- Email: [email protected]
Structural Gray Matter Differences between First-Episode Schizophrenics and Normal Controls Using Voxel-Based Morphometry
NeuroImage 17, 880 – 889 (2002) doi:10.1006/nimg.2002.1180
Structural Gray Matter Differences between First-Episode Schizophrenics and Normal Controls Using Voxel-Based Morphometry 1 Dominic E. Job,* Heather C. Whalley,* Sarah McConnell,* Mike Glabus,† Eve C. Johnstone,* and Stephen M. Lawrie* *Department of Psychiatry, University of Edinburgh, Kennedy Tower, Royal Edinburgh Hospital, Morningside Park, Edinburgh EH10 5HF, Scotland, United Kingdom; and †Unit on Integrative Neuroimaging, Clinical Brain Disorders Branch, National Institute of Mental Health, National Institutes of Health, Bethesda, Maryland 20892 Received October 30, 2001
The aim of this study was to compare the gray matter segments from T1 structural MR images of the brain in first-episode schizophrenic subjects (n ⴝ 34) and normal control subjects (n ⴝ 36) using automated voxel-based morphometry (VBM). This study is novel in that few studies have examined subjects in their first episode of schizophrenia. The subjects were recruited for the Edinburgh High Risk project and regional brain volumes were previously measured using a semi-automated volumetric region of interest (ROI) method of analysis. The primary interest was to compare the results from the compatible parts of the ROI study and the primary VBM approach. Our secondary interest was to compare the results of a study-specific template that was constructed from the control group to those using the generic T1 template (152 Montreal Neurological Institute brains) supplied with SPM99 (statistical parametric mapping). The images were processed and statistically analyzed using the SPM99 program. VBM analysis identified significant decreases in gray matter in the schizophrenics relative to the normal control group at the corrected voxel level (P < 0.05) in the right anterior cingulate, right medial frontal lobe, left middle temporal gyrus, left postcentral gyrus, and the left limbic lobe. There were no increases in gray matter in the schizophrenics relative to the control group. The construction of a customized template appeared to improve the detection of structural abnormalities. The analyses were subsequently restricted to voxels within the amygdala– hippocampal complex using the SPM small-volume correction. This identified gray matter decreases in the schizophrenics, at the corrected voxel level (P < 0.05), in the left and right uncus and parahippocampal gyri and the right amygdala. These results are compatible with and extend the relevant findings of the previous volumetric ROI analysis, when allowing for the differ-
1
This study was funded by a program grant from the Medical Research Council of Great Britain. 1053-8119/02 $35.00 © 2002 Elsevier Science (USA) All rights reserved.
ences between the methods and interpretation of their results. © 2002 Elsevier Science (USA)
INTRODUCTION Many studies examining structural magnetic resonance (MR) images of the brain in people with schizophrenia have indicated that they have subtle anatomical differences compared to normal control groups. Quantitative reviews of these studies have suggested that, in general, the temporal lobes, medial temporal lobe structures, and superior temporal gyri are smaller and the ventricles are larger in schizophrenics (Lawrie and Abukmeil, 1998; Wright et al., 2000). Most of these studies have employed region of interest (ROI) measurement of brain structures by manually drawing around the regions of interest. This type of analysis is prone to measurement error and is very time consuming. This restricts the number of regions that can be practically measured and means that relatively large structures of interest must be defined at the outset of the study. Voxel-based morphometry (VBM) is an automated method of examining structural MR images of the brain (Ashburner and Friston, 2000; Wright et al., 1995) based upon the statistical parametric mapping (SPM) techniques originally designed for analysis of functional images (Friston et al., 1995a). This method requires no manual tracing around regions and has the advantage of examining the brain as a whole. However, VBM also has limitations, resulting in difficulty in interpretation of VBM results. Other studies have used the same or variations of these methods to examine structural MR images from a variety of subject groups, including schizophrenic patients (Chua et al., 1997; Gaser et al., 1999; Wilkie et al., 2001; Wright et al., 1995, 1999a), people with depression (Shah et al., 1998), children and adolescents (Sowell et al., 1999), subjects with headache (May et al., 1999) and epilepsy (Woermann et al., 1999),
880
881
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
and taxi drivers as they acquire spatial knowledge (Maguire et al., 2000). The aim of this study was to compare the gray matter segments in first-episode schizophrenics and normal controls using this automated technique. The hypotheses tested were that the schizophrenic group would differ from the control group in the regions predicted by the literature and as indicated by the relevant findings in our previous ROI study, which showed volume reduction in the schizophrenic group, relative to the control group, in the left (P ⫽ 0.0003, uncorrected) and right (P ⫽ 0.01, uncorrected) amygdala– hippocampal complex (AHC) (Lawrie et al., 2001). METHODS Participants The subjects were recruited for the Edinburgh High Risk Study (Hodges et al., 1999; Johnstone et al., 2000). The subject groups consisted of 34 subjects in their first episode of schizophrenia (American Psychiatric Association, 1994) and 36 normal control subjects matched groupwise for age. The first-episode schizophrenia patients were identified from admissions to the Royal Edinburgh Hospital and associated psychiatric hospitals in Lothian region, Scotland. These subjects have no known family history of schizophrenia in their firstor second-degree relatives. Their case notes were reviewed with a structured assessment and the patients were examined with structured psychiatric interviews to confirm the diagnosis. One hundred fifty subjects at high risk of developing schizophrenia were also recruited but are not included in this paper. Three of the control subjects had isolated psychotic symptoms (mainly isolated hallucinations) in psychiatric interviews but were not excluded from the study to avoid biasing the control group (Miller et al., 2002). Control subjects had been recruited on the basis that there were no major psychiatric disorders or family history of the same. The inclusion of three controls with symptoms but no disorder is a strength of this study, as a similar proportion of the general population will have such symptoms (Johns et al., 2002). Details of the subject groups are presented in Table 1. Further details of the subject groups can be found in Johnstone et al. (2000). Ethical approval was obtained from the relevant health authorities and all subjects gave written informed consent. Scanning Protocol MR brain scanning was performed on a 42 SPE Siemens (Erlangen, Germany) Magnetom operating at 1.0 T. Midline sagittal localization was followed by two sequences to image the whole brain. The first scan was a double spin echo sequence which gave simultaneous
TABLE 1 Details of the Subject Groups
Age, years (sd) Male:female Handedness, mixed:left:right Height (sd) Paternal social class at birth, manual:nonmanual:unknown
Control (n ⫽ 36)
Schizophrenic (n ⫽ 34)
21.17 (2.37) 17:19 2:3:31 170.21 (9.09) 19:15:2
21.35 (3.66) 23:11 2:1:31 174.16 (11.85) 14:11:9
proton density and T2-weighted images (TR ⫽ 3565 ms, TE ⫽ 20 and 90 ms, 31 contiguous 5-mm slices in the Talairach plane, field of view 250 ⫻ 250 mm) to identify any gross brain lesions. The second scan was a three-dimensional magnetization prepared rapid acquisition gradient echo sequence consisting of a 180° inversion pulse followed by a fast low angle shot collection (flip angle 12°, TR ⫽ 10 ms, TE ⫽ 4 ms, TI ⫽ 200 ms, and relaxation delay time 500 ms, field of view 250 ⫻ 250 mm) to give 128 contiguous “slices” of 1.88-mm thickness. Immediately after each subject’s scan a large plain test object filled with light oil was scanned in exactly the same place in the coil and in the same orientation as the subject’s head. The test object data were used to correct for inhomogeneity in the radiofrequency coil. The mean intensity of a 5 ⫻ 5 pixel square in the center slice of the test object data was taken as the optimum intensity. The whole data set was normalized to this value. By dividing the patient data by the normalized data any deviation from the optimum coil response could be corrected for global differences in signal inhomogeneities. To address intensity shifts across each subject’s image the phantom image for each subject was superimposed upon each subject’s brain image. This standardizes (calibrates) the signal intensity of the scanner and counteracts falloff in intensity that is due to the scanner coil inhomogeneity. This produces images whose variations in signal intensity are due to the composition of the brain and not to the magnetic coil of the scanner. Details of the volumetric analysis on the same subject groups have been described previously (Lawrie et al., 2001; Whalley et al., 1999). In brief, three blinded raters with high reliability used automated edge detection and manual editing to measure the volumes of the whole brain, prefrontal and temporal lobes, the AHC, thalamus, lentiform, caudate, and the ventricular system for controls, high-risk, and first-episode schizophrenics.
882
JOB ET AL.
Image Processing Data were transferred to a Sun SPARC 1 workstation and then analyzed using Analyze version 7.5.5 (Mayo Foundation, Rochester, MN) and Statistical Parametric Mapping version 99 (SPM99) (Wellcome Department of Cognitive Neurology, Institute of Neurology, London) (Friston et al., 1995a,b) running in MATLAB version 5.3 (The MathWorks, Natick, MA). Images were analysed using voxel-based morphometry methods as described by Ashburner and Friston (Ashburner and Friston, 2000). All MR images were initially loaded into Analyze, corrected for field inhomogeneity as described above, and converted from 16-bit to 8-bit images. The default settings in SPM were then altered to read in the subject’s scans which were in the coronal plane and in radiological convention. Two templates were used in the study. The first template used was the generic T1 template included in the SPM package, composed of 152 subjects, mean age 25 years, 43% female, and 90% right-handed. The second template, the Edinburgh High Risk Study (EHRS) T1 template, was made from 33 healthy control images using SPM99, mean age 21 years, 53% female, and 86% right-handed. The three controls that had psychotic symptoms, principally fragmentary hallucinations, in the context of drug abuse, were not used in the construction of the EHRS T1 template. To construct the EHRS T1 template the healthy control images were spatially normalized to the generic T1 image using linear normalization and nearest neighbor interpolation. Extracerebral voxels were removed using the “Xtract brain” render function. This renders the gray and white matter segments and produces an “extracted brain.” Multiplying the original gray matter segment by the extracted brain file removes the extracerebral voxels. A mean image was then calculated and smoothed at 8 mm full width at half maximum (FWHM). All images were spatially normalized to the generic T1 and to the EHRS T1 template. This process involved minimizing the residual sum of squares differences between the images and each template and was performed using a 12-point linear affine transformation. Nearest neighbor interpolation was used to preserve the original intensity values and gray/white matter contrasts of the images. The images were then segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using the SPM modified clustering algorithm and SPM’s GM, WM and CSF a priori templates. This was performed with no inhomogeneity correction as these images had previously been corrected (as described above). The extracerebral voxels were then removed from the gray and white matter segments using the Xtract
brain render function in SPM99. The spatially normalized and segmented images were then smoothed with a 12-mm FWHM isotropic Gaussian kernel to remove the differences in sulcal/gyral patterns among individuals that may not be accounted for by normalization (Ashburner and Friston, 2000). The 12-mm width was chosen for the smoothing kernel to reduce the number of false positives, to conform with the width nascent in the use of the SPM99 package, and to facilitate comparison with other studies (Chua et al., 1997; Wilkie et al., 2001; Wright et al., 1999a). The 12-mm Gaussian smoothing filter promotes the detection of differences in structures of around 12-mm spatial extent (Ashburner and Friston, 2001). As all brain images are normalized to a standard template prior to group analysis, large-scale differences among the groups are eliminated. The resulting gray matter segments for each of the subjects were then entered into statistical analysis. A randomization analysis is generally desirable (Ashburner and Friston, 2001) but was not necessary for the ROI–VBM confirmation. Randomization testing is most pertinent in studies with low degrees of freedom (Nichols and Holmes, 2001). In this work the primary VBM contrast has 62° of freedom, and randomization testing is therefore not expected to have a significant impact on the results. However, randomization analysis and scale space search (Worsley et al., 1996) may be part of future work. Statistical Analysis Group comparisons were performed using the a random effects analysis “single subject: conditions and covariates,” a statistical model in SPM99 based on the general linear model. Nuisance or confound variables were entered into the analyses to remove global differences between the images (ANCOVA). In the volumetric ROI study, the whole brain volume (WBV) was measured (in native space) to control for global effects, so this measure was also used in the primary VBM study to replicate the conditions from the ROI study. Gender, handedness, height, age, and paternal social class at birth were also entered as nuisance variables into the model, to replicate the analysis performed in the ROI study (Lawrie et al., 2001). Handedness was determined by the subjects preferred writing hand. In the conventional VBM approach, the automated number of voxels (NVOX) measure is commonly used as a nuisance variable to remove global differences between the images, while in ROI studies the WBV measure is typically used. To illustrate any differences between the two approaches in implementing this nuisance variable, a further analysis to compare the effects of NVOX and WBV was performed. These two measures were compared in VBM using the images normalized to the EHRS T1 template. In the volumetric ROI study the WBV measure included white mat-
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
883
TABLE 2 The Primary VBM Study: EHRS T1 Template, Covariates Matching Those in the ROI Study Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
The primary VBM study: Results for the whole brain 0.001 0.031 0.017 0.019 0.052 0.042 0.054 0.084
3.96 3.96 ⫺55.44 ⫺63.36 ⫺45.54 ⫺11.88 ⫺19.80 55.44
40.87 45.83 ⫺4.46 ⫺12.64 ⫺0.42 ⫺0.84 ⫺5.05 ⫺22.98
1.64 ⫺15.75 ⫺11.55 19.06 ⫺8.39 ⫺16.78 ⫺23.30 6.68
Right, anterior cingulate, Brodmann area 32 Right, medial frontal gyrus, Brodmann area 11 Left, middle temporal gyrus Left, postcentral gyrus, Brodmann area 43 Left, superior temporal gyrus Left, limbic lobe (medial border of amygdala) Left, uncus, Brodmann area 28 Right, superior temporal gyrus, Brodmann area 41
The primary VBM study: Small-volume correction amygdala–hippocampus 0.001 0.002 0.028 0.038
⫺19.80 ⫺13.86 23.76 21.78
⫺5.05 ⫺0.84 ⫺5.14 ⫺6.90
⫺23.30 ⫺16.78 ⫺24.98 ⫺21.52
Left, uncus, Brodmann area 28 Left, parahippocampal gyrus Right, parahippocampal gyrus Right, uncus, amygdala
Note. EHRS, Edinburgh High Risk Study. Nuisance covariates: whole brain volume, handedness, sex, age, height, and paternal social class.
ter. The global gray matter, NVOX, was determined for use in the VBM study using a matlab script “get_globals.m” (J. Ashburner), a measure that does not include white matter. This calculated the number of voxels within each image and was performed on segmented raw images, derived using the internal 12point affine normalization only, in SPM99. Two contrasts were constructed, one to examine regional decreases and the other to examine regional increases in gray matter between the two groups. The results of these contrasts were then displayed as a statistical parametric map thresholded at a significance level of P ⬍ 0.05 (corrected). All results described are based on the voxel level of significance. The analysis was then restricted to voxels contained within the amygdala– hippocampus (4.9 resels, 12-mm FWHM smoothing). This was performed using the “small volume correction” (SVC) function in SPM99 and was also thresholded at a significance level of P ⬍ 0.05 (corrected). The coordinates of the significant voxels were converted from MNI (Montreal Neurological Institute) space to Talairach and Tournoux (Talairach and Tournoux, 1988) coordinates using the “mni2tal.m” matlab script, a nonlinear transform approach described by Brett (1999). RESULTS Table 2 and Fig. 1 present the results for the primary VBM approach, using the same confounds as used in the ROI study: whole brain volume, gender, handedness, height, age, and paternal social class at birth.
The results show reductions in gray matter in schizophrenics versus controls in the right anterior cingulate, right medial frontal lobe, left middle temporal gyrus, left postcentral gyrus, and left limbic lobe. The SVC results indicate reductions in gray matter in schizophrenics versus controls in the left and right uncus, left and right parahippocampal gyri, and the right amygdala. Table 3 presents the results for the EHRS T1 template, using the NVOX confound only. Here, the schizophrenic group showed a relative decrease in gray matter compared to the control group similar to the regions described above, i.e., the right anterior cingulate, right medial frontal gyrus, left middle temporal gyrus, and left postcentral gyrus and at the SVC level the left uncus, left and right parahippocampal gyri, and left amygdala. With the generic T1 template (see Table 4), using the NVOX confound only, the standard SPM approach, the schizophrenic group showed a relative decrease in gray matter compared to the control group in the right anterior cingulate, right medial frontal gyrus, left postcentral gyrus, and left uncus at the SVC level. The comparison to illustrate the differences between the two measures of nuisance variables (NVOX and WBV) showed relatively little difference in the two measures. The NVOX measure, with the EHRS T1 template (Table 3), showed changes as given above. The WBV measure, with the EHRS T1 template (Table 5), showed changes in the right anterior cingulate, left postcentral gyrus, left middle temporal gyrus, left uncus, and left and right parahippocampal gyri. The differences between the effects of the two measures,
884
JOB ET AL.
FIG. 1. Areas of gray matter reduction in schizophrenics versus healthy controls at a P-uncorrected level of 0.001. The color bar shows Z values corresponding to colors in the figure. The numbers beside each slice show the height on the z-axis in millimeters of the slice.
NVOX and WBV, can be seen in the differences between Tables 3 and 5. At the whole brain level Table 3 shows a difference in the right medial frontal gyrus
Brodman area 11 not seen in Table 5. At the whole brain level Table 5 shows a difference in the left and right anterior cingulate, but only in the right anterior
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
885
TABLE 3 EHRS T1 Template, with Number of Voxels (NVOX) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
EHRS T1, number of voxels: Results for the whole brain 0.002 0.019 0.033 0.003 0.027
3.96 40.97 3.48 3.96 39.86 20.11 3.96 43.896 ⫺15.65 ⫺63.36 ⫺14.58 19.15 ⫺63.36 ⫺44.38 5.91
Right, anterior cingulate, Brodmann area 32 Right, anterior cingulate Right, medial frontal gyrus, Brodmann area 11 Left, postcentral gyrus, Brodmann area 43 Left, middle temporal gyrus, Brodmann area 22
EHRS T1, number of voxels, small-volume correction amygdala–hippocampus ⫺13.86 ⫺21.78 ⫺19.80 23.76
0.007 0.029 0.032 0.043
⫺0.84 ⫺3.20 ⫺4.97 ⫺5.14
⫺16.78 ⫺25.07 ⫺21.62 ⫺24.98
Left, parahippocampal gyrus Left, uncus Left, uncus, amygdala Right, parahippocampal gyrus
Note. EHRS, Edinburgh High Risk Study.
cingulate as in Table 3. At the SVC AHC level Table 3 shows a difference in the left uncus and amygdala not seen in Table 5. There were no regions where there was an increase in gray matter in the schizophrenics relative to the control group in any of the different approaches. DISCUSSION Region of interest analysis is the gold standard for assessing structural MR data, but it is slow and not perfectly reliable. VBM is faster and less labor intensive and can be applied to the whole brain, but also has limitations. In this study we found significant differences between the controls and the schizophrenic groups in the left anterior cingulate, right medial frontal gyrus, left postcentral gyrus, and left medial temporal gyrus. The postcentral gyrus difference could not have been discovered by our volumetric ROI method as it lies outside any of our regions of interest (other than whole brain), while the other additional results—right medial frontal gyrus, anterior cingulate, and left me-
dial temporal gyrus—were only included within larger structures (prefrontal and temporal lobe volumes, respectively) and are therefore not comparable to the primary VBM study. An additional ROI study would be necessary to confirm the extended findings in the primary VBM study. In the ROI study we only found differences specifically between the first-episode patients and control subjects bilaterally in the amygdala– hippocampal complex. The VBM method can use a prior hypothesis for a region to strengthen the power of the statistical analysis of that region and gives a specific point of maximal change, rather than only confirming an expected change over a prespecified search volume. The ROI volume decrease in AHC gives little detail compared to the VBM findings with small-volume correction, but the detailed findings of the VBM study can be difficult to interpret. The point of maximal change must be carefully interpreted as a statistically significant difference of the probabilities of the voxels in each group being GM. Additionally, the 12-mm smoothing step can obscure the exact location of one or more points of change.
TABLE 4 SPM T1 Generic Template, with Number of Voxels (NVOX) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
Generic T1, number of voxels: Results for the whole brain. 0.004 0.007 0.013
3.96 40.97 3.48 3.96 47.85 ⫺14.17 ⫺65.34 ⫺16.43 21.09
Right, anterior cingulate, Brodmann area 32 Right, medial frontal gyrus, Brodmann area 11 Left, postcentral gyrus, Brodmann area 43
Generic T1, number of voxels, small-volume correction amygdala–hippocampus. 0.042 0.048
⫺21.78 ⫺17.82
⫺3.20 ⫺25.07 ⫺4.97 ⫺21.61
Left, uncus Left, uncus
886
JOB ET AL.
TABLE 5 EHRS T1 Template, with Whole Brain Volume (WBV) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
EHRS T1, whole brain volume: Results for the whole brain 0.001 0.010 0.003 0.015 0.084
3.96 40.97 3.96 39.76 ⫺63.36 ⫺12.64 ⫺63.36 ⫺44.38 ⫺1.98 7.33
3.48 18.28 19.06 5.91 ⫺8.78
Right, anterior cingulate, Brodmann area 32 Right, anterior cingulate Left, postcentral gyrus, Brodmann area 43 Left, middle temporal gyrus, Brodmann area 22 Left, anterior cingulate
EHRS T1, whole brain volume, small-volume correction for amygdala–hippocampus. 0.018 0.047 0.038
⫺13.86 ⫺.84 ⫺21.78 ⫺3.20 23.76 ⫺5.14
⫺16.78 ⫺25.07 ⫺23.98
Left, parahippocampal gyrus Left, uncus Right, parahippocampal gyrus
Note. EHRS, Edinburgh High Risk Study.
The differences in gray matter between the schizophrenic group and the normal control group are consistent with previous VBM reports of gray matter reductions in the superior temporal gyrus, amygdala, and insula (Wright et al., 1999a,b) and in a previous deformation-based approach which found gray matter reductions bilaterally in the frontal lobes and superior temporal gyri (Gaser et al., 1999). Our comparable ROI results and primary contrast VBM (Table 2) results are consistent for the AHC, taking into account that the findings are so similar that the inevitable methodological differences between the two methods, and each method’s shortcomings, outweigh the differences between the results from each method. The primary VBM approach shows a difference in the right amygdala and bilaterally in the parahippocampal gyrus, whereas the ROI study shows differences bilaterally in the AHC. It is of note that the VBM method shows differences mainly around the borders of the AHC or hippocampus, not only within the AHC and not in the hippocampus proper. Small movements in the point of maximal change can easily fall either side of a border line. For example, in Table 2 the left uncus maximal point is only 0.08 mm on the y axis and 1.68 mm on the z axis different to the left uncus/amygdala in Table 3. In our study, the small-volume correction function in SPM99 used images created by editing the relevant nonsmoothed EHRS T1 template to define the small volume, which contained parts of some bordering structures since the template is a mean image. Results from the secondary SVC contrasts are less consistent between different statistical and preprocessing manipulations. This shows that VBM is sensitive to the statistical confounders that are used (see Tables 3 and 5, for a comparison of two methodologically equivalent confounds), but more differences are found using study-specific templates in preprocessing the images (see Tables 3 and 4, for a template comparison). This
template sensitivity is probably attributable to a combination of using a template from matched controls and the same scanner, although we cannot tell whether one or other effect is more important. The primary VBM results show a relative rather than absolute replication of the comparable ROI results, but the results between different approaches (Tables 3, 4, and 5) are certainly compatible. An interesting inconsistency between the treatments is that in Table 2, the primary VBM contrast shows a significant reduction in the right uncus and amygdala, whereas Table 3 shows reductions in the left uncus and amygdala and Tables 4 and 5 show reductions in the left uncus. This hemisphere switch must be due to the confounders used in the primary VBM study. In the primary study, confounders include age, gender, handedness, height, and paternal social class at birth as confounds. It is well established that handedness and gender have some structural brain correlates, and therefore it is not unexpected that the statistical inference that includes these confounds shows a change in laterality of the amygdala finding, compared to those that do not. In our study, it may be that the ventricles in the schizophrenic subjects are reduced during spatial normalization and therefore the gray matter near them is reduced. It is therefore possible that structural differences in ventricular volumes can show up in a VBM study of gray matter volume. To avoid this problem spatial normalization can be based on only the segmented gray matter, where any significant differences are much more likely to be attributable to actual gray matter differences (Ashburner and Friston, 2001). Gray matter only normalization (after Good et al., 2001) is desirable and may be part of future work. Standard volumetric analyses are prone to measurement error. In the volumetric study all three raters obtained good interrater reliability figures (ICC range 0.75– 0.99) and assessed the scans blind to the clinical
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
status of the subject, but reliability was generally the lowest for small structures (Whalley et al., 1999). It should also be noted that while the ROI method measures differences over a specified region, the VBM method gives the points of maximal change for clusters of points of change (Wright et al., 1999a). Our VBM analysis shows differences mainly around the borders of the AHC. The ROI and VBM approaches can both suffer from partial volume effects, e.g., near the borders of the amygdala/parahippocampal gyrus, where some voxels are half gray matter and half white matter, it is difficult to threshold voxels into one class. While VBM is consistent in classification of partial volumes, it is also the case that white matter tracts such as the alveus and fimbria can be misclassified as gray matter during segmentation. CSF from the temporal horns of the lateral ventricles surrounding much of the hippocampal surface is often misclassified as gray matter, given that the signal intensities do not threshold as cleanly as they do in the cortex. The choroid plexus in the temporal horns could further obscure the tissue boundaries of the hippocampus. In medial temporal regions signal intensity values from different tissue types may be indistinguishable. However, this misclassification is the same for both subject groups, and smoothing should reduce any effects this has on the final statistics. These factors may account for subtle differences in the results between our two methods and among other ROI studies. This study has also shown compatible results for images spatially normalized to both the generic SPM99 T1 template and the specific EHRS T1 template constructed from images obtained in age-matched control subjects with the same scanner. However, constructing a customized template appears to improve the detection of structural abnormalities. The EHRS T1 template probably gives greater sensitivity due to this template being a study-specific template, matching the scanner and age of the subjects, an effect also described by Good et al. (2001). The comparison of the NVOX (Table 3) and WBV (Table 5) analyses shows that they give similar results. The NVOX measure could differ from the ROI approach as the NVOX measure is calculated over GM only, whereas the WBV measure includes WM. Segmentation for calculation of the NVOX measure in VBM can be poor as the process relies on overlaying a priori probability maps (in stereotactic space) onto internally normalized raw images to calculate the probability of any given voxel being GM/WM or CSF. This is susceptible to registration error as discussed in detail by Ashburner and Friston (Ashburner and Friston, 2000). Even when the raw images have been normalized using a 12-point linear affine transformation, it is possible that voxels containing a mixture of GM and WM can be classified as GM or WM, when strictly speaking they are a mixture, leading to further poten-
887
tial discrepancies between the ROI and VBM studies. Our confirmation of ROI results suggests, however, that such misclassification is not a serious problem, as do other VBM–ROI validations (Baron et al., 2001; Wright et al., 1999a). Its has been shown that VBM is capable of detecting differences in small areas such as the hippocampus (Maguire et al., 2000; Wright et al., 1999a), but it may be a less sensitive measure than ROI approaches for detecting volume differences. VBM is also sensitive to the confounds and templates used, and that makes the results difficult to interpret, especially for the AHC. We believe that the two approaches are complementary, with the ROI giving more detail regarding the difference in volumes of interest and the VBM having more detail regarding the specific points of maximal change. One cannot say from our results whether the differences between schizophrenics and controls are in the amygdala– hippocampal complex or surrounding structures or, as seems most likely, both, but they are certainly in the medial temporal lobe and include the amygdala (see meta-analyses and other VBM studies cited below). Indeed, the postmortem literature in schizophrenia has also less commonly found differences in the amygdala– hippocampal complex and more commonly in the parahippocampal gyrus (Harrison, 1999). A further possible limitation of VBM in this study is that a smoothing kernel of 12-mm isotropic FWHM Gaussian kernel could obscure small differences in hippocampus. However, we used linear rather than nonlinear normalization, leaving higher variability, such that a smaller smoothing filter would be inappropriate. Ashburner and Friston (2000) recommend a 12-mm filter and smoothing with a 12-mm FWHM Gaussian will best find differences in areas of 12-mm spatial extent (such as the amygdala– hippocampal complex rather than the hippocampus alone). The segmentation procedure may well include parts of the alveus and fimbria, potentially rendering some of their white matter as GM voxels, but it is the same for all images in all groups and therefore one would expect that the differences shown in the contrasts are real differences in structure. Smoothing is done after segmentation and this would reduce the impact of a small area, such as the alveus and fimbria, on the results. However, the potential misclassification of WM as GM does require cautious interpretation of the results. Bookstein (2001) points out that misregistration of small regions of interest in VBM can lead to the detection of systematic shape differences between groups. In this work we have not used nonlinear normalization, so there is no local registration and smoothing is intended to account for variability at a local level, i.e., in small regions of interest. In their reply to Bookstein (2001), Ashburner and Friston (2001) state that careful consideration of what may be the cause of any detected
888
JOB ET AL.
difference is required. Both Bookstein (2001) and Ashburner and Friston (2001) point out that many systematic differences can arise. For example, in our study, the schizophrenic subjects may have moved more or been systematically positioned differently in the scanner, producing a motion artifact in the images, resulting in biased segmentation. However, we have previously shown that the patients’ brains were not significantly more yawed than the controls, and there were no gross movement artifacts (Whalley et al., 1999). We specifically did not find differences in the left or right hippocampi in our VBM study. This could be due to its small size and spatial variability between subjects, which may have been implicitly accounted for by visual inspection in the ROI study. Such spatial variability has been described by Good et al. (2001). The ROI analysis was performed on scans without segmentation, such that the volumetric measurement of the hippocampus included regions of surrounding white matter (alveus and fimbria) which would not be analyzed in the VBM method (assuming accurate segmentation and adequate smoothing) which was performed solely on the thresholded gray matter segments. This, and the other limitations of VBM already discussed, may explain why some previous VBM studies in schizophrenia have found hippocampal reductions (Sigmundsson et al., 2001; Hulshoff et al., 2001) and others have not (Wright et al., 1999; Paillere-Martinot et al., 2001). All of these previous studies have, however, found reductions in the amygdala and/or parahippocampal gyrus. These VBM studies show precedence of the abnormalities we found in anterior cingulate (Sigmundsson et al., 2001), medial frontal gyri (Sigmundsson et al., 2001; Paillere-Martinot et al., 2001), left limbic lobe/amygdala (Hulshoff et al., 2001), right uncus/ amygdala (Wright et al., 1999; Hulshoff et al., 2001), and left parahippocampal gyrus (Sigmundsson et al., 2001; Paillere-Martinot et al., 2001). The findings in these studies closely match findings in our primary VBM contrast (Table 2 results). Abnormalities of the anterior cingulate/medial frontal lobe in schizophrenia are commonly reported in functional imaging studies of schizophrenia (e.g., Ebmeier et al., 1995), may represent part of the frontotemporal dysconnectivity that is currently thought to be the basis of the disorder (e.g., Fletcher et al., 1999), and may specifically be related to the negative symptoms of a loss of affect and drive (e.g., Chua et al., 1997). Volume reductions of the medial temporal lobe are the most consistently replicated ROI and VBM findings in schizophrenia (Lawrie and Abukmeil, 1998; Wright et al., 2000) and may underlie frontolimbic dysconnectivity and the positive symptoms of hallucinations, delusions, and thought disorder in particular (Wright et al., 1995; Chua et al., 1997).
One strength of our study of first-episode schizophrenics is that treatment effects are less likely to confound the results than might be the case in patients who have been on antipsychotic drugs for years. While this suggests a neurodevelopmental origin of the findings, we do not think that we can separate trait and state effects in the current study. In our study of highrisk subjects this may be possible as all subjects are antipsychotic free and repeated scans have been conducted over a period of symptomatic change. Future ROI and VBM studies in patients with schizophrenia and related populations should be able to specify the structural brain abnormalities in greater detail and link them to particular etiological factors and symptoms. REFERENCES American Psychiatric Association. 1994. Diagnostic and Statistical Manual of Mental Disorders, 4th Edition. American Psychiatric Association, Washington, DC. Ashburner J., Functional Imaging Laboratory. Wellcome Department of Cognitive Neurology, London, UK. Ashburner, J., and Friston, K. J. 2000. Voxel-based morphometry— The methods. NeuroImage 11: 805– 821. Ashburner, J., and Friston, K. J. 2001. Comments and controversies—Why voxel-based morphometry should be used. NeuroImage 14: 1238 –1243. Baron, J. C., Chetelat, G., Desgranges, B., Perchey, G., Landeau, B., de la Sayette, V., and Eustache, F. 2001. In vivo mapping of gray matter loss with voxel-based morphometry in mild Alzheimer’s disease. NeuroImage 14: 298 –309. Bookstein, F. L. 2001. “Voxel-based morphometry” should not be used with imperfectly registered images. NeuroImage 14: 1454 – 1462. Brett, M. 1999. The MNI brain and the Talairach atlas. MRC Cognition and Brain Sciences Unit. http://www.mrc-cbu.cam.ac.uk/ Imaging/. Chua, S. E., Wright, I. C., Poline, J. B., Liddle, P. F., Murray, R. M., Frackowiak, R. S. J., Friston, K. J., and McGuire, P. K. 1997. Grey matter correlates of syndromes in schizophrenia—A semi-automated analysis of structural magnetic resonance images. Br. J. Psychiatry 170: 406 – 410. Ebmeier, K. P., Lawrie, S. M., Blackwood, D. H. R., Johnstone, E. C., and Goodwin, G. M. 1995. Hypofrontality revisited: A high-resolution single photon emission tomography study in schizophrenia. J. Neurol. Neurosur. Psychiatry 58: 452– 456. Fletcher, P. C., McKenna, P. J., Friston, K. J., Frith, C. D. and Dolan, R. J. 1999. Abnormal cingulate modulation of fronto-temporal connectivity in schizophrenia. NeuroImage 9(3): 337–342. Friston, K. J., Ashburner, J., Poline, J. B., Frith, C. D., Heather, J. D., and Frackowiak, R. S. J. 1995a. Spatial registration and normalisation of images. Hum. Brain Mapping 2: 165–189. Friston, K. J., Holmes, A. P., Worsley, K. J., Poline, J. B., Frith, C. D., and Frackowiak, R. S. J. 1995b. Statistical parametric maps in functional imaging. Hum. Brain Mapping 2: 189 –210. Gaser, C., Volz, H. P., Kiebel, S., Riehemann, S., and Sauer, H. 1999. Detecting structural changes in whole brain based on nonlinear deformations—Application to schizophrenia research. NeuroImage 10: 107–113. Good, C. D., Johnsrude, I. S., Ashburner, J., Henson, R. N., Friston, K. J., and Frackowiak, R. S. 2001. A voxel-based morphometric
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA study of ageing in 465 normal adult human brains. NeuroImage 14: 21–36. Harrison, P. J. 1999. The neuropathology of schizophrenia. A critical review of the data and their interpretation. Brain 122: 593– 624. Hodges, A., Byrne, M., Grant, E., and Johnstone, E. 1999. People at risk of schizophrenia. Sample characteristics of the first 100 cases in the Edinburgh high-risk study. Br. J. Psychiatry 174: 547–553. Hulshoff Pol, H. E., Schnack, H. G., Mandl, R. C. W., van Haren, N. E. M., Konig, H., Collins, D. L., Evans, A. C., and Kahn, R. S. 2001. Focal gray matter density changes in schizophrenia. Arch. Gen. Psychiatry 58(12): 1118 –1125. Johns, L. J., Nazroo, J. Y., Bebbington, P., and Kuipers, E. 2002. Occurrence of hallucinatory experiences in a community sample and ethnic variations. Br. J. Psychiatry 180: 174 –178. Johnstone, E. C., Abukmeil, S. S., Byrne, M., Clafferty, R., Grant, E., Hodges, A., Lawrie, S. M., and Owens, D. G. C. 2000. Edinburgh high risk study—Findings after four years: Demographic, attainment and psychopathological issues. Schizophrenia Res. 46: 1–15. Lawrie, S. M., and Abukmeil, S. S. 1998. Brain abnormality in schizophrenia. A systematic and quantitative review of volumetric magnetic resonance imaging studies. Br. J. Psychiatry 172: 110 – 120. Lawrie, S. M., Whalley, H. C., Abukmeil, S. S., Kestelman, J. N., Donnelly, L., Miller, P., Best, J. J. K., Owens, D. G., and Johnstone, E. C. 2001. Brain structure, genetic liability and psychotic symptoms in subjects at high risk of developing schizophrenia. Biol. Psychaitry 49: 811– 823. Maguire, E. A., Gadian, D. G., Johnsrude, I. S., Good, C. D., Ashburner, J., Frackowiak, R. S., and Frith, C. D. 2000. Navigationrelated structural change in the hippocampi of taxi drivers. Proc. Natl. Acad. Sci. USA 97: 4398 – 4403. May, A., Ashburner, J., Buchel, C., McGonigle, D. J., Friston, K. J., Frackowiak, R. S., and Goadsby, P. J. 1999. Correlation between structural and functional changes in brain in an idiopathic headache syndrome. Nat. Med. 5: 836 – 838. Miller, P., Byme, M., Hodges, A., Lawrie, S., Owens, D. G. C., and Johnstone, E. C. 2002. Schizotypal components in people at high risk of developing schizophrenia: Early findings from the Edinburgh High Risk study. Br. J. Psychiatry 180: 179 –184. Nichols, T. E., and Holmes, A. P. 2001. Nonparametric analysis of PET functional neuroimaging experiments: A primer. Hum. Brain Mapping, 15: 1–25. Paillere-Martinot, M., Caclin, A., Artiges, E., Poline, J., Joliot, M., Mallet, L., et al. 2001. Cerebral gray and white matter reductions and clinical correlates in patients with early onset schizophrenia. Schizophrenia Res. 50: 19 –26.
889
Shah, P. J., Ebmeier, K. P., Glabus, M. F., and Goodwin, G. M. 1998. Cortical grey matter reductions associated with treatment-resistant chronic unipolar depression. Controlled magnetic resonance imaging study. Br. J. Psychiatry 172: 527–532. Sigmundsson, T., Suckling, J., Maier, M., Williams, S. C. R., Bullmore, E. T., Greenwood, K. E., Fukuda, R., Ron, M. A., and Toone, B. K. 2001. Structural abnormalities in frontal, temporal, and limbic regions and interconnecting white matter tracts in schizophrenic patients with prominent negative symptoms. Am. J. Psychiatry 158: 234 –243. Sowell, E. R., Thompson, P. M., Holmes, C. J., Batth, R., Jernigan, T. L., and Toga, A. W. 1999. Localizing age-related changes in brain structure between childhood and adolescence using statistical parametric mapping. NeuroImage 9: 587–597. Talairach, J., and Tournoux, P. 1988. A Coplanar Stereotaxic Atlas of a Human Brain. Thieme, Stuttgart. Whalley, H. C., Kestelman, J. N., Rimmington, J. E., Kelso, A., Abukmeil, S. S., Best, J. J., Johnstone, E. C., and Lawrie, S. M. 1999. Methodological issues in volumetric magnetic resonance imaging of the brain in the Edinburgh High Risk Project. Psychiatry Res. 91: 31– 44. Wilkie, M., Kaufmann, C., Grabner, A., Pu¨ tt, B., Wetter, T. C., and Auer, D. P. 2001. Gray matter changes and correlates of disease severity in schizophrenia: A statistical parametric mapping study. NeuroImage 13: 814 – 824. Woermann, F. G., Free, S. L., Koepp, M. J., Ashburner, J., and Duncan, J. S. 1999. Voxel-by-voxel comparison of automatically segmented cerebral gray matter—A rater-independent comparison of structural MRI in patients with epilepsy. NeuroImage 10: 373– 384. Worsley, K. J., Marrett, S., Neelin, P., and Evans, A. C. 1996. Searching scale space for activation in PET images. Hum. Brain Mapping 4: 74 –90. Wright, I. C., McGuire, P. K., Poline, J. B., Travere, J. M., Murray, R. M., Frith, C. D., Frackowiak, R. S., and Friston, K. J. 1995. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. NeuroImage 2: 244 –252. Wright, I. C., Ellison, Z. R., Sharma, T., Friston, K. J., Murray, R. M., and McGuire, P. K. 1999a. Mapping of grey matter changes in schizophrenia. Schizophrenia Res. 35: 1–14. Wright, I. C., Sharma, T., Ellison, Z. R., McGuire, P. K., Friston, K. J., Brammer, M. J., Murray, R. M., and Bullmore, E. T. 1999b. Supra-regional brain systems and the neuropathology of schizophrenia. Cereb. Cortex 9: 366 –378. Wright, I. C., Rabe-Hesketh, S., Woodruff, P. W., David, A. S., Murray, R. M., and Bullmore, E. T. 2000. Meta-analysis of regional brain volumes in schizophrenia. Am. J. Psychiatry 157: 16 –25.
Structural Gray Matter Differences between First-Episode Schizophrenics and Normal Controls Using Voxel-Based Morphometry 1 Dominic E. Job,* Heather C. Whalley,* Sarah McConnell,* Mike Glabus,† Eve C. Johnstone,* and Stephen M. Lawrie* *Department of Psychiatry, University of Edinburgh, Kennedy Tower, Royal Edinburgh Hospital, Morningside Park, Edinburgh EH10 5HF, Scotland, United Kingdom; and †Unit on Integrative Neuroimaging, Clinical Brain Disorders Branch, National Institute of Mental Health, National Institutes of Health, Bethesda, Maryland 20892 Received October 30, 2001
The aim of this study was to compare the gray matter segments from T1 structural MR images of the brain in first-episode schizophrenic subjects (n ⴝ 34) and normal control subjects (n ⴝ 36) using automated voxel-based morphometry (VBM). This study is novel in that few studies have examined subjects in their first episode of schizophrenia. The subjects were recruited for the Edinburgh High Risk project and regional brain volumes were previously measured using a semi-automated volumetric region of interest (ROI) method of analysis. The primary interest was to compare the results from the compatible parts of the ROI study and the primary VBM approach. Our secondary interest was to compare the results of a study-specific template that was constructed from the control group to those using the generic T1 template (152 Montreal Neurological Institute brains) supplied with SPM99 (statistical parametric mapping). The images were processed and statistically analyzed using the SPM99 program. VBM analysis identified significant decreases in gray matter in the schizophrenics relative to the normal control group at the corrected voxel level (P < 0.05) in the right anterior cingulate, right medial frontal lobe, left middle temporal gyrus, left postcentral gyrus, and the left limbic lobe. There were no increases in gray matter in the schizophrenics relative to the control group. The construction of a customized template appeared to improve the detection of structural abnormalities. The analyses were subsequently restricted to voxels within the amygdala– hippocampal complex using the SPM small-volume correction. This identified gray matter decreases in the schizophrenics, at the corrected voxel level (P < 0.05), in the left and right uncus and parahippocampal gyri and the right amygdala. These results are compatible with and extend the relevant findings of the previous volumetric ROI analysis, when allowing for the differ-
1
This study was funded by a program grant from the Medical Research Council of Great Britain. 1053-8119/02 $35.00 © 2002 Elsevier Science (USA) All rights reserved.
ences between the methods and interpretation of their results. © 2002 Elsevier Science (USA)
INTRODUCTION Many studies examining structural magnetic resonance (MR) images of the brain in people with schizophrenia have indicated that they have subtle anatomical differences compared to normal control groups. Quantitative reviews of these studies have suggested that, in general, the temporal lobes, medial temporal lobe structures, and superior temporal gyri are smaller and the ventricles are larger in schizophrenics (Lawrie and Abukmeil, 1998; Wright et al., 2000). Most of these studies have employed region of interest (ROI) measurement of brain structures by manually drawing around the regions of interest. This type of analysis is prone to measurement error and is very time consuming. This restricts the number of regions that can be practically measured and means that relatively large structures of interest must be defined at the outset of the study. Voxel-based morphometry (VBM) is an automated method of examining structural MR images of the brain (Ashburner and Friston, 2000; Wright et al., 1995) based upon the statistical parametric mapping (SPM) techniques originally designed for analysis of functional images (Friston et al., 1995a). This method requires no manual tracing around regions and has the advantage of examining the brain as a whole. However, VBM also has limitations, resulting in difficulty in interpretation of VBM results. Other studies have used the same or variations of these methods to examine structural MR images from a variety of subject groups, including schizophrenic patients (Chua et al., 1997; Gaser et al., 1999; Wilkie et al., 2001; Wright et al., 1995, 1999a), people with depression (Shah et al., 1998), children and adolescents (Sowell et al., 1999), subjects with headache (May et al., 1999) and epilepsy (Woermann et al., 1999),
880
881
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
and taxi drivers as they acquire spatial knowledge (Maguire et al., 2000). The aim of this study was to compare the gray matter segments in first-episode schizophrenics and normal controls using this automated technique. The hypotheses tested were that the schizophrenic group would differ from the control group in the regions predicted by the literature and as indicated by the relevant findings in our previous ROI study, which showed volume reduction in the schizophrenic group, relative to the control group, in the left (P ⫽ 0.0003, uncorrected) and right (P ⫽ 0.01, uncorrected) amygdala– hippocampal complex (AHC) (Lawrie et al., 2001). METHODS Participants The subjects were recruited for the Edinburgh High Risk Study (Hodges et al., 1999; Johnstone et al., 2000). The subject groups consisted of 34 subjects in their first episode of schizophrenia (American Psychiatric Association, 1994) and 36 normal control subjects matched groupwise for age. The first-episode schizophrenia patients were identified from admissions to the Royal Edinburgh Hospital and associated psychiatric hospitals in Lothian region, Scotland. These subjects have no known family history of schizophrenia in their firstor second-degree relatives. Their case notes were reviewed with a structured assessment and the patients were examined with structured psychiatric interviews to confirm the diagnosis. One hundred fifty subjects at high risk of developing schizophrenia were also recruited but are not included in this paper. Three of the control subjects had isolated psychotic symptoms (mainly isolated hallucinations) in psychiatric interviews but were not excluded from the study to avoid biasing the control group (Miller et al., 2002). Control subjects had been recruited on the basis that there were no major psychiatric disorders or family history of the same. The inclusion of three controls with symptoms but no disorder is a strength of this study, as a similar proportion of the general population will have such symptoms (Johns et al., 2002). Details of the subject groups are presented in Table 1. Further details of the subject groups can be found in Johnstone et al. (2000). Ethical approval was obtained from the relevant health authorities and all subjects gave written informed consent. Scanning Protocol MR brain scanning was performed on a 42 SPE Siemens (Erlangen, Germany) Magnetom operating at 1.0 T. Midline sagittal localization was followed by two sequences to image the whole brain. The first scan was a double spin echo sequence which gave simultaneous
TABLE 1 Details of the Subject Groups
Age, years (sd) Male:female Handedness, mixed:left:right Height (sd) Paternal social class at birth, manual:nonmanual:unknown
Control (n ⫽ 36)
Schizophrenic (n ⫽ 34)
21.17 (2.37) 17:19 2:3:31 170.21 (9.09) 19:15:2
21.35 (3.66) 23:11 2:1:31 174.16 (11.85) 14:11:9
proton density and T2-weighted images (TR ⫽ 3565 ms, TE ⫽ 20 and 90 ms, 31 contiguous 5-mm slices in the Talairach plane, field of view 250 ⫻ 250 mm) to identify any gross brain lesions. The second scan was a three-dimensional magnetization prepared rapid acquisition gradient echo sequence consisting of a 180° inversion pulse followed by a fast low angle shot collection (flip angle 12°, TR ⫽ 10 ms, TE ⫽ 4 ms, TI ⫽ 200 ms, and relaxation delay time 500 ms, field of view 250 ⫻ 250 mm) to give 128 contiguous “slices” of 1.88-mm thickness. Immediately after each subject’s scan a large plain test object filled with light oil was scanned in exactly the same place in the coil and in the same orientation as the subject’s head. The test object data were used to correct for inhomogeneity in the radiofrequency coil. The mean intensity of a 5 ⫻ 5 pixel square in the center slice of the test object data was taken as the optimum intensity. The whole data set was normalized to this value. By dividing the patient data by the normalized data any deviation from the optimum coil response could be corrected for global differences in signal inhomogeneities. To address intensity shifts across each subject’s image the phantom image for each subject was superimposed upon each subject’s brain image. This standardizes (calibrates) the signal intensity of the scanner and counteracts falloff in intensity that is due to the scanner coil inhomogeneity. This produces images whose variations in signal intensity are due to the composition of the brain and not to the magnetic coil of the scanner. Details of the volumetric analysis on the same subject groups have been described previously (Lawrie et al., 2001; Whalley et al., 1999). In brief, three blinded raters with high reliability used automated edge detection and manual editing to measure the volumes of the whole brain, prefrontal and temporal lobes, the AHC, thalamus, lentiform, caudate, and the ventricular system for controls, high-risk, and first-episode schizophrenics.
882
JOB ET AL.
Image Processing Data were transferred to a Sun SPARC 1 workstation and then analyzed using Analyze version 7.5.5 (Mayo Foundation, Rochester, MN) and Statistical Parametric Mapping version 99 (SPM99) (Wellcome Department of Cognitive Neurology, Institute of Neurology, London) (Friston et al., 1995a,b) running in MATLAB version 5.3 (The MathWorks, Natick, MA). Images were analysed using voxel-based morphometry methods as described by Ashburner and Friston (Ashburner and Friston, 2000). All MR images were initially loaded into Analyze, corrected for field inhomogeneity as described above, and converted from 16-bit to 8-bit images. The default settings in SPM were then altered to read in the subject’s scans which were in the coronal plane and in radiological convention. Two templates were used in the study. The first template used was the generic T1 template included in the SPM package, composed of 152 subjects, mean age 25 years, 43% female, and 90% right-handed. The second template, the Edinburgh High Risk Study (EHRS) T1 template, was made from 33 healthy control images using SPM99, mean age 21 years, 53% female, and 86% right-handed. The three controls that had psychotic symptoms, principally fragmentary hallucinations, in the context of drug abuse, were not used in the construction of the EHRS T1 template. To construct the EHRS T1 template the healthy control images were spatially normalized to the generic T1 image using linear normalization and nearest neighbor interpolation. Extracerebral voxels were removed using the “Xtract brain” render function. This renders the gray and white matter segments and produces an “extracted brain.” Multiplying the original gray matter segment by the extracted brain file removes the extracerebral voxels. A mean image was then calculated and smoothed at 8 mm full width at half maximum (FWHM). All images were spatially normalized to the generic T1 and to the EHRS T1 template. This process involved minimizing the residual sum of squares differences between the images and each template and was performed using a 12-point linear affine transformation. Nearest neighbor interpolation was used to preserve the original intensity values and gray/white matter contrasts of the images. The images were then segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using the SPM modified clustering algorithm and SPM’s GM, WM and CSF a priori templates. This was performed with no inhomogeneity correction as these images had previously been corrected (as described above). The extracerebral voxels were then removed from the gray and white matter segments using the Xtract
brain render function in SPM99. The spatially normalized and segmented images were then smoothed with a 12-mm FWHM isotropic Gaussian kernel to remove the differences in sulcal/gyral patterns among individuals that may not be accounted for by normalization (Ashburner and Friston, 2000). The 12-mm width was chosen for the smoothing kernel to reduce the number of false positives, to conform with the width nascent in the use of the SPM99 package, and to facilitate comparison with other studies (Chua et al., 1997; Wilkie et al., 2001; Wright et al., 1999a). The 12-mm Gaussian smoothing filter promotes the detection of differences in structures of around 12-mm spatial extent (Ashburner and Friston, 2001). As all brain images are normalized to a standard template prior to group analysis, large-scale differences among the groups are eliminated. The resulting gray matter segments for each of the subjects were then entered into statistical analysis. A randomization analysis is generally desirable (Ashburner and Friston, 2001) but was not necessary for the ROI–VBM confirmation. Randomization testing is most pertinent in studies with low degrees of freedom (Nichols and Holmes, 2001). In this work the primary VBM contrast has 62° of freedom, and randomization testing is therefore not expected to have a significant impact on the results. However, randomization analysis and scale space search (Worsley et al., 1996) may be part of future work. Statistical Analysis Group comparisons were performed using the a random effects analysis “single subject: conditions and covariates,” a statistical model in SPM99 based on the general linear model. Nuisance or confound variables were entered into the analyses to remove global differences between the images (ANCOVA). In the volumetric ROI study, the whole brain volume (WBV) was measured (in native space) to control for global effects, so this measure was also used in the primary VBM study to replicate the conditions from the ROI study. Gender, handedness, height, age, and paternal social class at birth were also entered as nuisance variables into the model, to replicate the analysis performed in the ROI study (Lawrie et al., 2001). Handedness was determined by the subjects preferred writing hand. In the conventional VBM approach, the automated number of voxels (NVOX) measure is commonly used as a nuisance variable to remove global differences between the images, while in ROI studies the WBV measure is typically used. To illustrate any differences between the two approaches in implementing this nuisance variable, a further analysis to compare the effects of NVOX and WBV was performed. These two measures were compared in VBM using the images normalized to the EHRS T1 template. In the volumetric ROI study the WBV measure included white mat-
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
883
TABLE 2 The Primary VBM Study: EHRS T1 Template, Covariates Matching Those in the ROI Study Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
The primary VBM study: Results for the whole brain 0.001 0.031 0.017 0.019 0.052 0.042 0.054 0.084
3.96 3.96 ⫺55.44 ⫺63.36 ⫺45.54 ⫺11.88 ⫺19.80 55.44
40.87 45.83 ⫺4.46 ⫺12.64 ⫺0.42 ⫺0.84 ⫺5.05 ⫺22.98
1.64 ⫺15.75 ⫺11.55 19.06 ⫺8.39 ⫺16.78 ⫺23.30 6.68
Right, anterior cingulate, Brodmann area 32 Right, medial frontal gyrus, Brodmann area 11 Left, middle temporal gyrus Left, postcentral gyrus, Brodmann area 43 Left, superior temporal gyrus Left, limbic lobe (medial border of amygdala) Left, uncus, Brodmann area 28 Right, superior temporal gyrus, Brodmann area 41
The primary VBM study: Small-volume correction amygdala–hippocampus 0.001 0.002 0.028 0.038
⫺19.80 ⫺13.86 23.76 21.78
⫺5.05 ⫺0.84 ⫺5.14 ⫺6.90
⫺23.30 ⫺16.78 ⫺24.98 ⫺21.52
Left, uncus, Brodmann area 28 Left, parahippocampal gyrus Right, parahippocampal gyrus Right, uncus, amygdala
Note. EHRS, Edinburgh High Risk Study. Nuisance covariates: whole brain volume, handedness, sex, age, height, and paternal social class.
ter. The global gray matter, NVOX, was determined for use in the VBM study using a matlab script “get_globals.m” (J. Ashburner), a measure that does not include white matter. This calculated the number of voxels within each image and was performed on segmented raw images, derived using the internal 12point affine normalization only, in SPM99. Two contrasts were constructed, one to examine regional decreases and the other to examine regional increases in gray matter between the two groups. The results of these contrasts were then displayed as a statistical parametric map thresholded at a significance level of P ⬍ 0.05 (corrected). All results described are based on the voxel level of significance. The analysis was then restricted to voxels contained within the amygdala– hippocampus (4.9 resels, 12-mm FWHM smoothing). This was performed using the “small volume correction” (SVC) function in SPM99 and was also thresholded at a significance level of P ⬍ 0.05 (corrected). The coordinates of the significant voxels were converted from MNI (Montreal Neurological Institute) space to Talairach and Tournoux (Talairach and Tournoux, 1988) coordinates using the “mni2tal.m” matlab script, a nonlinear transform approach described by Brett (1999). RESULTS Table 2 and Fig. 1 present the results for the primary VBM approach, using the same confounds as used in the ROI study: whole brain volume, gender, handedness, height, age, and paternal social class at birth.
The results show reductions in gray matter in schizophrenics versus controls in the right anterior cingulate, right medial frontal lobe, left middle temporal gyrus, left postcentral gyrus, and left limbic lobe. The SVC results indicate reductions in gray matter in schizophrenics versus controls in the left and right uncus, left and right parahippocampal gyri, and the right amygdala. Table 3 presents the results for the EHRS T1 template, using the NVOX confound only. Here, the schizophrenic group showed a relative decrease in gray matter compared to the control group similar to the regions described above, i.e., the right anterior cingulate, right medial frontal gyrus, left middle temporal gyrus, and left postcentral gyrus and at the SVC level the left uncus, left and right parahippocampal gyri, and left amygdala. With the generic T1 template (see Table 4), using the NVOX confound only, the standard SPM approach, the schizophrenic group showed a relative decrease in gray matter compared to the control group in the right anterior cingulate, right medial frontal gyrus, left postcentral gyrus, and left uncus at the SVC level. The comparison to illustrate the differences between the two measures of nuisance variables (NVOX and WBV) showed relatively little difference in the two measures. The NVOX measure, with the EHRS T1 template (Table 3), showed changes as given above. The WBV measure, with the EHRS T1 template (Table 5), showed changes in the right anterior cingulate, left postcentral gyrus, left middle temporal gyrus, left uncus, and left and right parahippocampal gyri. The differences between the effects of the two measures,
884
JOB ET AL.
FIG. 1. Areas of gray matter reduction in schizophrenics versus healthy controls at a P-uncorrected level of 0.001. The color bar shows Z values corresponding to colors in the figure. The numbers beside each slice show the height on the z-axis in millimeters of the slice.
NVOX and WBV, can be seen in the differences between Tables 3 and 5. At the whole brain level Table 3 shows a difference in the right medial frontal gyrus
Brodman area 11 not seen in Table 5. At the whole brain level Table 5 shows a difference in the left and right anterior cingulate, but only in the right anterior
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
885
TABLE 3 EHRS T1 Template, with Number of Voxels (NVOX) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
EHRS T1, number of voxels: Results for the whole brain 0.002 0.019 0.033 0.003 0.027
3.96 40.97 3.48 3.96 39.86 20.11 3.96 43.896 ⫺15.65 ⫺63.36 ⫺14.58 19.15 ⫺63.36 ⫺44.38 5.91
Right, anterior cingulate, Brodmann area 32 Right, anterior cingulate Right, medial frontal gyrus, Brodmann area 11 Left, postcentral gyrus, Brodmann area 43 Left, middle temporal gyrus, Brodmann area 22
EHRS T1, number of voxels, small-volume correction amygdala–hippocampus ⫺13.86 ⫺21.78 ⫺19.80 23.76
0.007 0.029 0.032 0.043
⫺0.84 ⫺3.20 ⫺4.97 ⫺5.14
⫺16.78 ⫺25.07 ⫺21.62 ⫺24.98
Left, parahippocampal gyrus Left, uncus Left, uncus, amygdala Right, parahippocampal gyrus
Note. EHRS, Edinburgh High Risk Study.
cingulate as in Table 3. At the SVC AHC level Table 3 shows a difference in the left uncus and amygdala not seen in Table 5. There were no regions where there was an increase in gray matter in the schizophrenics relative to the control group in any of the different approaches. DISCUSSION Region of interest analysis is the gold standard for assessing structural MR data, but it is slow and not perfectly reliable. VBM is faster and less labor intensive and can be applied to the whole brain, but also has limitations. In this study we found significant differences between the controls and the schizophrenic groups in the left anterior cingulate, right medial frontal gyrus, left postcentral gyrus, and left medial temporal gyrus. The postcentral gyrus difference could not have been discovered by our volumetric ROI method as it lies outside any of our regions of interest (other than whole brain), while the other additional results—right medial frontal gyrus, anterior cingulate, and left me-
dial temporal gyrus—were only included within larger structures (prefrontal and temporal lobe volumes, respectively) and are therefore not comparable to the primary VBM study. An additional ROI study would be necessary to confirm the extended findings in the primary VBM study. In the ROI study we only found differences specifically between the first-episode patients and control subjects bilaterally in the amygdala– hippocampal complex. The VBM method can use a prior hypothesis for a region to strengthen the power of the statistical analysis of that region and gives a specific point of maximal change, rather than only confirming an expected change over a prespecified search volume. The ROI volume decrease in AHC gives little detail compared to the VBM findings with small-volume correction, but the detailed findings of the VBM study can be difficult to interpret. The point of maximal change must be carefully interpreted as a statistically significant difference of the probabilities of the voxels in each group being GM. Additionally, the 12-mm smoothing step can obscure the exact location of one or more points of change.
TABLE 4 SPM T1 Generic Template, with Number of Voxels (NVOX) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
Generic T1, number of voxels: Results for the whole brain. 0.004 0.007 0.013
3.96 40.97 3.48 3.96 47.85 ⫺14.17 ⫺65.34 ⫺16.43 21.09
Right, anterior cingulate, Brodmann area 32 Right, medial frontal gyrus, Brodmann area 11 Left, postcentral gyrus, Brodmann area 43
Generic T1, number of voxels, small-volume correction amygdala–hippocampus. 0.042 0.048
⫺21.78 ⫺17.82
⫺3.20 ⫺25.07 ⫺4.97 ⫺21.61
Left, uncus Left, uncus
886
JOB ET AL.
TABLE 5 EHRS T1 Template, with Whole Brain Volume (WBV) Confound Voxel, P corrected
x, y, z {mm} Talairach coordinates
Point of maximal change
EHRS T1, whole brain volume: Results for the whole brain 0.001 0.010 0.003 0.015 0.084
3.96 40.97 3.96 39.76 ⫺63.36 ⫺12.64 ⫺63.36 ⫺44.38 ⫺1.98 7.33
3.48 18.28 19.06 5.91 ⫺8.78
Right, anterior cingulate, Brodmann area 32 Right, anterior cingulate Left, postcentral gyrus, Brodmann area 43 Left, middle temporal gyrus, Brodmann area 22 Left, anterior cingulate
EHRS T1, whole brain volume, small-volume correction for amygdala–hippocampus. 0.018 0.047 0.038
⫺13.86 ⫺.84 ⫺21.78 ⫺3.20 23.76 ⫺5.14
⫺16.78 ⫺25.07 ⫺23.98
Left, parahippocampal gyrus Left, uncus Right, parahippocampal gyrus
Note. EHRS, Edinburgh High Risk Study.
The differences in gray matter between the schizophrenic group and the normal control group are consistent with previous VBM reports of gray matter reductions in the superior temporal gyrus, amygdala, and insula (Wright et al., 1999a,b) and in a previous deformation-based approach which found gray matter reductions bilaterally in the frontal lobes and superior temporal gyri (Gaser et al., 1999). Our comparable ROI results and primary contrast VBM (Table 2) results are consistent for the AHC, taking into account that the findings are so similar that the inevitable methodological differences between the two methods, and each method’s shortcomings, outweigh the differences between the results from each method. The primary VBM approach shows a difference in the right amygdala and bilaterally in the parahippocampal gyrus, whereas the ROI study shows differences bilaterally in the AHC. It is of note that the VBM method shows differences mainly around the borders of the AHC or hippocampus, not only within the AHC and not in the hippocampus proper. Small movements in the point of maximal change can easily fall either side of a border line. For example, in Table 2 the left uncus maximal point is only 0.08 mm on the y axis and 1.68 mm on the z axis different to the left uncus/amygdala in Table 3. In our study, the small-volume correction function in SPM99 used images created by editing the relevant nonsmoothed EHRS T1 template to define the small volume, which contained parts of some bordering structures since the template is a mean image. Results from the secondary SVC contrasts are less consistent between different statistical and preprocessing manipulations. This shows that VBM is sensitive to the statistical confounders that are used (see Tables 3 and 5, for a comparison of two methodologically equivalent confounds), but more differences are found using study-specific templates in preprocessing the images (see Tables 3 and 4, for a template comparison). This
template sensitivity is probably attributable to a combination of using a template from matched controls and the same scanner, although we cannot tell whether one or other effect is more important. The primary VBM results show a relative rather than absolute replication of the comparable ROI results, but the results between different approaches (Tables 3, 4, and 5) are certainly compatible. An interesting inconsistency between the treatments is that in Table 2, the primary VBM contrast shows a significant reduction in the right uncus and amygdala, whereas Table 3 shows reductions in the left uncus and amygdala and Tables 4 and 5 show reductions in the left uncus. This hemisphere switch must be due to the confounders used in the primary VBM study. In the primary study, confounders include age, gender, handedness, height, and paternal social class at birth as confounds. It is well established that handedness and gender have some structural brain correlates, and therefore it is not unexpected that the statistical inference that includes these confounds shows a change in laterality of the amygdala finding, compared to those that do not. In our study, it may be that the ventricles in the schizophrenic subjects are reduced during spatial normalization and therefore the gray matter near them is reduced. It is therefore possible that structural differences in ventricular volumes can show up in a VBM study of gray matter volume. To avoid this problem spatial normalization can be based on only the segmented gray matter, where any significant differences are much more likely to be attributable to actual gray matter differences (Ashburner and Friston, 2001). Gray matter only normalization (after Good et al., 2001) is desirable and may be part of future work. Standard volumetric analyses are prone to measurement error. In the volumetric study all three raters obtained good interrater reliability figures (ICC range 0.75– 0.99) and assessed the scans blind to the clinical
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA
status of the subject, but reliability was generally the lowest for small structures (Whalley et al., 1999). It should also be noted that while the ROI method measures differences over a specified region, the VBM method gives the points of maximal change for clusters of points of change (Wright et al., 1999a). Our VBM analysis shows differences mainly around the borders of the AHC. The ROI and VBM approaches can both suffer from partial volume effects, e.g., near the borders of the amygdala/parahippocampal gyrus, where some voxels are half gray matter and half white matter, it is difficult to threshold voxels into one class. While VBM is consistent in classification of partial volumes, it is also the case that white matter tracts such as the alveus and fimbria can be misclassified as gray matter during segmentation. CSF from the temporal horns of the lateral ventricles surrounding much of the hippocampal surface is often misclassified as gray matter, given that the signal intensities do not threshold as cleanly as they do in the cortex. The choroid plexus in the temporal horns could further obscure the tissue boundaries of the hippocampus. In medial temporal regions signal intensity values from different tissue types may be indistinguishable. However, this misclassification is the same for both subject groups, and smoothing should reduce any effects this has on the final statistics. These factors may account for subtle differences in the results between our two methods and among other ROI studies. This study has also shown compatible results for images spatially normalized to both the generic SPM99 T1 template and the specific EHRS T1 template constructed from images obtained in age-matched control subjects with the same scanner. However, constructing a customized template appears to improve the detection of structural abnormalities. The EHRS T1 template probably gives greater sensitivity due to this template being a study-specific template, matching the scanner and age of the subjects, an effect also described by Good et al. (2001). The comparison of the NVOX (Table 3) and WBV (Table 5) analyses shows that they give similar results. The NVOX measure could differ from the ROI approach as the NVOX measure is calculated over GM only, whereas the WBV measure includes WM. Segmentation for calculation of the NVOX measure in VBM can be poor as the process relies on overlaying a priori probability maps (in stereotactic space) onto internally normalized raw images to calculate the probability of any given voxel being GM/WM or CSF. This is susceptible to registration error as discussed in detail by Ashburner and Friston (Ashburner and Friston, 2000). Even when the raw images have been normalized using a 12-point linear affine transformation, it is possible that voxels containing a mixture of GM and WM can be classified as GM or WM, when strictly speaking they are a mixture, leading to further poten-
887
tial discrepancies between the ROI and VBM studies. Our confirmation of ROI results suggests, however, that such misclassification is not a serious problem, as do other VBM–ROI validations (Baron et al., 2001; Wright et al., 1999a). Its has been shown that VBM is capable of detecting differences in small areas such as the hippocampus (Maguire et al., 2000; Wright et al., 1999a), but it may be a less sensitive measure than ROI approaches for detecting volume differences. VBM is also sensitive to the confounds and templates used, and that makes the results difficult to interpret, especially for the AHC. We believe that the two approaches are complementary, with the ROI giving more detail regarding the difference in volumes of interest and the VBM having more detail regarding the specific points of maximal change. One cannot say from our results whether the differences between schizophrenics and controls are in the amygdala– hippocampal complex or surrounding structures or, as seems most likely, both, but they are certainly in the medial temporal lobe and include the amygdala (see meta-analyses and other VBM studies cited below). Indeed, the postmortem literature in schizophrenia has also less commonly found differences in the amygdala– hippocampal complex and more commonly in the parahippocampal gyrus (Harrison, 1999). A further possible limitation of VBM in this study is that a smoothing kernel of 12-mm isotropic FWHM Gaussian kernel could obscure small differences in hippocampus. However, we used linear rather than nonlinear normalization, leaving higher variability, such that a smaller smoothing filter would be inappropriate. Ashburner and Friston (2000) recommend a 12-mm filter and smoothing with a 12-mm FWHM Gaussian will best find differences in areas of 12-mm spatial extent (such as the amygdala– hippocampal complex rather than the hippocampus alone). The segmentation procedure may well include parts of the alveus and fimbria, potentially rendering some of their white matter as GM voxels, but it is the same for all images in all groups and therefore one would expect that the differences shown in the contrasts are real differences in structure. Smoothing is done after segmentation and this would reduce the impact of a small area, such as the alveus and fimbria, on the results. However, the potential misclassification of WM as GM does require cautious interpretation of the results. Bookstein (2001) points out that misregistration of small regions of interest in VBM can lead to the detection of systematic shape differences between groups. In this work we have not used nonlinear normalization, so there is no local registration and smoothing is intended to account for variability at a local level, i.e., in small regions of interest. In their reply to Bookstein (2001), Ashburner and Friston (2001) state that careful consideration of what may be the cause of any detected
888
JOB ET AL.
difference is required. Both Bookstein (2001) and Ashburner and Friston (2001) point out that many systematic differences can arise. For example, in our study, the schizophrenic subjects may have moved more or been systematically positioned differently in the scanner, producing a motion artifact in the images, resulting in biased segmentation. However, we have previously shown that the patients’ brains were not significantly more yawed than the controls, and there were no gross movement artifacts (Whalley et al., 1999). We specifically did not find differences in the left or right hippocampi in our VBM study. This could be due to its small size and spatial variability between subjects, which may have been implicitly accounted for by visual inspection in the ROI study. Such spatial variability has been described by Good et al. (2001). The ROI analysis was performed on scans without segmentation, such that the volumetric measurement of the hippocampus included regions of surrounding white matter (alveus and fimbria) which would not be analyzed in the VBM method (assuming accurate segmentation and adequate smoothing) which was performed solely on the thresholded gray matter segments. This, and the other limitations of VBM already discussed, may explain why some previous VBM studies in schizophrenia have found hippocampal reductions (Sigmundsson et al., 2001; Hulshoff et al., 2001) and others have not (Wright et al., 1999; Paillere-Martinot et al., 2001). All of these previous studies have, however, found reductions in the amygdala and/or parahippocampal gyrus. These VBM studies show precedence of the abnormalities we found in anterior cingulate (Sigmundsson et al., 2001), medial frontal gyri (Sigmundsson et al., 2001; Paillere-Martinot et al., 2001), left limbic lobe/amygdala (Hulshoff et al., 2001), right uncus/ amygdala (Wright et al., 1999; Hulshoff et al., 2001), and left parahippocampal gyrus (Sigmundsson et al., 2001; Paillere-Martinot et al., 2001). The findings in these studies closely match findings in our primary VBM contrast (Table 2 results). Abnormalities of the anterior cingulate/medial frontal lobe in schizophrenia are commonly reported in functional imaging studies of schizophrenia (e.g., Ebmeier et al., 1995), may represent part of the frontotemporal dysconnectivity that is currently thought to be the basis of the disorder (e.g., Fletcher et al., 1999), and may specifically be related to the negative symptoms of a loss of affect and drive (e.g., Chua et al., 1997). Volume reductions of the medial temporal lobe are the most consistently replicated ROI and VBM findings in schizophrenia (Lawrie and Abukmeil, 1998; Wright et al., 2000) and may underlie frontolimbic dysconnectivity and the positive symptoms of hallucinations, delusions, and thought disorder in particular (Wright et al., 1995; Chua et al., 1997).
One strength of our study of first-episode schizophrenics is that treatment effects are less likely to confound the results than might be the case in patients who have been on antipsychotic drugs for years. While this suggests a neurodevelopmental origin of the findings, we do not think that we can separate trait and state effects in the current study. In our study of highrisk subjects this may be possible as all subjects are antipsychotic free and repeated scans have been conducted over a period of symptomatic change. Future ROI and VBM studies in patients with schizophrenia and related populations should be able to specify the structural brain abnormalities in greater detail and link them to particular etiological factors and symptoms. REFERENCES American Psychiatric Association. 1994. Diagnostic and Statistical Manual of Mental Disorders, 4th Edition. American Psychiatric Association, Washington, DC. Ashburner J., Functional Imaging Laboratory. Wellcome Department of Cognitive Neurology, London, UK. Ashburner, J., and Friston, K. J. 2000. Voxel-based morphometry— The methods. NeuroImage 11: 805– 821. Ashburner, J., and Friston, K. J. 2001. Comments and controversies—Why voxel-based morphometry should be used. NeuroImage 14: 1238 –1243. Baron, J. C., Chetelat, G., Desgranges, B., Perchey, G., Landeau, B., de la Sayette, V., and Eustache, F. 2001. In vivo mapping of gray matter loss with voxel-based morphometry in mild Alzheimer’s disease. NeuroImage 14: 298 –309. Bookstein, F. L. 2001. “Voxel-based morphometry” should not be used with imperfectly registered images. NeuroImage 14: 1454 – 1462. Brett, M. 1999. The MNI brain and the Talairach atlas. MRC Cognition and Brain Sciences Unit. http://www.mrc-cbu.cam.ac.uk/ Imaging/. Chua, S. E., Wright, I. C., Poline, J. B., Liddle, P. F., Murray, R. M., Frackowiak, R. S. J., Friston, K. J., and McGuire, P. K. 1997. Grey matter correlates of syndromes in schizophrenia—A semi-automated analysis of structural magnetic resonance images. Br. J. Psychiatry 170: 406 – 410. Ebmeier, K. P., Lawrie, S. M., Blackwood, D. H. R., Johnstone, E. C., and Goodwin, G. M. 1995. Hypofrontality revisited: A high-resolution single photon emission tomography study in schizophrenia. J. Neurol. Neurosur. Psychiatry 58: 452– 456. Fletcher, P. C., McKenna, P. J., Friston, K. J., Frith, C. D. and Dolan, R. J. 1999. Abnormal cingulate modulation of fronto-temporal connectivity in schizophrenia. NeuroImage 9(3): 337–342. Friston, K. J., Ashburner, J., Poline, J. B., Frith, C. D., Heather, J. D., and Frackowiak, R. S. J. 1995a. Spatial registration and normalisation of images. Hum. Brain Mapping 2: 165–189. Friston, K. J., Holmes, A. P., Worsley, K. J., Poline, J. B., Frith, C. D., and Frackowiak, R. S. J. 1995b. Statistical parametric maps in functional imaging. Hum. Brain Mapping 2: 189 –210. Gaser, C., Volz, H. P., Kiebel, S., Riehemann, S., and Sauer, H. 1999. Detecting structural changes in whole brain based on nonlinear deformations—Application to schizophrenia research. NeuroImage 10: 107–113. Good, C. D., Johnsrude, I. S., Ashburner, J., Henson, R. N., Friston, K. J., and Frackowiak, R. S. 2001. A voxel-based morphometric
GRAY MATTER REDUCTIONS IN SCHIZOPHRENIA study of ageing in 465 normal adult human brains. NeuroImage 14: 21–36. Harrison, P. J. 1999. The neuropathology of schizophrenia. A critical review of the data and their interpretation. Brain 122: 593– 624. Hodges, A., Byrne, M., Grant, E., and Johnstone, E. 1999. People at risk of schizophrenia. Sample characteristics of the first 100 cases in the Edinburgh high-risk study. Br. J. Psychiatry 174: 547–553. Hulshoff Pol, H. E., Schnack, H. G., Mandl, R. C. W., van Haren, N. E. M., Konig, H., Collins, D. L., Evans, A. C., and Kahn, R. S. 2001. Focal gray matter density changes in schizophrenia. Arch. Gen. Psychiatry 58(12): 1118 –1125. Johns, L. J., Nazroo, J. Y., Bebbington, P., and Kuipers, E. 2002. Occurrence of hallucinatory experiences in a community sample and ethnic variations. Br. J. Psychiatry 180: 174 –178. Johnstone, E. C., Abukmeil, S. S., Byrne, M., Clafferty, R., Grant, E., Hodges, A., Lawrie, S. M., and Owens, D. G. C. 2000. Edinburgh high risk study—Findings after four years: Demographic, attainment and psychopathological issues. Schizophrenia Res. 46: 1–15. Lawrie, S. M., and Abukmeil, S. S. 1998. Brain abnormality in schizophrenia. A systematic and quantitative review of volumetric magnetic resonance imaging studies. Br. J. Psychiatry 172: 110 – 120. Lawrie, S. M., Whalley, H. C., Abukmeil, S. S., Kestelman, J. N., Donnelly, L., Miller, P., Best, J. J. K., Owens, D. G., and Johnstone, E. C. 2001. Brain structure, genetic liability and psychotic symptoms in subjects at high risk of developing schizophrenia. Biol. Psychaitry 49: 811– 823. Maguire, E. A., Gadian, D. G., Johnsrude, I. S., Good, C. D., Ashburner, J., Frackowiak, R. S., and Frith, C. D. 2000. Navigationrelated structural change in the hippocampi of taxi drivers. Proc. Natl. Acad. Sci. USA 97: 4398 – 4403. May, A., Ashburner, J., Buchel, C., McGonigle, D. J., Friston, K. J., Frackowiak, R. S., and Goadsby, P. J. 1999. Correlation between structural and functional changes in brain in an idiopathic headache syndrome. Nat. Med. 5: 836 – 838. Miller, P., Byme, M., Hodges, A., Lawrie, S., Owens, D. G. C., and Johnstone, E. C. 2002. Schizotypal components in people at high risk of developing schizophrenia: Early findings from the Edinburgh High Risk study. Br. J. Psychiatry 180: 179 –184. Nichols, T. E., and Holmes, A. P. 2001. Nonparametric analysis of PET functional neuroimaging experiments: A primer. Hum. Brain Mapping, 15: 1–25. Paillere-Martinot, M., Caclin, A., Artiges, E., Poline, J., Joliot, M., Mallet, L., et al. 2001. Cerebral gray and white matter reductions and clinical correlates in patients with early onset schizophrenia. Schizophrenia Res. 50: 19 –26.
889
Shah, P. J., Ebmeier, K. P., Glabus, M. F., and Goodwin, G. M. 1998. Cortical grey matter reductions associated with treatment-resistant chronic unipolar depression. Controlled magnetic resonance imaging study. Br. J. Psychiatry 172: 527–532. Sigmundsson, T., Suckling, J., Maier, M., Williams, S. C. R., Bullmore, E. T., Greenwood, K. E., Fukuda, R., Ron, M. A., and Toone, B. K. 2001. Structural abnormalities in frontal, temporal, and limbic regions and interconnecting white matter tracts in schizophrenic patients with prominent negative symptoms. Am. J. Psychiatry 158: 234 –243. Sowell, E. R., Thompson, P. M., Holmes, C. J., Batth, R., Jernigan, T. L., and Toga, A. W. 1999. Localizing age-related changes in brain structure between childhood and adolescence using statistical parametric mapping. NeuroImage 9: 587–597. Talairach, J., and Tournoux, P. 1988. A Coplanar Stereotaxic Atlas of a Human Brain. Thieme, Stuttgart. Whalley, H. C., Kestelman, J. N., Rimmington, J. E., Kelso, A., Abukmeil, S. S., Best, J. J., Johnstone, E. C., and Lawrie, S. M. 1999. Methodological issues in volumetric magnetic resonance imaging of the brain in the Edinburgh High Risk Project. Psychiatry Res. 91: 31– 44. Wilkie, M., Kaufmann, C., Grabner, A., Pu¨ tt, B., Wetter, T. C., and Auer, D. P. 2001. Gray matter changes and correlates of disease severity in schizophrenia: A statistical parametric mapping study. NeuroImage 13: 814 – 824. Woermann, F. G., Free, S. L., Koepp, M. J., Ashburner, J., and Duncan, J. S. 1999. Voxel-by-voxel comparison of automatically segmented cerebral gray matter—A rater-independent comparison of structural MRI in patients with epilepsy. NeuroImage 10: 373– 384. Worsley, K. J., Marrett, S., Neelin, P., and Evans, A. C. 1996. Searching scale space for activation in PET images. Hum. Brain Mapping 4: 74 –90. Wright, I. C., McGuire, P. K., Poline, J. B., Travere, J. M., Murray, R. M., Frith, C. D., Frackowiak, R. S., and Friston, K. J. 1995. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. NeuroImage 2: 244 –252. Wright, I. C., Ellison, Z. R., Sharma, T., Friston, K. J., Murray, R. M., and McGuire, P. K. 1999a. Mapping of grey matter changes in schizophrenia. Schizophrenia Res. 35: 1–14. Wright, I. C., Sharma, T., Ellison, Z. R., McGuire, P. K., Friston, K. J., Brammer, M. J., Murray, R. M., and Bullmore, E. T. 1999b. Supra-regional brain systems and the neuropathology of schizophrenia. Cereb. Cortex 9: 366 –378. Wright, I. C., Rabe-Hesketh, S., Woodruff, P. W., David, A. S., Murray, R. M., and Bullmore, E. T. 2000. Meta-analysis of regional brain volumes in schizophrenia. Am. J. Psychiatry 157: 16 –25.