Skip Navigation


Schizophrenia Bulletin Advance Access originally published online on November 7, 2008
Schizophrenia Bulletin 2009 35(1):82-95; doi:10.1093/schbul/sbn150
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
35/1/82    most recent
sbn150v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Segall, J. M.
Right arrow Articles by Calhoun, V. D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Segall, J. M.
Right arrow Articles by Calhoun, V. D.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The Author 2008. Published by Oxford University Press on behalf of the Maryland Psychiatric Research Center. All rights reserved. For permissions, please email: journals.permissions@oxfordjournals.org.

Voxel-based Morphometric Multisite Collaborative Study on Schizophrenia

Judith M. Segall2, Jessica A. Turner3, Theo G.M. van Erp4, Tonya White5, H. Jeremy Bockholt2, Randy L. Gollub6, Beng C. Ho7, Vince Magnotta8, Rex E. Jung2, Robert W. McCarley9, S. Charles Schulz5, John Lauriello10, Vince P. Clark2, James T. Voyvodic11, Michele T. Diaz11 and Vince D. Calhoun1,12,13
2 The Mind Research Network, Albuquerque, NM
3 Department of Psychiatry and Human Behavior, University of California Irvine, Irvine, CA
4 Department of Psychology, University of California Los Angeles, CA
5 Department of Psychiatry, University of Minnesota, Minneapolis, MN
6 Department of Psychiatry, Massachusetts General Hospital, Boston, MA
7 Department of Psychiatry, University of Iowa, Iowa City, IA
8 Department of Radiology, University of Iowa, Iowa City, IA
9 Department of Psychiatry, VAMC & Harvard Medical School, Brockton, MA
10 Department of Psychiatry, University of New Mexico, Albuquerque, NM
11 Brain Imaging and Analysis Center, Duke University, Durham, NC
12 Deptartment of ECE, University of New Mexico, Albuquerque, NM
13 Department of Psychiatry, Yale University, New Haven, CT


   Abstract
Top
 Abstract
Introduction
 Materials and Methods
 Results
 Discussion
 Funding
 References
 
Regional gray matter (GM) abnormalities are well known to exist in patients with chronic schizophrenia. Voxel-based morphometry (VBM) has been previously used on structural magnetic resonance images (MRI) data to characterize these abnormalities. Two multisite schizophrenia studies, the Functional Biomedical Informatics Research Network and the Mind Clinical Imaging Consortium, which include 9 data collection sites, are evaluating the efficacy of pooling structural imaging data across imaging centers. Such a pooling of data could yield the increased statistical power needed to elucidate effects that may not be seen with smaller samples. VBM analyses were performed to evaluate the consistency of patient versus control gray matter concentration (GMC) differences across the study sites, as well as the effects of combining multisite data. Integration of data from both studies yielded a large sample of 503 subjects, including 266 controls and 237 patients diagnosed with schizophrenia, schizoaffective or schizophreniform disorder. The data were analyzed using the combined sample, as well as analyzing each of the 2 multisite studies separately. A consistent pattern of reduced relative GMC in schizophrenia patients compared with controls was found across all study sites. Imaging center-specific effects were evaluated using a region of interest analysis. Overall, the findings support the use of VBM in combined multisite studies. This analysis of schizophrenics and controls from around the United States provides continued supporting evidence for GM deficits in the temporal lobes, anterior cingulate, and frontal regions in patients with schizophrenia spectrum disorders.

Keywords: schizophrenia / VBM / gray matter concentration / multisite / multicenter / multiscanner


   Introduction
Top
 Abstract
 Introduction
Materials and Methods
 Results
 Discussion
 Funding
 References
 
Volumetric differences in gray matter (GM) and specific subcortical regions have been well characterized in patients with schizophrenia compared with age-matched controls.1 For example, a meta-analysis done by Wright et al 2 of 58 volumetric schizophrenia studies found that schizophrenia patients on average had a 2% reduction of global cerebral volume compared with matched controls. Specific subcortical structures, such as the amygdala and hippocampus, also showed reductions in patients with schizophrenia.3 One method that has been used to compare GM in schizophrenia patients to healthy controls is voxel-based morphometry (VBM).4–6

VBM is a fully automated alternative to volumetric studies that use manual delineation of regions of interest (ROIs) to detect differences between patients and controls.7 More traditional morphometric approaches to measuring brain volumes, such as manual delineation are time intensive, specific to particular brain regions, and can be prone to rater bias and error. While VBM also has some inherent biases, it is a consistent and relatively fast way to measure brain morphometry. VBM was developed to detect group differences in the relative concentration of GM tissue across the whole brain in a voxel-wise manner. In VBM, relative concentration, sometimes referred to as density, is the proportion of the type of tissue, such as GM, to the proportion of all other tissues within a region.8 Essentially, VBM is a test for residual tissue concentration differences that remain after MRI scans are spatially normalized into standardized stereotactic space.9 Performing a meta-analysis of 15 published VBM articles on schizophrenia, Honea et al1 found consistent differences between relative GM deficits in schizophrenia patients compared with matched healthy cohorts. In particular, ROIs that were most often reported to have reduced GM in patients with schizophrenia were in the left superior temporal gyrus and the left medial temporal lobe.1 Significant differences in other ROIs have been found less consistently in schizophrenia. This may be due to the different sizes of smoothing kernels, the heterogeneity of schizophrenia, and/or the inability for these small patient sample sizes to provide adequate power to detect group differences.1,10

Most imaging studies of neurological and psychiatric illness have compared relatively small numbers of subjects, particularly in disorders with difficult-to-recruit patients and in disorders with a low prevalence rate. Two multisite schizophrenia imaging studies, the Functional Biomedical Informatics Research Network (fBIRN) and the Mind Clinical Imaging Consortium (MCIC), are evaluating the potential of pooling data across sites in order to increase the statistical power needed to detect subtle, but meaningful anatomical differences that may not be seen with smaller samples. For instance, by using a larger number of subjects, regions that show smaller GM volume differences can be observed. Also, differences between patient subgroups, based on neuropsychological profile and/or clinical features (eg, severity of positive or negative symptoms), as well as the effects of gender or other demographic variable of interest usually require a larger data set to observe statistically significant differences. A recent VBM study performed at the Clinical Brain Disorders Branch/National Institute of Mental Health Genetic Study of Schizophrenia had a subject size of 169 patients with schizophrenia, 213 unaffected siblings, and 212 healthy volunteers.11 Such large scale data collection is not feasible in most single-center studies, due to the difficulties in recruiting adequate numbers of patients and the increased level of infrastructure, personnel, and analysis support required. For the purpose of this article, multisite refers to multiple imaging centers/universities with scans performed on different magnetic resonance scanners. Due to the fact that some of the same imaging centers/universities were used in both of the fBIRN and MCIC studies, "site" in this article refers to scanning protocol, which includes the location (ie, university) and pulse-sequence. The ability to use multisite data provides a method for obtaining a larger data set than would otherwise be possible.

A primary purpose of the fBIRN consortium is to evaluate the possibility of merging neuroimaging data from multiple collections, centers, and databases with a specific application to schizophrenia. fBIRN aims to develop standardized imaging protocols, as well as tools for processing multicenter functional MRI (fMRI) and structural MRI (sMRI) data collections.12,13 The morphometry test bed of BIRN has assessed the prospect of combining structural legacy data from 3 BIRN sites, but with specific applications to Alzheimer's disease (AD).14 The fBIRN project collects multicenter fMRI data to study regional brain dysfunction related to schizophrenia.15–18 Similarly, the MCIC study was established to better elucidate the underlying pathophysiological mechanisms of schizophrenia though fMRI, sMRI, diffusion-tensor imaging (DTI), and a large number of clinical and neuropsychological measures.17,19–21

In this article we present results from a morphometric analysis using VBM of the sMRI scans that were acquired from 9 of the fBIRN collaboration imaging centers, with additional data obtained from 4 of these imaging centers that were also involved in the MCIC study. This yielded a total of 503 unique subjects who were scanned on Siemens and GE scanners at 1.5T, 3T, or 4T field strengths and 13 sites. A primary objective was to evaluate whether VBM produces consistent results, with the increased power from a large sample, to detect the subtle group differences reported in schizophrenia studies, when applied to multisite structural imaging data, in which there are differences as to how the structural scans were acquired. We hypothesized that this would be possible because a recent single-site multiscanner VBM study on AD reported that the neurodegeneratiave changes in brain anatomy associated with AD were substantially larger than the effect introduced by the variability in scanner hardware and software changes that occurred at their center.22 The secondary objective was to evaluate the consistency of gray matter concentration (GMC) changes across site in ROIs that have been reported in prior VBM studies on schizophrenia. The third and final objective was to use this large data set to explore the effect of gender, which is potentially associated with GM abnormalities in schizophrenia, but is difficult to evaluate with small patient samples.

This study reports the results from analyzing both the fBIRN and MCIC multisite structural imaging data sets with VBM. In the following sections, we first describe the multisite data collection and show several VBM results of cohorts consisting of fBIRN-only, MCIC-only, and a combined fBIRN+MCIC data set. We close with a discussion of the implications of the resulting findings for both schizophrenia studies and multisite VBM.


   Materials and Methods
Top
 Abstract
 Introduction
 Materials and Methods
Results
 Discussion
 Funding
 References
 
Subjects
This study combines data from 2 multisite collaborative projects, fBIRN and MCIC, which comprise 13 separate patient and control samples from the University of Minnesota (MINN), the Mind Research Network (MRN), Harvard's Massachusetts General Hospital (MGH) and Brigham and Women's Hospital (BWH), the University of California-Irvine (UCI), the University of California-Los Angeles (UCLA), the University of Iowa (IOWA), Duke University, and Yale University. Four of the sites (MRN, MGH, MINN, and IOWA) participated in both studies and some of the subjects at these sites took part in both the FBIRN and MICIC studies, for details see Results. Given that the scans performed are not necessarily representative of the optimized structural imaging protocols used at each site, and to prevent this report from being used to assess overall imaging data quality at the individual data collection sites, this article lists the data obtained from these different sites by site code rather than by name. The total number of subjects in the fBIRN data set includes 121 schizophrenia patients (including schizophrenia and schizoaffective disorder) and 123 healthy controls. The total number of subjects in the MCIC data set includes 163 healthy controls (63 females, mean age = 32.2 SD = 11.3, range 19–60; 100 males, mean age = 31.3, SD = 11.1, range 18–58) and 149 patients with chronic or first-episode (FE) schizophrenia (38 females, mean age = 34.6, SD = 12.0, range 18–60; 111 males, mean age = 32.6, SD = 10.5, range 18–59). In the MCIC data set, there is not a consistent number of FE patients at all 4 sites; therefore, some sites have more FE patients than others. Subjects with schizoaffective disorder were also included in the cohorts as studies have shown that those with schizoaffective disorder have functional outcome similar to schizophrenia, and there is a significant overlap between the 2 diagnoses. The combined fBIRN and MCIC data set consists of 237 unique patients with schizophrenia and 266 unique healthy controls as outlined in table 1. Thus, the subjects used for this article include the fBIRN set (fBIRN), the MCIC set (MCIC), and a combined MCIC and fBIRN set (fBIRN+MCIC), which excludes MCIC scans from subjects already represented in the fBIRN cohort.


View this table:
[in this window]
[in a new window]

  		Table 1. Demographics for fBIRN +MCIC and MCIC– only studies

 
Subjects in the fBIRN study were screened using the structured clinical interview for DSM-IV-TR (SCID-I, November 2002 Patient and Non-Patient version). Healthy controls were free from any Axis-1 disorder, had no history of drug dependence, no major untreated illness, no first-degree relative with history of psychotic illness, and had an IQ above 75. Only chronic patients that met the DSM-IV-TR criteria for schizophrenia or schizoaffective disorder were included in the study, this excluded schizophreniform subjects. Patients were also excluded if they were not clinically stable for 2 months prior to enrollment, had alcohol or substance dependence 2 months prior, or had an IQ less than 75. Demographic and other socioeconomic information was collected at interview. Other measures that were included for all subjects were the Edinburgh Handedness Inventory,23 Fagerstrom Test for Nicotine Dependence,24,25 and the North American Adult Reading Test (NAART).26–28 In addition, all patients were administered Scales for the Assessment of Positive (SAPS)29 and Negative Symptoms (SANS)30 as well as other measures of clinical symptoms.

The MCIC study participants were screened and selected in a manner similar to the participants in the fBIRN study. The healthy controls were screened using the SCID, and subjects were excluded who were diagnosed with substance abuse/dependence, medical, psychiatric, or neurological illnesses. Healthy controls were not excluded if they had been medicated with antidepressants, antianxiety, or sleep deprivation medications, so long as these medications had not been taken for at least 6 months prior to the scan and for not more than 2 months of continuous use at any time. The MCIC patient group comprised subjects that met DSM-IV-TR criteria for schizophrenia, schizophreniform disorder, or schizoaffective disorder. The diagnoses were based on DSM-IV criteria using the SCID.31 Patients were excluded if they had a history of neurologic or psychiatric disease other than schizophrenia, head injuries, lifetime history of substance dependence or abuse within the past month, or an IQ less than or equal to 70. Sociodemographics, medical history, cognitive assessments, and symptom data were also collected. Patients were recruited from the public sector/university and outpatient clinics that were affiliated with the site. Some FE patients may have been identified while still inpatient and joined the study following discharge. The FE patients could have received other psychotropics if they were given as part of a prepsychotic diagnosis.

Control subjects for both studies were recruited via flyers, newspaper ads, and word-of-mouth. Every effort was made to match controls to patients based on gender, age, handedness, ethnicity, and parental socioeconomic status; however, overall controls were often more educated than patients. For both studies written informed consent was obtained as required by the Institutional Review Board of record for all sites. Additional demographic data for IQ, handedness, and SAPS/SANS scores were located in table 2a,b. Although, we did not use these additional measures as cofactors it maybe helpful to the readers to see the differences in patients and controls across sites. For the fBIRN data set, the IQ measure is the FSIQ derived from the NAART and from the MCIC data set they are raw IQ scores from the WAIS Reading Total. We did not use IQ as a covariate because of this acquisition difference.


View this table:
[in this window]
[in a new window]

  		Table 2a. Additional demographics for fBIRN and MCIC studies

 

View this table:
[in this window]
[in a new window]

  		Table 2b. Additional demographics for fBIRN and MCIC studies

 
MR Imaging Parameters
The sites differed with respect to the scanner manufacturer (General Electric, Philips, and Siemens), field strength (5 sites at 1.5 T, 7 sites at 3.0 T, and 1 site at 4.0 T) and implemented pulse sequences. A T1-weighted sMRI was acquired at each site (table 3). For the fBIRN study, slice thickness was 1.2 mm and the slice orientation was sagittal, except for 3 subjects from site FB05 whose slice orientation was coronal. The scanning parameters were for the most part consistent for the entire study; the few deviations can be seen in table 2. The parameters for the MCIC study were the same for the 3 sites that used a 1.5 scanner. The remaining MCIC site used a 3T scanner and the parameters were comparable to those that were also scanned on a 3T in the fBIRN study. All scans in the MCIC study were collected in a coronal orientation.


View this table:
[in this window]
[in a new window]

  		Table 3. Scanners pulse sequences, slice, location fBIRN, and MCIC

 
Data Analysis
Data were preprocessed using the SPM5 software package (http://www.fil.ion.ucl.ac.uk/spm/software/spm5/), MATLAB version 7.2. The voxel size for all images was resliced to 2 x 2 x 2 mm. SPM5 was used to segment the brain into white matter (WM), GM, and cerebral spinal fluid with unmodulated normalized parameters. Unmodulated parameters test for voxel-wise differences in GMC, whereas modulated parameters test for voxel-wise differences in the relative volume of GM.32 Spatial normalization requires warping, an expansion and contraction of some brain regions. Modulation can be used to compensate for the effects of this warping to ensure that the total amount of GM in a region is the same before and after spatial normalization; therefore, yielding relative volume for each voxel. A recent study by Meda et al10 compared modulated and unmodulated normalization methods on a data set primarily collected at one site and showed the same pattern of less GM for schizophrenia patients in the temporal, frontal, and subcortical/limbic regions, for both methods. However, when using modulated results, there was a much larger intersubject variance for patients relative to controls.10 The current version of SPM5 uses a new VBM method for segmenting brain images called unified segmentation.33 Previous brain image segmentation typically used 2 procedures in series including a tissue classification approach and registration to a template. In order to prevent misinterpretation of volumetric differences caused by normalization and not by structural differences, a newer technique was developed, known as optimized VBM. Optimized VBM spatially normalizes GM and WM-segmented images to a standard space by matching images, respectively, to their GM and WM templates.32 These optimized images are then used for a secondary segmentation. Ashburner and Friston33 find this method to be circular because the registration requires an initial tissue classification, and the tissue classification requires an initial registration. Unified segmentation33 deploys a framework where tissue classification, bias correction, and image registration are integrated within the same model. It also produces an estimate of the tissue class intensities from a fitting of spatial priors to the image, which allows for differences in voxel intensities. After segmentation, the GM images were smoothed to a full-width half maximum (FWHM) Gaussian kernel of 10 mm. In general, the smoothing kernel should be comparable to the size of the expected regional differences between the groups of brains. The importance of the smoothing kernel has been shown empirically in the case of DTI data34 and also for sMRI data.1 It has been suggested that for sMRI with a balanced sample set of our size, a kernel of FWHM of 10 mm is reasonable.35,36

Subject outlier detection was performed using a Pearson correlation, which compared the degree to which subjects are related to the average smoothed GM map. We performed outlier detection to ensure that all subjects segmented properly, because the segmentation and normalization steps were processed through automation scripts and not manually. This correlation was done within sites, as well as overall across sites. Subjects were considered outliers when they deviated more than 2% from the mean GMC. Some sites had a greater number of outliers than others, but this was mainly due to orientation and not the scans themselves. When outliers were identified, a visual check was performed and the origin was manually reoriented to the anterior commissure-posterior commissure line. The subject was then rerun through the preprocessing, that is, unified segmentation and smoothing. If after being reprocessed and there were no other obvious problems with the scan, the subject remained in the final data set to avoid the subjectivity associated with removing outliers.

Statistical Analysis
Voxel-wise ANCOVAs were performed to compare relative GMC group differences between controls, and patients were performed for the MCIC, fBIRN, and fBIRN+MCIC data sets. For all 3 data sets group, site, and gender were entered into the model as between-subject measures, and age and total intracranial volume (TIV) were entered as covariates. TIV used in VBM studies to control for individual differences in head size. A previous multisite VBM study also used TIV as a covariate and we followed this approach.22 Group differences were examined for both control > patients and patients > controls. Additional analyses of gender effects were performed only on the fBIRN+MCIC data set given that this data set has most power to detect differences. Statistical results for the fBIRN, MCIC, and fBIRN+MCIC group comparisons were thresholded at P < .05, correcting for multiple comparisons using the family-wise error rate (FWE).37 For all the voxel-wise analyses in this article FWE with a P < .05 was used. In addition, because the degrees of freedom (df) were different for the 3 analyses, we also performed a comparison of the pattern of differences at each site by converting the T-maps to Z-scores. Fifty-three of the participants were enrolled in both the MCIC and fBIRN studies, and for these participants only the fBIRN data were included in the final fBIRN+MCIC data set. Because the MCIC set had a larger subject pool than the fBIRN set, we consequently decided to use the fBIRN subjects when there was a group overlap so that the combined fBIRN+MCIC data set would be a more even distribution of the fBIRN and MCIC subjects. An additional comparison was performed on 29 pairs who were scanned at the same site, under both the fBIRN and MCIC protocols (14 controls and 15 patients), to investigate possible interactions between pulse-sequence and group. A paired t test was also done on the 14 controls.

Gender differences were explored using several different contrasts, which were male > female, female > male, patient male > control male, control male > patient male, control female > patient female and patient female > control female. The gender analysis was performed on the combined fBIRN+MCIC data set and TIV, site, and age were controlled for. In addition, contrasts were generated to examine possible interactions between group (patient and control) and gender. Finally, a voxel-wise F test was also performed to detect the effects of site difference with groups using the same design matrix as the ANCOVA of the overall fBIRN+MCIC data set.

ROI Analyses
ROIs were used to compare differences in GMC across groups and sites in the fBIRN+MCIC data set. ROI selection was based on findings from prior VBM studies on schizophrenia. Honea et al1 reported that more than 50% of the 15 studies they reviewed showed significantly lower GM volume in patients compared with controls in the left superior temporal gyrus and the left medial temporal lobe. In 50% of the studies, they also noted GM volumetric decreases in the right superior temporal gyrus, left inferior frontal gyrus, left medial frontal gyrus, and the left parahippocampal gyrus. The fusiform gyrus was selected as a pseudo-control area, not expected to show a volume difference in schizophrenics, because it was only found to have significant volume deficits in less than 30% of the 15 VBM studies evaluated by Honea et al1. Further numerous studies have shown GM deficits in the anterior cingulate in patients with schizophrenia compared with controls.3,38,39 Based on these findings, we used WFU PickAtlas40 labeled masks to define the ROIs and investigate group and site differences in the combined fBIRN+MCIC data set for the average relative GMCs in bilateral superior temporal, inferior frontal, fusiform, and anterior cingulate gyri.


   Results
Top
 Abstract
 Introduction
 Materials and Methods
 Results
Discussion
 Funding
 References
 
Effects of Diagnosis
For all 3 data sets, we evaluated relative voxel-wise GMC differences between diagnostic groups (patients vs. controls). We used a FWE correction for the fBIRN, MCIC, and fBIRN+MCIC data sets. In figure 1d are axial images of z-score normalized maps for all 3 data sets. Regions that are circled in white show the similarities between the 3 sets, which are ROIs that had reduced GMC. Figures 1a–c shows surface renderings of the significant differences in each of the 3 data sets. These images show that, although each subject data set contains different multisite subjects, the overall pattern of GMC reduction is similar between patients and controls. The significant regions showing group GM differences for fBIRN, (P < .05 FWE corrected, T = 4.66, 231 df), for MCIC (P < .05 FWE corrected, T = 4.89, 307 df), and fBIRN+MCIC, (P < .05 FWE corrected, T = 4.77, 486 df) data sets are reported in table 4. Group difference T (figure 1) maps are overlaid on the average-rendered MNI template for the fBIRN, MCIC, and FBIRN+MCIC data sets. We did not find any significant regions where schizophrenia patients had more GMC than controls.


Figure 1
View larger version (51K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 1. Rendered images of thresholded T-maps FWE, P < .05 for FBIRN (a), MCIC (b), fBIRN+MCIC (c) overlaid on MNI average template, which are regions that have lower GMC for Schizophrenics. (d) Z-score normalized maps for all 3 data sets and regions that are circled in white show the similarities in reduced GMC for all data sets.

 

View this table:
[in this window]
[in a new window]

  		Table 4. Talairach coordinates for significant regions in fBIRN, MCIC, and fBIRN+MCIC; n.s. (nonsignificant)

 
Although, there were differences in the relative GMC, the comparison of the 29 pairs, from one site, that were in both fBIRN and MCIC studies showed that there was no interaction between pulse-sequence and group. Of the 4 sites there is overlap between 53 subjects (20 healthy controls and 33 schizophrenia patients) from the MCIC and fBIRN studies. The paired t test on the 14 overlap controls showed no significant differences between the 2 different studies.

Gender Results
Gender differences were found; however, there were no significant interactions between gender and patient group. In the fBIRN+MCIC set, males, regardless of group, tended to have less GM than females. Figure 2 is a T-map of the regions where men had reduced GMC. The contrast used was males < females (P < .05 FWE corrected, t = 4.66, 484 df). Table 5 lists the regions in which there is the greatest GMC reduction difference between females and males.


Figure 2
View larger version (47K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 2. Rendered T-map of regions where males had reduced GMC, P < .05. These are regions where males had less GMC in comparison to females.

 

View this table:
[in this window]
[in a new window]

  		Table 5. Talairach coordinates for significant regions (P < .05) in gender analysis; n.s. (nonsignificant)

 
ROI Results
GMC in the anterior cingulate, inferior frontal, superior temporal, and the fusiform gyri all showed a significant effect of diagnosis. Figure 3 shows patient and control differences for all sites averaged together and for all 4 ROIs patients consistently have reduced GMC. The 4 graphs in figure 4 show patient and control differences in the selected ROIs for each site. Although the GM values are different across the sites, the overall trend of patients having reduced GMC is the same for all the ROIs mentioned across all sites. These results are consistent with the meta-analytic findings.1 The standard error (SE) was also calculated and can be seen above each bar. Compared with controls, patients showed a 5% reduction in mean GMC in the superior temporal gyrus [0.48, SE = 0.01 patients, (SP); 0.51, SE = 0.01 controls, (HC)] and inferior frontal gyrus [0.44, SE = 0.01 SP; 0.46, SE = 0.01 HC), a 4% reduction in mean GMC for the anterior cingulate gyrus [0.47, SE = 0.01 SP; 0.49, SE = 0.09 HC), and 3.5% mean reduction in GMC in the fusiform gyrus (0.59, SE = 0.01 SP; 0.61, SE = 0.01 HC).


Figure 3
View larger version (20K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 3. Mean GMC differences by diagnostic group for anterior cingulate cortex, superior temporal gyrus, inferior frontal gyrus, and fusiform gyrus. SE is above each bar for each ROI. In all 4 ROIs, which all sites have been combined, patients with schizophrenia have reduced GMC.

 

Figure 4
View larger version (42K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		Fig. 4. Mean GMC differences by diagnostic group and site, for anterior cingulate cortex, superior temporal gyrus, inferior frontal gyrus, and fusiform gyrus. SE is above each bar for each site. In this figure, there are site differences between controls and patients; however, even though it is not as significant in some sites, the overall trend of reduced GMC in patients in the same.

 

   Discussion
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
Funding
 References
 
The principle finding of this study is that in a whole brain voxel-based analysis, 3 multisite data sets display similar regions where relative GMC was significantly decreased in schizophrenia patients compared with demographically matched healthy controls. There are differences in the degree to which patients and controls differ, but the overall patterns were strikingly similar. Prior multisite and multiscanner studies have investigated the degree to which the scanner affects imaging results. Several of these studies emphasize that the most important confound to explore is that of the scanner/site.22,42,43 To our knowledge, this is the largest multisite imaging study that examines the impact of imaging center–associated biases in a VBM analysis. Before group differences between patients and controls are discussed in a meaningful way, it is important to establish that there are no significant interactions between scanners and the effect of group differences.

In a recent multiscanner study of AD, the effects of site (scanner) and effect of interest (disease status) were evaluated.22 Although that study reported site (scanner) differences, the overall group differences were more significant than the effect of interactions between the scanners. The Stonnington et al22 study included 136 subjects scanned with similar parameters, except for minor variations in the TR/TE and flip angle, on six 1.5T GE scanners with software differences due to upgrades. In our multisite study, there were more extreme scanner and parameter variations by design. Thus, we might expect site biases in the results. However, even if steps are taken to control for variability, such as phantoms and reliability testing, scanner effects cannot be fully removed.42 A recent multisite VBM study on childhood absence epilepsy by Pardoe et al44 found that explicitly modeling for site as a factor in statistical analysis provides adequate control over intersite variability, which in turn allows the combination of multisite/multiscanner data for VBM.

The effectiveness of multisite studies is limited by "image intensity variability, which can translate into accuracy errors for derived morphometric data used to characterize brain structures"42. This intensity variability in image intensities, which is not limited to multiscanner studies, is caused by a range of factors from partial volume effects, image gradient nonlinearities, noise from the electronics of the MRI system, as well as the subject's physiology, and inhomogeneities in the magnetic field. There is likewise similar image intensity variability in this study, which has a greater effect in multisite studies; however, given the large N of our study, the statistical magnitude should make our results more compelling.

Testing scanner reliability,42 found that the same subject's positioning from one scanner to another was inconsistent, which could be a problem for multicenter studies, cross-sectional studies, and longitudinal studies. In our study, subjects scanned at both fBIRN and MCIC sites had site-related variation in GM values; however, the patients still showed reduced GMC compared with controls. The only differences between the repeat scans were the time interval between scans and minor differences in the sequence parameters. There was no significant interaction between the scanner and the effect of group. This demonstrates that although there can be differences across scanners and scanning methods, the differences between patient and control groups were greater.

In our fBIRN+MCIC data set, we observed few effects of site, and the difference between patients and controls appeared to be less sensitive to site differences. However, we did find that there were significant site differences. The individual site differences could be due to the disparate number of subjects for each site, differences in age matching, and patient differences in duration of illness. But the group differences at each site were consistent with one another, in that patients consistently had less GM than controls. This is shown in figure 3 where the relative GMC is shown for specific ROIs at each site in the overall fBIRN+MCIC study. Beyond the standard matching criteria that are an issue for all studies, design recommendations for a multisite study would ideally include a similar number of patients and controls at each site, consistent magnet strength, and a consistent pulse-sequence.

The fBIRN, MCIC, and fBIRN+MCIC sample sets showed similar GM reductions in the patients groups in the ROI analyses. This was most pronounced in the inferior frontal gyrus, insula, parahippocampal gyrus, superior temporal gyrus, and rectal gyrus. These regions have also been reported in other VBM studies.1 In figure 1, it appears that the MCIC set has more significant differences; however, this could be due to the larger MCIC sample size as compared with fBIRN. The individual differences site differences that are seen in the ROI analysis is possibly due to the disparate number of subjects from each site. Another possible cause for site differences is that the MCIC study recruited FE patients and within that study some sites had more FE patients than others. This difference between duration of illness may have also contributed to ROI differences for each site.

For the gender differences, males had reduced GMC compared with females, regardless of group. Previous gender studies have looked at sexual dimorphism in GM in a variety of different cohorts.45,46 Chen et al47, in a VBM study of sex differences in regional GM of middle-aged healthy individuals found that sexual dimorphism appears to exist independent of aging and age-related pathology.47 However, in a study of healthy elderly, Lemaitre et al. found that brain atrophy in the seventh and eighth decades of life is ubiquitous and is not modulated by sex.41,48

One criticism of VBM is that when comparing 2 groups, some regions that are found to be significant may not necessarily be the most important for characterizing group differences.49 Our aim of a multisite study is to increase the statistical possibility of finding more subtle group differences through a large sample set. This large subject set may then reflect a more accurate representation of group differences. An important issue to remember when discussing the validity of VBM is that it has been described as, "a research tool that enables people to ask specific questions about their data and is not a diagnostic or classification device.50" VBM may in turn be a method that aids in the appropriate direction to take when comparing groups. This information obtained from VBM may be a stepping stone for future GM schizophrenia research.

Future extensions of this study could include parsing out the clinical heterogeneity of schizophrenia (ie, positive and negative symptoms, medication effects, duration of illness) to assess regional differences within more homogeneous clinical domains. Investigating FE versus chronic patients could also clarify how to combine patient data.

To our knowledge, this is the largest multisite structural data set in schizophrenia to date. In combining these data, we find that the previously identified areas of volumetric differences in schizophrenia were identified—for example, the superior, inferior, and medial temporal lobe, thalamus, hippocampal regions and anterior cingulate—and also that specific areas of the frontal and parietal lobe were implicated as well. The value of a large, multisite data set resides in the power to find the more subtle volumetric differences in a representative sample. We have also determined that it is feasible to combine data from different sites (and potentially legacy) data for a VBM analysis. Even though there are differences between sites, scanners, and demographics, patients diagnosed with schizophrenia showed consistently reduced GM in comparison to controls across all 13 sites.


   Funding
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 Funding
References
 
NIH U24-RR021992 to the Functional Biomedical Informatics Research Network (FBIRN, http://www.fbirn.org), and NIH U24-RR021382 to the Morphometry BIRN, by the National Center for Research Resources; and DOE Grant No. DE-FG02-99ER62764.


  Footnotes
 
1 To whom correspondence should be addressed; The Mind Research Network, 1101 Yale Boulvard NE, Albuquerque, NM 87131; tel: 505-272-1817, fax: 505-272-8002 e-mail: vcalhoun{at}unm.edu.


   Acknowledgments
 
Segall and Calhoun performed the data analyses and interpretation and wrote the manuscript; Turner, van Erp, Bockholt, Gollub, and Ho helped design and implement the experiment and provided editorial comments on the analyses and interpretation; the remaining authors designed and implemented the experiment, collected the data, and facilitated the imaging and behavioral data sharing to make these analyses possible. Also, thanks to Adam Scott for demographics help.


   References
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 Funding
 References
 

  1. Honea R, Crow TJ, Passingham D, Mackay CE. Regional deficits in brain volume in schizophrenia: a meta-analysis of voxel-based morphometry studies. Am J Psychiatry (2005) 162::2233–2245.[Abstract/Free Full Text]

  2. Wright IC, Rabe-Hesketh S, Woodruff PW, David AS, Murray RM, Bullmore ET. Meta-analysis of regional brain volumes in schizophrenia. Am J Psychiatry (2000) 157:(1):16–25.[Abstract/Free Full Text]

  3. Shenton ME, Dickey CC, Frumin M, McCarley RW. A review of MRI findings in schizophrenia. Schizophr Res. (2001) 49:(1–2):1–52. Apr 15.[Web of Science][Medline]

  4. Kubicki M, Shenton ME, Salisbury DF, et al. Voxel-based morphometric analysis of gray matter in first episode schizophrenia. Neuroimage (2002) 17::1711–1719.[CrossRef][Web of Science][Medline]

  5. Wilke M, Kaufmann C, Grabner A, Putz B, Wetter TC, Auer DP. Gray matter-changes and correlates of disease severity in schizophrenia: a statistical parametric mapping study. Neuroimage (2001) 13::814–824.[Web of Science][Medline]

  6. Shapleske J, Rossell S, Chitnis X, et al. A computational morphometric MRI study of schizophrenia: effects of hallucinations. Cereb Cortex (2002) 12::1331–1341.[Abstract/Free Full Text]

  7. Giuliani N, Calhoun VD, Pearlson GD, Francis A, Buchanan RW. Voxel-based morphometry versus regions of interest: a comparison of two methods for analyzing gray matter disturbances in schizophrenia. Schizophr Res. (2005) 74:(2–3):135–147.[CrossRef][Web of Science][Medline]

  8. Mechelli A, Price CJ, Friston KJ, Ashburner J. Voxel-based morphometry of the human brain: methods and applications. Curr Med Imaging Rev. (2005) 1:(2):105–113.[CrossRef]

  9. Ashburner J, Friston KJ. Voxel-based morphometry–the methods. NeuroImage (2000) 11::805–821.[CrossRef][Web of Science][Medline]

  10. Meda SA, Giuliani NR, Calhoun VD, et al. A large scale (N=400) investigation of gray matter differences in schizophrenia using optimized voxel-BASED morphometry. Schizophr Res. (2008) 101:(1–3):95–105.[CrossRef][Web of Science][Medline]

  11. Honea R, Meyer-Lindenberg A, Hobbs K, et al. Is gray matter volume an intermediate phenotype for schizophrenia? A voxel-based morphometry study of patients with schizophrenia and their healthy siblings. Biol Psychiatry (2008) 63::465–474.[CrossRef][Web of Science][Medline]

  12. Van Erp T, Segall JM, Turner J, et al. Voxel-based morphometric analysis of a multi-site study on schizophrenia. In: Proc Soc Neurosci (2007) San Diego, CA.

  13. BIRN. Biomedical Informatics Research Network. Available at. http://www.nbirn.net/.

  14. Fennema-Notestine C, Gamst AC, Quinn BT, et al. Feasibility of multi-site clinical structural neuroimaging studies of aging using legacy data. Neuroinformatics (2007) 5:(4):235–245.[Medline]

  15. Friedman L, Glover G, Krenz D, Magnotta V, The FIRST BIRN. Reducing inter-scanner variability of activation in a multicenter fMRI study: role of smoothness equalization. NeuroImage (2006) 32::1656–1668.[CrossRef][Web of Science][Medline]

  16. Friedman L, Glover G, The FIRST BIRN Consortium. Reducing interscanner variability of activation in a multicenter fMRI study: controlling for signal-to-fluctuation-noise-ratio (SFNR) differences. NeuroImage (2006) 33::471–481.[Medline]

  17. Benavidez M, Clark VP, Kuperberg G, Lim K, Calhoun VD. Functional networks identified in an auditory oddball task of chronic and first episode schizophrenia patients (N=261) collected from the MIND clinical imaging consortium. In: Proc Soc Neurosci (2007) San Diego, CA.

  18. Turner J, Potkin S, Brown G, et al. Multi-center fMRI methods and design: function BIRN. International Congress of Schizophrenia Research. (2005) Savannah, GA.

  19. Demirci O, Clark V, Calhoun V. A projection pursuit algorithm to classify individuals using fMRI data: application to schizoprenia. NeuroImage (2008) 39::1774–1782.[Medline]

  20. Gollub RL, Manoach DS, Fries J, et al. Multi-site fMRI study of DLPFC activation differences between subjects with schizophrenia and controls. In: Proc Soc Neurosci (2007) San Diego, CA.

  21. Machado G, Clark VP, Gollub RL, Magnotta V, White T, Calhoun VD. Probing schizophrenia with a sensorimotor task: large-scale (N=273) independent component analysis of first episode and chronic schizophrenia patients. In: Proc Soc Neurosci (2007) San Diego, CA.

  22. Stonnington CM, Tan G, Kloppel S, Chu C, et al. Interpreting scan data acquired from multiple scanners: a study with Alzheimer's disease. NeuroImage (2008) 39::1180–1185.[Medline]

  23. Oldfield R. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia (1971) 9::97–113.[CrossRef][Web of Science][Medline]

  24. Pomerleau CS, Majchrzak MJ, Pomerleau OF. Nicotine dependence and the Fagerstrom Tolerance Questionnaire: a brief review. J Subst Abuse (1989) 1::471–477.[Medline]

  25. Steinberg ML, Williams JM, Steinberg HR, Krejci JA, Ziedonis DM. Applicability of the Fagerstrom test for nicotine dependence in smokers with schizophrenia. Addic Behav. (2005) 30::49–59.[CrossRef]

  26. Nelson H. National Adult Reading Test (NART): Test Manual (1982) Windsor, UK: NFER Nelson.

  27. Nelson HE, O'Connell A. Dementia: the estimation of premorbid intelligence levels using the new adult reading test. Cortex (1978) 142::234–244.

  28. Blair JR, Spreen O. Predicting premorbid IQ: a revision of the National Adult Reading Test. Clin Neuropsychol. (1989) 3::129–136.[Medline]

  29. Andreasen N. The Scale for the Assesment of Positive Symptoms (SAPS) (1984) Iowa City, IA: University of Iowa.

  30. Andreasen N. The Scale for the Assessment of Negatve Symptoms (1983) Iowa City, IA: University of Iowa.

  31. First M, Spitzer R, Gibbbon M, Williams J. Structured Clinical Interview for DSM-IV-TR Axis I Disorders (1997) Washington, D.C: American Psychiatric Press, Inc.

  32. Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, Frackowiak RS. A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage (2001) 14:(1 Pt 1):21–36.[CrossRef][Web of Science][Medline]

  33. Ashburner J, Friston KJ. Unified segmentation. NeuroImage (2005) 26::839–851.[CrossRef][Web of Science][Medline]

  34. Jones DK, Symms MR, Cercignani M, Howard RJ. The effect of filter size on VBM analyses of DT-MRI data. Neuroimage (2005) 26::546–554.[CrossRef][Web of Science][Medline]

  35. Salmond CH, Ashburner J, Vargha-Khadem F, Connelly A, Gadian DG, Friston KJ. Distributional assumptions in voxel-based morphometry. NeuroImage (2002) 17::1027–1030.[CrossRef][Web of Science][Medline]

  36. White T, O'Leary D, Magnotta V, Arndt S, Flaum M, Andreasen N. Anatomic and functional variability: the effects of filter size in group fMRI data analysis. Neuroimage (2001) 13::577–588.[Web of Science][Medline]

  37. Nichols T, Hayasaka S. Controlling the familywise error rate in functional neuroimaging: a comparative review. Stat Methods Med Res. (2003) 12::419–446.[Abstract/Free Full Text]

  38. Fornito A, Yücel M, Wood S, et al. Surface-based morphometry of the anterior cingulate cortex in first episode schizophrenia. Hum Brain Mapp. (2008) 29::478–489.[CrossRef][Medline]

  39. Ellison-Wright I, Glahn D, Laird A, Thelen S, Bullmore E. The anatomy of first-episode and chronic schizophrenia: an anatomical likelihood estimation meta-analysis. Am J Psychiatry (2008) 165:(8):1015–1023.[Abstract/Free Full Text]

  40. WFU Pickatlas [computer program]. Version. 2003;.

  41. Lemaitre H, Crivello F, Grassiot B, Alperovitch A, Tzourio C, Mazoyer B. Age- and sex-related effects on the neuroanatomy of healthy elderly. NeuroImage (2005) 26::900–911.[CrossRef][Web of Science][Medline]

  42. Jovicich J, Czanner S, Greve D, et al. Reliability in multi-site structural MRI studies: effects of gradient non-linearity correction on phantom and human data. Neuroimage (2006) 30::436–443.[CrossRef][Web of Science][Medline]

  43. Han X, Jovicich J, Salat D, et al. Reliability of MRI-derived measurements of human cerebral cortical thickness: the effects of field strength, scanner upgrade and manufacturer. NeuroImage (2006) 32:(1):180–194.[CrossRef][Web of Science][Medline]

  44. Pardoe H, Pell GS, Abbott DF, Berg AT, Jackson GD. Multi-site voxel-based morphometry: methods and a feasibility demonstration with childhood absence epilepsy. NeuroImage (2008) 42::611–616.[CrossRef][Web of Science][Medline]

  45. Sowell ER, Peterson BS, Kan E, et al. Sex differences in cortical thickness mapped in 176 healthy individuals between 7 and 87 years of age. Cereb Cortex (2006) 17::1550–1560.[CrossRef][Web of Science][Medline]

  46. Good CD, Johnsrude I, Ashburner J, Henson RNA, Friston KJ, Frackowiak RSJ. Cerebral asymmetry and the effects of sex and handedness on brain structure: a voxel-based morphometric analysis of 465 normal adult human brains. NeuroImage (2001) 14::685–700.[CrossRef][Web of Science][Medline]

  47. Chen X, Sachdev PS, Wen W, Anstey KJ. Sex differences in regional gray matter in healthy individuals aged 44-48 years: a voxel-based morphometric study. Neuroimage (2007) 36::691–699.[CrossRef][Web of Science][Medline]

  48. Nopoulos P, Flaum M, Andreasen NC. Sex differences in brain morphology in schizophrenia. Am J Psychiatry (1997) 154::1648–1654.[Abstract/Free Full Text]

  49. Davatzikos C. Why voxel-based morphometric analysis should be used with great caution when characterizing group differences. Neuroimage (2004) 23:(1):17–20.[CrossRef][Web of Science][Medline]

  50. Friston KJ, Ashburner J. Generative and recognition models for neuroanatomy. Neuroimage (2004) 23:(1):21–24.[CrossRef][Web of Science][Medline]


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?



This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
35/1/82    most recent
sbn150v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrowRequest Permissions
Google Scholar
Right arrow Articles by Segall, J. M.
Right arrow Articles by Calhoun, V. D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Segall, J. M.
Right arrow Articles by Calhoun, V. D.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?