assistant or research nurse. The RDC diagnoses were established by a process of consensus involving the same clinicians involved in the DSM-III-R diagnosis along with the SADS interviewer. Of the 22 patients in this study, 19 met RDC criteria for chronic schizophrenia of the following sub¬ types: paranoid (n 10), disorganized (n 4), undifferentiated (n 2), and residual (n 3). The three remaining patients met RDC criteria for schizoaffective disorder with the following sub¬ types: chronic, depressed type, mainly schizophrenic; chronic, depressed type, other; and subchronic, manic type, mainly =
=
=
=
schizophrenic. mean
The (±SD) duration of illness for the schizophrenic pa¬ tients was 11.3±6.1 years (range, 1 to 25 years), with a mean age at onset of 22.8±4.3 years (range, 17 to 32 years). All patients had been treated in the past with antipsychotic medications, except for one patient, who received his first dose the day before his MRI scan. One patient had been a subject in an earlier CT study of schizophrenia conducted at this laboratory.6 The clinical con¬
dition of each patient was evaluated when the patient was con¬ sidered to be clinically stable using the BPRS38 at a mean interval of 8 ±10 days (range, 1 to 47 days) from the time of MRI scanning (Table 1). The BPRS was administered by two raters with estab¬ lished reliability, and the average score was used for analysis. The BPRS yielded three scores: the total score for all 18 items combined (possible scores of 18 to 126), the score for the Withdrawal-Retardation factor (possible scores of 3 to 21), and the score for the Thinking Disturbance factor (possible scores of 3 to 21). In all cases, higher scores reflected greater symptom se¬ verity. The mean total BPRS score was 41.4±9.6 (range, 25.5 to 67.6), while the means for the Thinking Disturbance and Withdrawal-Retardation factors were 8.2±3.1 (range, 3 to 12) and 8.2±2.5 (range, 3.5 to 11.5), respectively. Healthy Control Subjects.—Healthy control subjects in¬ cluded 20 right-handed male veterans, also between 23 and 45 years of age, recruited from the local community. The exclusion criteria used for patients were also applied to con¬ trols. Of the 20 subjects, 11 were recruited specifically to serve as controls for MRI studies of schizophrenia and alco¬ holism being conducted at this laboratory. These subjects were interviewed by a psychiatrist or psychiatric research fellow using the SADS-Lifetime version as a guideline. An additional nine subjects were recruited as reference subjects for a separate protocol that included a lumbar puncture, sleep studies, cognitive testing, a CT scan, and an MRI scan. These nine subjects were given the SADS by a trained research assistant and also underwent a clinical interview by a psychi¬ atrist, psychologist, or psychiatric research fellow before en¬ tering the study. None of the controls had participated in our previously reported CT study of schizophrenia.6 Demographic Data for Patient and Control Groups.—The two groups were matched for age in mean and range. All subjects were self-declared right-handers and were matched on quantified handedness.39 The National Adult Reading Test (NART)40 and the age-scaled Vocabulary subtest of the Wechsler Adult Intelligence Scale-Revised (WAIS-R)41 were administered to each patient at a time when they were clini¬ cally stable (Table 1). The NART is considered to reflect premorbid IQ while the Vocabulary subtest provides an estimate of current verbal IQ and is commonly related to educational experience. Although the schizophrenic group had signifi¬ cantly fewer years of education than the control group, NART scores were equivalent, a match we considered more meaningful than years of formal education because the latter is frequently affected by premature dropout from school brought on by prodromal symptoms of schizophrenia. The patients had lower age-scaled Vocabulary scores than con¬ trols, which could reflect the difference in educational experi¬ ence or, alternatively, an illness-related deficit. Lifetime alcohol consumption, quantified for each subject us¬ ing a modification of Skinner's procedure,42 did not differ signif-
icantly between the two groups (Table 1). Although the groups were of similar height, the patient group weighed significantly less than the control group
(Table 1).
Procedure MRI Scan Acquisition.—All patients underwent scanning using a 1.5-T MRI scanner (General Electric Signa, Milwaukee, Wis). Scanning values and procedures used in this study have been previously described in detail by Lim and Pfefferbaum.43 Axial MRI images were acquired using a spin-echo sequence with a field of view of 24 cm. Acquisition was gated to every other cardiac cycle for an effective repetition time of 1600 to 2200 mil¬ liseconds, with one excitation for each of 256 phase encodes. Early and late echoes were obtained at 20 and 80 milliseconds, respectively. All axial images were oriented in an oblique plane, perpendicular to the sagittal plane, and passed through the an¬ terior and posterior commissures, which were identified from a midsagittal image. Beginning interiorly at the base of the pons, 17 to 20 sections 5 mm thick were collected with a 2.5-mm interslice skip to avoid "aliasing" errors. The MRI scans of all patients and control subjects were inde¬ pendently evaluated by a clinical neuroradiologist. No subject was excluded on the basis of focal lesions or structural deformity. MRI Section Selection Criteria and Quantification.—All im¬ ages were stored on magnetic tape, transferred to a laboratory minicomputer, and coded to allow processing to be performed "blind" to the subject's identity, age, diagnosis, and neuroradiologist's report. For each scan, the most inferior section above the level of the orbits, where the anterior horns of the lateral ventricles could be seen, was identified as an index section. We chose this level as a standardized anchor point across all subjects because the orbits and frontal horns provide reliably identifiable landmarks despite varying degrees of ventricular and sulcal en¬ largement found in different subjects. Eight consecutive sec¬ tions, beginning at the section immediately inferior to the index section, were analyzed for each subject. The lowest section was used only for quantification of the third ventricle. The superior seven sections were used in volumetric quantification. Sections superior to this were excluded because, in many subjects, they contained only fluid and partially volumed cortical tissue, mak¬ ing quantification unreliable. Sections inferior to the index sec¬ tion were also excluded from all but the third-ventricle analysis because their irregularly shaped skull boundaries precluded stripping of skull and soft-tissue pixels from the images, a pre¬ processing requirement for automated tissue segmentation. Three-Compartment Image Segmentation.—Each of the MRI sections was segmented into cerebrospinal fluid (CSF), gray matter, and white matter compartments using a semiautomated image analysis technique developed in this laboratory.43 The technique consisted of the following steps. First, skull margins were identified, and skull and all pixels peripheral to it were stripped from each image. Then, to enhance CSF-tissue contrast, a composite image was created by subtracting late from early echo images. This image was filtered using a homomorphic dig¬ ital filter to reduce the effects of radiofrequency inhomogeneity (which results in spatial variations in baseline pixel intensity).43 Trained research assistants blind to subject identity, age, and diagnosis identified the image intensity value above which all pixels could be considered tissue and below which all pixels could be considered CSF. Interrater reliability for the different measures was assessed by having one of us (R.B.Z.) and each of four research assistants analyze a set of 15 MRI scans involving a total of 120 sections. Measures were then entered into a data set and randomly sorted across the five raters to ensure that the correlations measured agreement rather than simple association.44 A correlation matrix was calculated for each measure using the Spearman Rank Cor¬ relation Coefficient, and the mean value from the correlation matrix was calculated. Interrater reliability for the CSF-tissue "thresholding" was .98. Raters also delineated the third ven¬ tricle on the section in which it appeared largest (Table 2) with =
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Regions-of-lnterest Intracranial volume index (IVI) Lateral ventricles
Third ventricle
Subcortex
Total cortex
Definition Total volume of all fluid and tissue voxels in sections 2 through 8 Sum of all fluid voxels in inner 55% of all sections on which the lateral ventricles could be identified beginning with the index section (section 2 in Fig 1), expressed as a percentage of IVI Identified on the one section on which it appeared largest; an irregularly shaped region of interest was traced around the third ventricle and some of the adjacent thalamus but was limited anteriorly by the anterior horns of the lateral ventricles and posteriorly by the pineal gland; the third ventricle volume was defined as the sum of ail fluid voxels in this region of interest and expressed as a percentage of IVI Fluid and tissue voxels in inner 55% of the index section; gray matter and white matter volumes are expressed as percentages of total voxels in this volume Fluid and tissue voxels in outer 45% of the most superior 7 sections (sections 2 through 8 in Fig 1); cerebrospinal fluid; gray matter and white matter volumes are expressed as percentages of total voxels in this volume
reliability of p= .95. Midlines were drawn to sepa¬ right hemispheres of the brain for all brain sec¬ tions using a series of 17 points for all sections inferior to the centrum semiovale and nine points for all sections located more superiorly. Interrater reliability of the left-right hemisphere ra¬ an
interrater
rate the left and
tios for each section, based on these midlines, was .99. For each scan, the most anterior and posterior points of the corpus callosum were also identified. To enhance gray-white contrast, another set of composite im¬ ages was created by adding together pixel intensity values for the early and late echo images. These composite images were also filtered to reduce radiofrequency inhomogeneity. Gray-white segmentation was achieved using an automated procedure32 based on a nonparametric histogram analysis technique.45 All pixels identified as tissue during manual thresholding were subjected to histogram analysis, which determined the thresh¬ old value separating gray from white matter (Fig 1). This auto¬ matic gray-white thresholding technique was validated against our previously reported operator-driven thresholding method.43 Interrater reliability for the manual gray-white separation tech¬ nique was =.73, while comparing the automated technique with this manual technique yielded a value of p= .79. Regional Divisions of Segmented Images. —The images were divided according to anatomical landmarks and a priori geomet¬ ric rules in an effort to achieve standardized regional divisions of the brain images. To this end, each segmented brain section was first divided into an inner 55% region (to facilitate quanti¬ fication of central CSF including the ventricles) and an outer 45% (to facilitate quantification of the cortex). These proportions were empirically determined to maximize differentiation of ventricu¬ lar from cortical sulcal fluid.31 Using these divisions, estimates of the volume of CSF or white matter and gray matter were calcu-
lated for the intracranial volume, the lateral ventricles, the cere¬ bral cortex, and the subcortical region (Fig 1, Table 2). The total volume of all seven sections summed is referred to as the intra¬ cranial volume index (IVI). Each section was further divided into four regions by three coronal planes passing through (1) the most anterior extreme of the genu of the corpus callosum, (2) the most posterior extreme of the splenium of the corpus callosum, and (3) midway between them. The first of these planes was chosen to establish a bound¬ ary for the prefrontal region. The latter planes, though arbitrary, provided a more reliable basis for dividing cortical sections than specific cortical sulcal landmarks, which were difficult to estab¬ lish reliably on these images. These planes were projected through each section, perpendicular to the orientation of the ax¬ ial sections and the midline (Fig 1). The "total cortical" measure, which comprised the outer 45% of the seven MRI sections, was then subdivided empirically into the following anatomical mea¬ sures, by combining the areas labeled on Fig 1 across sections: prefrontal (a), frontotemporal (b), temporoparietal (c), frontal (d), parietal (e), and parieto-occipital (f). Within each of these regions, CSF, gray matter, and white matter voxels were summed bilaterally and then divided by the total volume of the region so that the measures could be expressed as the percent¬ age CSF, gray matter, and white matter in each region. This procedure was undertaken in order to minimize the effect of in¬ dividual differences in head size.
Statistical An age
age
on
Analysis
regression model was used to control for the effect of
all CSF, gray matter, and white matter measures.29 This
approach enables direct comparison of groups of different ages with the effect of age removed. Although the groups were of similar ages in this study, an age regression model reduces within-group variance and so enhances the ability to detect between-group differences. For each of the percentage CSF, gray matter, and white matter measures in all regions-of-interest, the values for controls were regressed against age. Values for the schizophrenics were then compared with the control regression and expressed as scores (the difference between the schizo¬ phrenic value and the control value predicted from the control regression, divided by the control SE of the regression). Patient scores were evaluated for statistical significance using a f sta¬ tistic of the form: [(mean y'-mean y)-i>(mean x'-mean x)] t=-
=
V
(mean a:'-mean
x)2}
-s^-)
where
mean y and mean y' are the mean percentage CSF, per¬ centage gTay matter, or percentage white matter measure for patients and controls, respectively; b, the slope of age vs percentage CSF, percentage gray matter, or percentage white matter for controls; Sp, the pooled SE of the estimate for the pa¬ tient and control regressions; Sxx, the sum of squared differences between the control ages and the mean age for controls; and n' and are the sample sizes for patients and controls, respectively.
Tests of differences in CSF measures were tested one-tailed while differences in gray matter and white matter measures were tested two-tailed, with =.05. The age regression model was compared with a more conventional statistical approach, analy¬ sis of covariance (ANCOVA), to control for age effects on these data. Correlations of global brain measures with demographic vari¬ ables were calculated with Pearson Product-Moment tests. Cor¬ relations of specific brain measures with clinical measures of the BPRS were assessed using the Spearman Rank Correlation test. To test whether the schizophrenics differed horn controls in in¬ tracranial volume, differences in IVI volume were assessed us¬ ing ANCOVA with height as a covariate. Because the multiplicity of statistical analyses used could re¬ sult in Type I errors, we have therefore explicitly described all
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Fig 1.—Final processed and segmented images for eight consecutive axial brain sections used for quantitative analysis (see "Patients and Methods" section). These brain images have been segmented into cerebrospinal fluid, gray matter, and white matter pixels. Definitions of regions-of-interest appear in Table 2. Section I or 2 was used to measure the third ventricle. Combinations from sections 2 to 8 were used for volumetric quantification of other regions-of-interest. Each segmented brain section was first divided into an inner 55% region, which included the ventricles and subcortical regions-of-interest, and an outer 45%, which included the cortical regions-of-interest. The cortical region was subdivided as follows: prefrontal (a), frontotemporal (b), temporoparietal (c), frontal (d), parietal (e), and parieto-occipital (f).
specifying multiple levels of statistical signifi¬ and have been conservative in interpreting the differences, taking this multiplicity into account. tests carried out
cance
mean z=
RESULTS Brain Volume Estimates of Normal
Control
ume was
Subjects
The percentage CSF, gray matter, and white matter for the IVI for controls is shown in Fig 2. The percentage gray matter was significantly negatively correlated with age (r= —.51, Ps.05), while the correlation of percentage CSF with age failed to reach statistical significance (r= .31, P^.10) and the percentage white matter was unrelated to age (r— .05). Third-ventricular volume was significantly correlated with age (r= .39, P=s.05), while lat¬ eral ventricular volume was not (r= .14). The percentage CSF, gray matter, and white matter for each of the brain regions stud¬ ied are presented in Table 3. —
Brain Volume Estimates of Schizophrenic Patients Within the schizophrenic group, the same pattern of agerelated changes in CSF, gray matter, and white matter were found; significant correlations were found between age and both CSF (r .42, P« .05) and IVI gray matter (r .48, P« .05) but not between age and IVI white matter (r .22). Neither lateral ventricular volume nor third-ventricular volume was signifi¬ cantly correlated with age in the schizophrenic group (r .07 and r— .26, respectively). The significant effect of age in the normal control group necessitated correcting the patients' MRI mea¬ sures for age. Accordingly, the raw percentage measures of CSF, gray matter, and white matter of the schizophrenic patients were converted to age-corrected scores derived from the control data (Fig 3). Mean percentage values and mean scores for schizo¬ phrenics for all CSF, gray matter, and white matter measures are listed in Tables 3 and 4, respectively. Mean scores for the con¬ trols are 0 for each measure. Values for schizophrenic patients that differed significantly from 0 are indicated in Table 4. For the brain as a whole (ie, IVI), the schizophrenic patients showed less gray matter (mean z= —1.53, P=£.001) but no significant differ=
=
in white matter (mean 0.55) or CSF (mean 0.48, P=s.25). Gray matter differences were apparent for both the left and right hemispheres of the TVI (mean z= —1.22, P=s.01, and enees
-
=
—
=
=
=
-1.62, P=£.001, respectively). Lateral ventricular vol¬ significantly larger in the schizophrenic group than in
the controls, while the differences in third-ventricular volume and total cortical CSF volume did not reach statistical significance. Cortical subregions were then analyzed to determine whether the global differences in gray matter could be localized. Gray matter volume scores were lower than expected from normal values in all six cortical subregions, with differences attaining statistical significance in all but the parietal region (area e). Re¬ gional cortical gray matter scores were significantly intercorrelated in 13 of 15 comparisons within the schizophrenic group and eight of 15 within the control group (Table 5). Percentage white matter of the schizophrenic group was significantly increased in the parieto-occipital region but did not differ from normal in any of the other brain areas. Cerebrospinal fluid volume was signif¬ icantly greater in the frontotemporal and temporoparietal re¬ gions (areas b and c) of the schizophrenic group. The ANCOVA, with age as a covariate, yielded results similar to those obtained from the age-regression analysis (Table 3). Diagnosis-by-age interactions were significant only for total cor¬ tical white matter and parieto-occipital CSF, thus precluding co¬ variate analysis. The schizophrenic group had significantly less gray matter over all seven sections (F 19.38, P=£.001), in the cortical region as a whole (F 10.62, P«s.01), in the subcortical region (F 6.25, P«£.05), and in each regional cortical measure except for the parietal region (Table 3). Ventricular volume was also significantly greater in the schizophrenic group than in the control group (F 5.30, P=£.05). White matter measures did not differ significantly between groups except in the parieto-occipital region, which was significantly greater in the schizophrenic as compared with the control group (F 5.28, P=s.05). As in the age regression analysis, cortical sulcal CSF volumes were signifi¬ cantly greater in the schizophrenic group only in the frontotem¬ poral and temporoparietal regions. When estimating intracranial volume (ie, IVI, the sum of all =
=
=
=
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
=
Intracranial volume index
Total cortical Subcortical
45 40
30 20
— -
25
30
40
35
45
44.6±3.5|| 47.3 ±3.4
44.7 ±5.0 43.4 ±3.9
47.2 + 5.7 50.7±4.0
29.2 + 5.4 27.7±3.5
18.4±10.7|| 50.6±6.6
31.1 ±9.1 31.8±9.0
...
...
33.2±2.9
17.8±9.0 16.3 ±6.5
46.5±5.0 49.5 ±5.3
35.7±5.2 34.2 ±3.5
Parietal
S C
14.7±6.4 14.7±5.8
46.8±4.4 47.6±5.9
38.4±3.5 37.7±3.5
Parieto-occipital
S C
17.3±8.2§
48.0±4.6 50.1 ±3.7
34.7±4.7||
S C
4.3±1.8||
S C
0.074±0.046 0.063±0.027
Third ventricle
50
35.0±3.7§
51.8±4.9* 37.0±5.5 55.2±3.9 37.1 ±4.3
Lateral ventricles
35
48.2±4.3 50.9±3.7
11.2±6.8||
:
sí-
16.8±6.9 15.9±4.7
S C S C
Frontal
55
42.4±2.6t 44.4±3.9 44.9 ±2.1 43.0±2.4
% White
55.2±5.8
Temporoparietal
(j 50
13.3±5.0 12.1 ±3.2
Gray
13.0±5.7
65-
>~
S C S C S C
%
23.6 ±9.7 21.6±6.5
Frontotemporal
60
% CSF
S C S C
Prefrontal
70
Group
7.8±3.6
19.6±8.0
30.3±6.4
3.2±1.0
*CSF indicates cerebrospinal fluid; S, schizophrenic; and C, control. Values are mean ±SD. tPss.001 vs control group. ^.01 vs control group. §Analysis of covariance was not calculated owing to differences in slopes between the two groups. || «.05 vs control group.
70 65 60 55 50-
not significantly correlated with duration of illness or age at onset of (r= .12 and r= .01, respectively). correlated Gray matter scores for the IVI were were
.-·.· I
45
schizophrenia
40 35
-
-
30 20
——
25
30
——
35
40
45
=
50
Age, y
Fig 2.—Age vs percentage of intracranial volume index made up of cerebrospinal fluid (CSF), gray matter, and white matter for 20 right-handed male community volunteers (control group). Each graph displays the linear regression line (solid line) as well as lines 1 and 2 SEs of the regression above and below the mean (dashed lines).
CSF, gray matter, and white matter pixels in the seven sections), the schizophrenic patients did not differ significantly from trols (542 vs 523 cm3, respectively; F 1.964).
con¬
=
Correlations of MRI Scores With Demographic and Behavioral Data
=
-
—
=
—
—
-
Thinking
=
The schizophrenic patients did not differ significantly from controls on estimates of total ufetime alcohol consumption (Ta¬ ble 1). In the schizophrenic group, IVI gray matter and white scores were not significantly correlated with the matter estimates of lifetime alcohol consumption (r= .07 and r= .06, respectively). The IVI gray matter scores for the schizophren¬ ics were not significantly correlated with years of education, NART IQ, or the WAIS-R Vocabulary age-scaled score (r= .16, r .05, r .02, respectively). Gray matter scores for the IVI =
significantly
-
with the BPRS Withdrawal-Retardation factor (p= .47, P«.05) but not with the total BPRS score (p= .21) or the BPRS Think¬ ing Disturbance factor (p .05). White matter scores for the IVI were not significantly correlated with any of the three BPRS measures: total score (p=.01), Withdrawal-Retardation factor (p= .16), or Thinking Disturbance factor (p= .26). The predicted association between scores on the Withdrawal-Retardation fac¬ tor and the prefrontal (area a) gray matter scores failed to reach the a priori significance level (p= .35, P= .053). The scores on the Disturbance factor were not significantly correlated with either the frontotemporal (area b) or temporoparietal (area c) gray matter scores (p=-.02 and -.22, respectively). Hypothesized brain-behavior relationships were further pur¬ sued by creating composite scores for the two frontal measures (areas a and d) and the two temporal measures (areas b and c). The association between the scores on the WithdrawalRetardation factor and the composite frontal measure was in the predicted direction (p .36, .051), while no association was apparent with the composite temporal measure (p= .11). The Disturbance factor was not significantly correlated with either the composite frontal or the temporal measure (p .17and p= —.14, respectively).
Thinking
-
=
COMMENT In this MRI study, quantitative estimates of gray mat¬ ter volume but not white matter volume were lower in all regions examined in patients with schizophrenia as com¬ pared with healthy control subjects. Thus, the diminished
Downloaded From: http://archpsyc.jamanetwork.com/ by a Penn State Milton S Hershey Med Ctr User on 05/24/2015
Regions-of-lnterest
CSF
Intracranial volume index
0.48 ±1.52
-1.53 ±1.24+ 0.55 ±1.56
Total cortical
0.31 ±1.41
-1.15±1.26
Gray
White
0.68±1.36
Subcortical Prefrontal
0.42 ±1.48
-1.12±1.47
0.41 ±1.51
Frontotemporal Temporoparietal
0.89±1.85§ -0.96±1.19 0.99±1.85§ -1.30±1.51
0.01 ±1.06
Frontal
0.41 ±1.39
-1.00+1.11* 0.46 ±1.47
0.09 ±1.04
-0.38 ±0.74
Parietal
Parieto-occipital
...
-0.21 ±0.90
-0.78±1.00§ 0.31 ±1.27
0.12 ±1.43 0.43 ±1.33
-0.74±1.12§ 0.65±0.68§
Lateral ventricles 1.02±1.78§ 0.55 ±1.74 Third ventricle *CSF indicates cerebrospinal fluid. Values are mean ± SD. Cerebrospinal fluid probability values are based on one-tailed tests; gray and white matter values are based on two-tailed tests. tPss.001 vs control group. «.01 vs control group. §P=s.05 vs control group.
Suddath et al23 and the postmortem study by Pakkenberg.30 In the Suddath et al study, the schizophrenic pa¬ tients had 67% larger ventricles and 19% smaller tempo¬ ral lobe gray matter volume than did controls. In the present study, the schizophrenic group's ventricular vol¬ ume was 34% larger than that of the control group. The schizophrenic group's gray matter volume was consis¬ tently smaller across the cortical subregions: 9% less for frontotemporal, 7% less for temporoparietal, 8% less for prefrontal, 8% less for frontal, 4% less for parietal, and 6% less for parieto-occipital. The percentage difference in gray matter volume for the cortical region as a whole was 7% less, which compares with the 12% mean reduction in cortical gray matter found in the postmortem study by
Pakkenberg.30
Fig 3.—Age vs percentage of intracranial volume index made up of cerebrospinal fluid (CSF), gray matter, and white matter for 22 right-handed male schizophrenic patients. The points for the schizophrenic patients are superimposed on the linear regression lines and the SEs of the regression lines determined for the control group (see Fig 2). The schizophrenic group differed significantly from the control group
not on the
3 and 4).
on
the percentage gray matter (P
