Schizophrenia (SZ) is a disabling mental disorder characterized by delusions and auditory hallucinations, as well as impairments in memory, attention, executive, and many other high-order cognitive functions. [1,2] The development of magnetic resonance imaging (MRI) has offered an effective and noninvasive approach to examine the anatomy of the brain and has prompted numerous scientists to explore the underlying neuropathology of SZ. The vast majority of studies show that gray matter (GM) reductions in the brains of SZ patients typically involve the temporal, frontal, and parietal lobes. [3–7] GM abnormalities in the occipital lobe, insula, striatum, thalamus, cerebellum, and cingulate regions have also been reported. [8–12] White matter (WM) fibers link spatially distinct cortical and subcortical regions. Several studies have suggested that SZ is a type of disconnection syndrome [13,14] ; thus, it is reasonable to hypothesize that WM aberrances are associated with this psychosis. [6,15] A number of studies have found white matter volume (WMV) reductions in the corpus callosum, bilateral frontal lobe, and internal capsule in SZ patients. [16–18]
However, these GM or WM deficits were identified using conventional univariate analyses at a group level, which can analyze group-wise differences in brain areas but generally cannot differentiate among the individuals in 2 or more groups. Recently, machine learning, which is a type of multivariate analysis that can automatically discriminate individuals within a sample group, was used to classify several neuropsychiatric disorders and addictions, such as SZ, [19–24] Alzheimer's disease , [25–28] depression, [29–31] attention-deficit hyperactivity disorder, [32–34] and smoking. 
Support vector machine (SVM) is a specific method of supervised machine learning that aims to classify data points by maximizing the margin between classes in a high-dimensional space.  Advantages of this method are the selection of training examples that are most informative for the classification and a good scaling for high dimensions. A previous study discriminated SZ patients from normal controls (NCs) by analyzing multimodal brain imaging data (EEG, structural MRI, and functional MRI) with a SVM classifier and achieved a good classification performance with a 91% accuracy and 100% prediction rate.  A more recent study applied the SVM method to the automatic classification of 72 SZ patients and 74 NCs and achieved an average accuracy of 80%.  These previous findings demonstrate the promising classification performance of the SVM method, as well as the potential for applying this method to identify biomarkers for the diagnosis of psychiatric diseases.
Recent studies have shown that feature reduction can increase computational speed and improve classification performance, by removing uninformative, irrelevant, or redundant features from the classification procedures. [27,39] A recursive feature elimination (RFE) algorithm is an iterative procedure that eliminates 1 backward feature at a time,  and thus, can prevent information loss that can occur by eliminating several features in 1 iteration.  Previous studies have indicated that RFE employed for dimensionality reduction can significantly improve classification accuracy of neuroimaging data  and yield a better classification performance than many other feature reduction methods. 
In this study, we employed voxel-based morphometry (VBM) and region of interest (ROI) analyses to compare differences in gray matter volume (GMV) and WMV between SZ patients and NCs. Significant between-group differences in GMV and WMV were selected as features in the discriminate analysis of 2 groups using SVM combined with RFE.
2 Materials and methods
Forty-one SZ patients (the SZ group) and 42 age- and sex-matched NCs (the NC group) were included in this study. The SZ patients were diagnosed by trained and experienced clinical psychiatrists using a structured clinical interview according to the DSM-IV-TR (SCID).  The SZ and NC groups were recruited from Guangzhou Brain Hospital and the local community, respectively. All subjects were aged between 18 and 45 years, and their biological parents were Han Chinese. Before scanning, a clinical assessment was performed by psychiatrists using the Positive and Negative Syndrome Scale (PANSS).  The subjects obtained a consensus score for each item on all 3 subscales (positive symptoms, negative symptoms, and general psychopathology) that was based on a 7-point scale indicating the severity of the symptom (1 = absent; 2 = minimal; 3 = mild; 4 = moderate; 5 = moderate severe; 6 = severe; 7 = extreme).  The inclusion criteria for all SZ patients included the following: (1) a total score of at least 60 for the 3 PANSS subscales and (2) at least 3 positive symptom items on the PANSS with a score of at least 4.
The exclusion criteria for all subjects included the following: (1) any other psychiatric Axis I disorder meeting DSM-IV criteria, including schizoaffective disorders, mental retardation, major depressive disorder, bipolar disorder, delirium, dementia, memory disorder, and other cognitive disorders; (2) mental disorder due to substance dependence, a serious unstable somatic disease, definite diabetes, thyroid diseases, hypertension or heart disease; (3) narrow angle glaucoma; (4) a history of epilepsy, except for febrile convulsions; (5) alcohol dependence meeting DSM-IV-TR criteria (excluding nicotine dependence); (6) receiving electroconvulsive therapy within the past 6 months; (7) a contraindication for MRI; (8) medical resource neuroleptic malignant syndrome or serious tardive dyskinesia; (9) a serious suicide attempt or an irritative state; (10) noncompliant drug administration or a lack of legal guardians; or (11) lactating, pregnant, or planning pregnancy. In addition, the NCs were excluded if they had a first- or second-degree relative with a psychiatric Axis I disorder according to the DSM-IV criteria. Before enrollment, all subjects or their legal guardians provided written informed consent. These studies were performed according to the Declaration of Helsinki and approved by the Ethics Committees of the Guangzhou Brain Hospital.
2.2 MRI data acquisition
Imaging data were acquired using Philips 3T MR systems (Philips, Best, The Netherlands) located at Guangzhou Brain Hospital. For each subject, an anatomical image was obtained using a sagittal 3-dimensional gradient-echo T1-weighted sequence (TR = 7.6 ms, TED = 3.7 ms, TI = 795 ms, flip angle = 8°, 180 slices, slice thickness = 1 mm, Gap = 0 mm, matrix = 256 × 256, inversion time = 0).
2.3 Image processing
Image processes were described in our previous studies. [29,30,45] Briefly, all T1-weighted MRI data processing was performed using the SPM8 software package (http://www.fil.ion.ucl.ac.uk/spm; Institute of Neurology, University College London, UK). First, each T1-weighted MRI was segmented into 3 tissue maps, including GM, WM, and cerebrospinal fluid (CSF) by using the new segmentation algorithm from SPM8. Second, a customized, population-specific template was created from the segmented tissue maps by using the DARTEL template-creation tool.  Third, all GM and WM maps were warped to the custom template space, using its corresponding smooth, reversible deformation parameters. A modulation was applied by locally multiplying tissue values by the Jacobian determinants derived from the special normalization step.  Finally, all warped, modulated GM and WM images were smoothed with an 8-mm full-width at half-maximum (FWHM) Gaussian kernel.
2.4 VBM analysis
Smoothed modulated GM and WM images were analyzed with SPM8 utilizing a general linear model (GLM). Voxel-wise GMV differences between the SZ and NC groups were investigated using analysis of variance (ANOVA). The covariates included in the model were total brain volume (TBV = total GMV + total WMV), age, gender, and years of education. The resulting statistical map was corrected for multiple comparisons to a significance level of P < 0.05 by combining individual voxels (P < 0.001) and using a cluster size of 173 voxels. This correction was confined within a whole-brain mask and determined using Monte Carlo simulations (FWHM = 8 mm, iterations = 1000) using the AFNI AlphaSim program (http://afni.nih.gov/afni/docpdf/AlphaSim.pdf).
2.5 ROI analysis
Recently, many studies have suggested the use of higher-resolution parcellation, of up to 1000 smaller parcels, instead of using a coarse parcellation scheme of AAL atlas including 90 brain regions. [48–50] Therefore, we parcellated the entire cerebrum into 1024 ROIs with equal sizes using a high-resolution automated anatomical labeling (AAL) atlas,  which was defined from the standard AAL atlas including 90 cerebral regions.  We then calculated the regional GMV (RGMV) and regional WMV (RWMV) in each of the 1024 ROIs from the modulated GM and WM images, respectively. ANOVAs were performed to compare the RGMV and RWMV in 1024 ROIs between the SZ and NC groups. The covariates included in the model were TBV, age, gender, and years of education. The significance level was P < 0.05, corrected by false discovery rate (FDR) correction.
2.6 Correlation between the PANSS and the RGMV or RWMV
We calculated the RGMV or RWMV of brain areas showing significant between-group differences in both the VBM and ROI analyses. The partial correlations between the PANSS scores (including positive, negative, and total scores of PANSS) and the RGMV or RWMV of brain areas were calculated, controlling for the age, gender, years of education, and TBV. The volumes of extracted brain areas and the partial correlations were calculated in Matlab 2010b using in-house scripts.
2.7 SVM with RFE
In this study, a classification method of SVM with RFE was applied to discriminate SZ patients from NCs using the significant between-group differences in RGMV and RWMV identified by the VBM and ROI analyses. SVM is an effective classification method based on the structural risk minimization principle.  It implements the following idea: input vectors are nonlinearly mapped to a high-dimension feature space.  In the feature space, the machine searches for an optimal hyperplane, which maximizes the distance from the hyperplane to the closest examples in each class to classify different groups. Once the hyperplane is found, it can be used to predict the group label of a new testing example. The symbols (+) and (–) are used to represent 2 classes. The input features and the class labels are given. A linear decision function is a simple weighted sum of the input vector plus a bias.  Given the nature of input features, a linear decision function was applied. The weight vector and bias value are represented by w and b, respectively. The decision function is:
If D(x)>0, x belonged to class (+), else if D(x)<0, x belonged to class (–), else x was on the decision boundary. A linear kernel SVM implemented in the LIBSVM toolbox was used in this study. 
In machine learning, some features are uninformative, irrelevant, or redundant for classification, and too many features may cause “over-fitting.”. Therefore, eliminating a number of features not only simplifies the classification model but also improves the classification accuracy.  The SVM-RFE algorithm uses the SVM classifier iteratively to remove redundant features while preserving discriminative features. RFE is an iterative procedure that trains the classifier and removes the smallest ranking criterion feature after ranking.  We chose the square of the feature weights wi (wi2) of the decision function as the ranking criterion. Clearly, the smaller weight of a feature means the less contribution to the decision function. In each iteration, we computed the ranking criteria wi2 for all features and eliminated the feature corresponding to the smallest wi2. We obtained a discriminate feature list after n iterations.
The performance of the classifier is evaluated by leave-one-out cross-validation (LOOCV). The leave-one-out method is a cross-validation error estimation method.  The superiority of the leave-one-out method has been proven. [58,59] The LOOCV involves using data from all but 1 subject as a training set, and the remaining data are used as a test set. This process should be repeated until each sample is used to calculate the overall accuracy of SVM.
Furthermore, we plotted the receiver operating characteristic (ROC) curves and calculated the area under the curve (AUC) to examine the possibility of discriminating SZ patients and NCs correctly. The value of the AUC is between 0 and 1. It is generally believed that a value close to 1 indicates a better prediction performance. In addition to the accuracy and AUC, we measured the sensitivity and specificity to quantify the performance of the SVM. The definitions of sensitivity, specificity, and accuracy are as follows:
where TP is true positive, the number of positive subjects classified as positive; FN is false negative, the number of positive subjects classified as negative; TN is true negative, the number of negative subjects classified as negative; and FP is false positive, the number of negative subjects classified as positive.
A permutation test was applied to evaluate the statistical significance of the classification results.  The permutation test is a type of nonparametric test and can be used to test a null hypothesis.  In our analysis, we permuted the class labels of the training data for 1000 times randomly and performed all classification process with each set of permuted class labels. Based on probability distributions, it is possible to test the null hypothesis indicated by a small P value. The P value was computed as the proportion of accuracies that are no less than the accuracy obtained by the original data. The statistical significance was set at P < 0.05.
3.1 Clinical and demographic characteristics
The clinical and demographic characteristics of all subjects are shown in Table 1. There was no significant difference in age, gender, years of education, and TBV between the SZ and NC groups (P > 0.05). However, the positive, negative, and total scores for the PANSS in the SZ group were significantly higher than those in the NC group (P < 0.0001).
3.2 VBM analysis of GMV and WMV
The VBM analysis indicated that the SZ group showed significant GMV reductions in the left hippocampus (HIP) (Fig. 1A) and the right middle occipital gyrus (MOG) (Fig. 1B), as well as significant GMV increases in the right posterior lobe of the cerebellum (Fig. 1C), left caudate (CAU) (Fig. 1D), and left thalamus (THM) (Fig. 1D) when compared with the NC group. The SZ group showed significant WMV reductions in the right lingual gyrus (LING) (Fig. 2A), right calcarine (CAL) (Fig. 2A), and right cuneus (CUN) (Fig. 2B), as well as significant WMV increases in the right precuneus (PCUN) (Fig. 2C) when compared with the NC group.
3.3 ROI analysis of GMV and WMV
The ROI analysis indicated that the SZ group showed significant GMV reductions in the left MOG (Fig. 3A), the parahippocampal gyrus (PHG) (Fig. 3B), right MOG (Fig. 3C), and right postcentral gyrus (PoCG) (Fig. 3D), as well as significant GMV increases in the right PCUN (Fig. 3E) when compared with the NC group. The SZ group showed significant WMV reductions in the right LING (Fig. 4A), bilateral CAL (Fig. 4A), bilateral CUN (Fig. 4A), and left superior temporal gyrus (STG) (Fig. 4B), as well as significant WMV increases in the left middle frontal gyrus (MFG) (Fig. 4C), right PCUN (Fig. 4D), and right fusiform gyrus (FFG) (Fig. 4E) when compared with the NC group.
3.4 Correlation between PANSS and RGMV or RWMV
We calculated the RGMV or RWMV of 17 brain areas (as shown in Figs. 1–4), which showed significant between-group differences in both the VBM and ROI analyses. The RGMV of the left PHG, as well as the RWMV of the right PCUN and left STG indicated significant positive correlations with positive PANSS scores, after controlling for age, sex, years of education, and TBV (Fig. 5).
3.5 Overall classifier performance
Automatic classifications were performed to distinguish SZ patients from NCs. First, we applied the significant between-group differences for both RGMV and RWMV in the VBM analysis as the input features for the classifier. The linear SVM without RFE classifier achieved an accuracy of 82.6% (P
< 0.05), a sensitivity of 83.8%, and a specificity of 81.3%; the linear SVM with RFE classifier yielded a better classification performance with an accuracy of 88.4% (P
< 0.05), a sensitivity of 91.9%, and a specificity of 84.4% for the same input. Second, we applied the significant between-group differences for both RGMV and RWMV by the ROI analysis as the input features for the classifier. The linear SVM without RFE classifier achieved an accuracy of 79.5% (P
< 0.05), a sensitivity of 71.4%, and a specificity of 87.8%; the linear SVM with RFE classifier achieved a slightly better performance with an accuracy of 80.7% (P
< 0.05), a sensitivity of 73.8%, and a specificity of 87.8%. ROC curves were computed taking each subject's discriminative score as a threshold and are presented in Fig. 6. The areas under the ROC curve for the 4 classifiers were 0.915, 0.948, 0.895, and 0.887 (Table 2).
In the present study, we applied a machine learning method that combined SVM with RFE to discriminate SZ patients from NCs using structural MRI data. Our results indicated that the SZ patients showed significant GM and WM abnormalities primarily in the emotion, memory, and visual systems according to both the VBM and ROI analyses. In the discriminative analyses of SZ patients, the SVM with RFE classifier used with the significant structural abnormalities identified by the VBM analysis as input features achieved the best performances (an accuracy of 88.4%, a sensitivity of 91.9%, and a specificity of 84.4%), which was better than the performances of previous discriminative analyses of SZ patients using structural MRI data. [23,24]
In this study, we achieved better performances in the discriminative analyses of SZ patients using the SVM with RFE classifier, when compared with those using the SVM without RFE classifier. The machine learning method of SVM has been used to obtain available biomarkers in the diagnoses of psychiatric diseases and addictions. [23,62–64] However, some input features are irrelevant or redundant for classification when performing classification with SVM. Thus, the exclusion of uninformative features from the dataset while retaining discriminative features can not only increase the computation speed but also improve the classification performance. [65,66] A previous study by our group selected features based on their rankings of discriminative power by computing F-scores individually and independently and achieved a good result in the discriminative analysis of depressive patients with structural MRI data.  A more recent study applied VBM with a multivariate classification method consisting of a SVM with RFE to discriminate smokers from nonsmokers using their structural MRI data and achieved the highest accuracy of 69.6%. 
In addition to using dimensionality reduction, the feature characters (i.e., structural and functional characteristics) and spatial resolution of brain areas might also affect the classification accuracy. A previous study introduced an M3 method to identify AD patients by combining multimodal imaging and multilevel measures and demonstrated promising classification performance (a classification accuracy of 89.47%, a sensitivity of 87.50%, and a specificity of 90.91%).  In this study, we compared both GM and WM structures between SZ patients and NCs and then applied these structural abnormalities as input features of discriminate analyses. Our results verified that multiple structural characteristics were useful in increasing the classification performance (see supplementary Table S1, http://links.lww.com/MD/B107). Moreover, in the previous study,  the SVM-RFE procedure was applied separately on raw GMVs, mean GMVs from the standard AAL atlas (including 90 cerebral areas) and from the high resolution AAL atlas (including 1024 cerebral areas), and it achieved the highest accuracy when the high-resolution AAL atlas was used. However, our results were inconsistent with this finding and indicated that the classifiers using the input features from the VBM analysis achieved better performances than those using the input features from the ROI analysis. The reason for this inconsistency might be the significant between-groups differences of GMVs in the VBM analysis but not raw GMVs that were adopted as the input features of classifiers in this study.
In this study, we found significant GM and WM abnormalities in many brain areas involved primarily in the emotion, memory, and visual systems, which were consistent with many previous studies. [67–75] More importantly, several important brain areas related to emotion processing and memory functions, such as the left PHG, left STG, and right PCUN, showed significant positive correlations between their RGMV or RWMV and positive PANSS scores. These results were largely consistent with a previous multimodal MRI studies that indicated that SZ patients showed lower values of canonical variants in subcortical regions, including the thalamus, striatum, hippocampus, parahippocampal gyrus, and visual cortex. 
It has been reported that SZ patients show emotional processing deficits in identifying emotions, understanding the feelings of others, inferring people's thoughts, and responding emotionally to others.  Our results indicated that SZ patients showed increased RGMV in the left CAU and left THM, which is consistent with previous findings in chronic SZ patients. [78,79] Interestingly, the increased RGMV in the CAU are thought to be related to antipsychotic usage.  The striatum is an essential neural system in emotional learning that receives sensory input and identifying stimuli that predict rewarding or reinforcing outcomes. [81,82] Moreover, our results indicated that SZ patients showed reduced RWMV in the left STG, which plays an important role in the recognition of learned social-emotional values, especially nonverbal social cues,  and is one of the most consistent regions showing significant structural abnormalities in SZ patients. [3,83] A recent functional near-infrared spectroscopy study showed that the fronto-temporal dysfunction is involved in the pathophysiology of abnormal emotional processing and cognitive inhibition in SZ patients.  We also found significant increases in RGMV and RWMV in the right PCUN, which is linked to self-awareness, mentalizing, and the theory of mind. [85,86] Reduced structural and functional connectivity among the amygdala, PCUN, and parietal regions might contribute to abnormalities in emotional processing in SZ patients. [87,88]
Pronounced deficits in memory, including episodic memory and semantic memory, are also affected in the daily life of SZ patients. [89,90] Our results indicated that the SZ patients showed reduced RGMV in the left HIP and left PHG, which is consistent with previous studies. [6,91] The HIP contributes to the recollection of contextual information in the service of relative memory tasks and is one of the most prominent impaired brain regions in both first-episode and chronic SZ patients. [91,92] In addition, a previous study demonstrated reduced functional and anatomical connectivity between HIP and some brain areas, such as the posterior cingulate cortex, medial prefrontal cortex, and PHG.  Reduced hippocampal volume and aberrant hippocampal modulation might work together to impair episodic memory in SZ patients.
SZ patients have impairments in higher-order processing and primary functions, such as visual processing.  In this study, we found significant GM and WM abnormalities in the visual cortex, including the bilateral MOG, bilateral CAL, bilateral CUN, right FFG, and right LING. GM and WM abnormalities in the primary visual areas and visual association areas of SZ patients have been well documented. [95–97] As for the clinical symptoms, SZ patients sometimes experience visual hallucinations, which is associated with activity in the visual association areas.  A previous ROI study revealed a significant GMV reduction in the bilateral visual association areas in chronic SZ patients.  More recently, a study showed that SZ patients had a lower magnetization transfer ratio and fractional anisotropy in WM adjacent to visual areas.  Thus, our findings may support the notion that both dysfunction of visual areas and disconnections between them and other brain areas contribute to the impaired visual processing in SZ patients.
There are several limitations to this study. First, the sample size was modest. The VBM and ROI analyses with relatively small sample sizes in this study might have insufficient statistical power and were at risk for false-positive errors. Moreover, a large number of subjects can train a more robust model to discriminate SZ patients from NCs. Second, the SZ patients included in this study were chronic SZ patients, and their brain structures might be affected by antipsychotic use. We plan to recruit first-episode, treatment-naive SZ patients in future studies. Third, although we analyzed both GM and WM structures and applied the abnormalities found in SZ patients into the machine learning study, multimodal MRI data including structural and functional MRI data will be useful for achieving better performance in a discriminative analysis. Fourth, this study suffered from that neuropsychological tests were not performed. In the future study, we plan to collect the data of the Measurement and Treatment Research to Improve Cognition in Schizophrenia (MATRICS) Consensus Cognitive Battery (MCCB), which provides a reliable and valid assessment of cognition across major cognitive domains.
In this study, we demonstrated that SZ patients showed GM and WM abnormalities in a variety of brain structures but primarily in the memory, visual, and emotion systems. The SVM-RFE classification showed good performance in distinguishing SZ patients from NCs by using structural features from structural MRI data. Distinct neuroanatomical profiles associated with SZ patients can provide a potential biomarker for disease diagnosis.
1. Wheeler AL, Voineskos AN. A review of structural neuroimaging in schizophrenia: from connectivity to connectomics. Front Hum Neurosci
2. van Os J, Kenis G, Rutten BP. The environment and schizophrenia. Nature
3. Gupta CN, Calhoun VD, Rachakonda S, et al Patterns of gray matter abnormalities in schizophrenia based on an international mega-analysis. Schizophr Bull
4. van den Heuvel MP, Fornito A. Brain networks in schizophrenia. Neuropsychol Rev
5. Vita A, et al Progressive loss of cortical gray matter in schizophrenia: a meta-analysis and meta-regression of longitudinal MRI studies. Transl Psychiatry
6. Sui J, Pearlson GD, Du Y, et al In search of multimodal neuroimaging biomarkers of cognitive deficits in schizophrenia. Biol Psychiatry
7. Li X, et al Structural abnormalities in language circuits in genetic high-risk subjects and schizophrenia patients. Psychiatry Res
8. Tohid H, Faizan M, Faizan U. Alterations of the occipital lobe in schizophrenia. Neurosciences (Riyadh)
9. Meisenzahl EM, Koutsouleris N, Bottlender R, et al Structural brain alterations at different stages of schizophrenia: A voxel-based morphometric study. Schizophrenia Res
10. Lui S, Deng W, Huang XQ, et al Neuroanatomical differences between familial and sporadic schizophrenia and their parents: An optimized voxel-based morphometry
study. Psychiatry Res Neuroimaging
11. Xu L, et al Source-based morphometry: the use of independent component analysis to identify gray matter differences with application to schizophrenia. Human Brain Mapping
12. Byne W, Hazlett EA, Buchsbaum MS, et al The thalamus and schizophrenia: current status of research. Acta Neuropathol
13. Friston KJ. Disconnection and cognitive dysmetria in schizophrenia. Am J Psychiatry
14. Stephan KE, Baldeweg T, Friston KJ. Synaptic plasticity and dysconnection in schizophrenia. Biol Psychiatry
15. Ellison-Wright I, Bullmore E. Meta-analysis of diffusion tensor imaging studies in schizophrenia. Schizophr Res
16. Lei W, et al White matter alterations in first episode treatment-naive patients with deficit schizophrenia: a combined VBM and DTI study. Sci Rep
17. Watson DR, et al A voxel based morphometry study investigating brain structural changes in first episode psychosis. Behav. Brain Res
18. Di X, Chan RCK, Gong QY. White matter reduction in patients with schizophrenia as revealed by voxel-based morphometry
: an activation likelihood estimation meta-analysis. Prog Neuropsychopharmacol Biol Psychiatry
19. Su L, Wang L, Shen H, et al Discriminative analysis of non-linear brain connectivity in schizophrenia: an fMRI Study. Front Hum Neurosci
20. Shen H, Wang L, Liu Y, et al Discriminative analysis of resting-state functional connectivity patterns of schizophrenia using low dimensional embedding of fMRI. Neuroimage
21. Kasparek T, Thomaz CE, Sato JR, et al Maximum-uncertainty linear discrimination analysis of first-episode schizophrenia subjects. Psychiatry Res
22. Tanaka S, Maezawa Y, Kirino E. Classification of schizophrenia patients and healthy controls using p100 event-related potentials for visual processing. Neuropsychobiology
23. Nieuwenhuis M, van Haren NE, Hulshoff Pol HE, et al Classification of schizophrenia patients and healthy controls from structural MRI scans in two large independent samples. Neuroimage
24. Greenstein D, Malley JD, Weisinger B, et al Using multivariate machine learning methods and structural MRI to classify childhood onset schizophrenia and healthy controls. Front Psychiatry
25. Kloppel S, Stonnington CM, Chu C, et al Automatic classification of MR scans in Alzheimers disease. Brain
26. Dai Z, Yan C, Wang Z, et al Discriminative analysis of early Alzheimer's disease using multi-modal imaging and multi-level characterization with multi-classifier (M3). Neuroimage
27. Zhou L, Wang L, Liu L, et al Discriminative brain effective connectivity analysis for Alzheimer's disease: a kernel learning approach upon sparse Gaussian Bayesian network. Conf Comput Vis Pattern Recognit Workshops
28. Li S, Shi F, Pu F, et al Hippocampal shape analysis of Alzheimer disease based on machine learning methods. Am J Neuroradiol
29. Chi M, Guo S, Ning Y, et al Discriminative analysis of major depressive disorder and anxious depression using support vector machine. J Comput Theor Nanosci
30. Chi MY, Guo SW, Ning YP, et al Using support vector machine to identify imaging biomarkers of major depressive disorder and anxious depression. Bio-Inspired Computing—Theories and Applications. Vol 472
2014; Berlin: Springer, 63–67.
31. Gong Q, Wu Q, Scarpazza C, et al Prognostic prediction of therapeutic response in depression using high-field MR imaging. Neuroimage
32. Iannaccone R, Hauser TU, Ball J, et al Classifying adolescent attention-deficit/hyperactivity disorder (ADHD) based on functional and structural imaging. Eur Child Adolesc Psychiatry
33. Zhu CZ, Zang YF, Liang M, et al Discriminative analysis of brain function at resting-state for attention-deficit/hyperactivity disorder. Medical Image Computing and Computer-Assisted Intervention: MICCAI International Conference on Medical Image Computing and Computer-Assisted Intervention. Vol 8
2005; Berlin: Springer, 468–475.
34. Zhu C-Z, Zang Y-F, Cao Q-J, et al Fisher discriminative analysis of resting-state brain function for attention-deficit/hyperactivity disorder. NeuroImage
35. Ding X, Yang Y, Stein EA, et al Multivariate classification of smokers and nonsmokers using SVM-RFE on structural MRI images. Human Brain mapping
36. Cherkassky V. The nature of statistical learning theory. IEEE Trans Neural Netw
37. Sui J, Castro E, He H, et al Combination of FMRI-SMRI-EEG data improves discrimination of schizophrenia patients by ensemble feature selection. Conf Proc IEEE Eng Med Biol Soc
38. Savio A, Graña M. Local activity features for computer aided diagnosis of schizophrenia on resting-state fMRI. Neurocomputing
39. Liang Y, Zhang F, Wang J, et al Prediction of drought-resistant genes in Arabidopsis thaliana using SVM-RFE. PLoS One
40. Guyon I, Weston J, Barnhill S. Gene selection for cancer classification using support vector machines. Mach Learn
41. Yoon S, Kim S. Mutual information-based SVM-RFE for diagnostic classification of digitized mammograms. Pattern Recognit Lett
42. First MB. User's Guide for the Structured Clinical Interview for DSM-IV Axis II Personality Disorders: SCID-II. 1997; Washington, DC: American Psychiatric Press, 91.
43. Kay SR, Fiszbein A, Opler LA. The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophr Bull
44. van Tol MJ, van der Meer L, Bruggeman R, et al Voxel-based gray and white matter morphometry correlates of hallucinations in schizophrenia: The superior temporal gyrus does not stand alone. Neuroimage Clin
45. Qi H, Ning Y, Li J, et al Gray matter volume abnormalities in depressive patients with and without anxiety disorders. Medicine
46. Ashburner J. A fast diffeomorphic image registration algorithm. Neuroimage
47. Good CD, Johnsrude I, Ashburner J, et al Cerebral asymmetry and the effects of sex and handedness on brain structure: a voxel-based morphometric analysis of 465 normal adult human brains. Neuroimage
48. Bai F, Shu N, Yuan Y, et al Topologically convergent and divergent structural connectivity patterns between patients with remitted geriatric depression and amnestic mild cognitive impairment. J Neurosci
49. Cao M, Wang JH, Dai ZJ, et al Topological organization of the human brain functional connectome across the lifespan. Dev Cogn Neurosci
50. Wang J, Zuo X, Dai Z, et al Disrupted functional brain connectome in individuals at risk for Alzheimer's disease. Biol Psychiatry
51. Zalesky A, Fornito A, Harding IH, et al Whole-brain anatomical networks: does the choice of nodes matter? Neuroimage
52. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, et al Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage
53. Cortes C, Vapnik V. Support-vector networks. Mach Learn
54. Duda RO, Hart PE. Pattern Classification and Scene Aanalysis. 1973; New York: Wiley, 482.
55. Chang CC, Lin CJ. LIBSVM: a library for support vector machines. ACM Trans Intell Syst Technol
2011; 2: Article 27.
56. Ding J, Shi J, Wu FX. SVM-RFE based feature selection for tandem mass spectrum quality assessment. Int J Data Min Bioinform
57. Kim DR, Ali M, Sur D, et al Determining optimal neighborhood size for ecological studies using leave-one-out cross validation. Int J Health Geogr
58. Elisseeff A, Pontil M. Leave one out error and stability of learning algorithms with applications. Advances in Learning Theory: Methods, Models and Application—NATO Science Series III: Computer and Systems Science
59. Kearns M, Ron D. Algorithmic stability and sanity-check bounds for leave-one-out cross-validation. Neural Comput
60. Golland P, Fischl B. Permutation tests for classification: towards statistical significance in image-based studies. Inf Process Med Imaging
61. Liu F, Guo W, Fouche JP, et al Multivariate classification of social anxiety disorder using whole brain functional connectivity. Brain Struct Funct
62. Ding X, Yang Y, Stein EA, et al Multivariate classification of smokers and nonsmokers using SVM-RFE on structural MRI images. Hum Brain Mapp
63. Fan Y, et al Unaffected family members and schizophrenia patients share brain structure patterns: A high-dimensional pattern classification study. Biol Psychiatry
64. Caprihan A, Pearlson GD, Calhoun VD. Application of principal component analysis to distinguish patients with schizophrenia from healthy controls based on fractional anisotropy measurements. Neuroimage
65. Pereira F, Mitchell T, Botvinick M. Machine learning classifiers and fMRI: a tutorial overview. Neuroimage
2009; 45 ((1 Suppl)):S199–S209.
66. Dosenbach NU, Nardos B, Cohen AL, et al Prediction of individual brain maturity using fMRI. Science
67. Guo W, Hu M, Fan X, et al Decreased gray matter volume in the left middle temporal gyrus as a candidate biomarker for schizophrenia: a study of drug naive, first-episode schizophrenia patients and unaffected siblings. Schizophr Res
68. Lyu H, Hu M, Eyler LT, et al Regional white matter abnormalities in drug-naive, first-episode schizophrenia patients and their healthy unaffected siblings. Aust N Z J Psychiatry
69. Hu M, Li J, Eyler L, et al Decreased left middle temporal gyrus volume in antipsychotic drug-naive, first-episode schizophrenia patients and their healthy unaffected siblings. Schizophr Res
70. Goghari VM, Rehm K, Carter CS, et al Regionally specific cortical thinning and gray matter abnormalities in the healthy relatives of schizophrenia patients. Cereb Cortex
71. Fornito A, Yucel M, Patti J, et al Mapping grey matter reductions in schizophrenia: an anatomical likelihood estimation analysis of voxel-based morphometry
studies. Schizophr Res
72. Bora E, Fornito A, Radua J, et al Neuroanatomical abnormalities in schizophrenia: a multimodal voxelwise meta-analysis and meta-regression analysis. Schizophr Res
73. Rajarethinam R, Sahni S, Rosenberg DR, et al Reduced superior temporal gyrus volume in young offspring of patients with schizophrenia. Am J Psychiatry
74. Kuroki N, Shenton ME, Salisbury DF, et al Middle and inferior temporal gyrus gray matter volume abnormalities in first-episode schizophrenia: an MRI study. Am J Psychiatry
75. Onitsuka T, Shenton ME, Salisbury DF, et al Middle and inferior temporal gyrus gray matter volume abnormalities in chronic schizophrenia: an MRI study. Am J Psychiatry
76. Sui J, Pearlson GD, Du Y, et al In search of multimodal neuroimaging biomarkers of cognitive deficits in schizophrenia. Biol Psychiatry
77. Green MF, Horan WP, Lee J. Social cognition in schizophrenia. Nat Rev Neurosci
78. Glahn DC, Laird AR, Ellison-Wright I, et al Meta-analysis of gray matter anomalies in schizophrenia: application of anatomic likelihood estimation and network analysis. Biol Psychiatry
79. Hulshoff Pol HE, Schnack HG, Mandl RC, et al Focal gray matter density changes in schizophrenia. Arch Gen Psychiatry
80. Okugawa G, Nobuhara K, Takase K, et al Olanzapine increases grey and white matter volumes in the caudate nucleus of patients with schizophrenia. Neuropsychobiology
81. Ochsner KN. The social-emotional processing stream: five core constructs and their translational potential for schizophrenia and beyond. Biol Psychiatry
82. Schultz W. Neural coding of basic reward terms of animal learning theory, game theory, microeconomics and behavioural ecology. Curr Opin Neurobiol
83. Shenton ME, Dickey CC, Frumin M, et al A review of MRI findings in schizophrenia. Schizophrenia Res
84. Egashira K, Matsuo K, Nakashima M, et al Blunted brain activation in patients with schizophrenia in response to emotional cognitive inhibition: a functional near-infrared spectroscopy study. Schizophr Res
85. Mar RA. The neural bases of social cognition and story comprehension. Annu Rev Psychol
86. Van Overwalle F, Baetens K. Understanding others’ actions and goals by mirror and mentalizing systems: a meta-analysis. Neuroimage
87. Mukherjee P, Whalley HC, McKirdy JW, et al Lower effective connectivity between amygdala and parietal regions in response to fearful faces in schizophrenia. Schizophrenia Res
88. Nazeri A, Chakravarty MM, Felsky D, et al Alterations of superficial white matter in schizophrenia and relationship to cognitive performance. Neuropsychopharmacology
89. Aleman A, Hijman R, de Haan EHF, et al Memory impairment in schizophrenia: a meta-analysis. Am J Psychiatry
90. Pender NP. From conditioning to conscious recollection: memory systems of the brain. Trends Cognitive Sci
91. Adriano F, Caltagirone C, Spalletta G. Hippocampal volume reduction in first-episode and chronic schizophrenia: a review and meta-analysis. Neuroscientist
92. Ranganath C, Minzenberg MJ, Ragland JD. The cognitive neuroscience of memory function and dysfunction in schizophrenia. Biol Psychiatry
93. Zhou Y, Shu N, Liu Y, et al Altered resting-state functional connectivity and anatomical connectivity of hippocampus in schizophrenia. Schizophr Res
94. Kim D, Zemon V, Saperstein A, et al Dysfunction of early-stage visual processing in schizophrenia: harmonic analysis. Schizophr Res
95. Ananth H, Popescu I, Critchley HD, et al Cortical and subcortical gray matter abnormalities in schizophrenia determined through structural magnetic resonance imaging with optimized volumetric voxel-based morphometry
. Am J Psychiatry
96. Onitsuka T, McCarley RW, Kuroki N, et al Occipital lobe gray matter volume in male patients with chronic schizophrenia: a quantitative MRI study. Schizophrenia Res
97. Palaniyappan L, Al-Radaideh A, Mougin O, et al Combined white matter imaging suggests myelination defects in visual processing regions in schizophrenia. Neuropsychopharmacology
98. Silbersweig DA, Stern E, Frith C, et al A functional neuroanatomy of hallucinations in schizophrenia. Nature