Machine learning techniques in a structural and functional MRI diagnostic approach in schizophrenia: a systematic review

Figures & data

Figure 1 PRISMA flow diagram of included studies.

Table 1 Summary of included studies of machine learning techniques classifying schizophrenia using structural neuroimaging

First author, Year	Data utilized	Sample size and diagnosis	Machine learning model	Best accuracy (%)	Other measures (sensibility, specificity, AUC, precision, error rate)	Data features	Brain regions and networks involved	Commentary
Greenstein, 2012^Citation17	sMRI 1.5T	197 subjects: -98 Childhood onset SCZ -99 HC	RF	73.7%	N.A.	Cortical thickness for 68 frontal, temporal, parietal, and occipital lobe regions, and bilateral lateral ventricle, thalamus, and hippocampus volumes to yield the 74 variables authors used as features	Left temporal lobes, bilateral dorsolateral prefrontal regions and left medial parietal lobes.	Authors have chosen to use regional brain measures that provide lower resolution compared to the higher resolution voxel-wise measures
Castellani, 2012^Citation15	sMRI 1.5T	108 subjects: -54 SCZ -54 HC	SVM	66.4–84.1%	N.A.	Landmark point detection of DLPFC and Feature vocabulary	Dorsolateral prefrontal cortex	Findings may have partially been limited by administration of AP or by length of illness
Iwabuchi, 2013^Citation20	sMRI 3T and 7T	39 subjects: -19 SCZ -20 HC	SVM	GM 3T 66.6% 7T 77% WM 3T 63.9% 7T 69.1%	Sn 63.2% Sp 70.0% Sn 78.9% Sp 75% Sn 57.9% Sp 70.0% Sn 63.2% Sp 75.0%	The normalized, modulated, unsmoothed WM, and GM images for the 3- and 7-T datasets were then used as inputs to the separate linear SVM classifiers	Insula, anterior cingulate cortex, thalamus, superior temporal cortex, parahippocampal gyrus	First study compares head-to-head GM and WM in SVM analysis
Lu, 2016^Citation14	sMRI 3T	83 subjects: -41 SCZ -42 HC	SVM RFE	SVM-RFE, 88.4%	Sn 91.9% Sp 84.4%	First, RGMV and RWMV in the VBM analysis; second, the significant between-group differences for both RGMV and RWMV by the ROI analysis as the input features for the classifier	Middle occipital gyrus, calcarine, cuneus, fusiform gyrus, lingual gyrus, hippocampus, parahippocampal gyrus	Chronic SCZ patients brain structures might be affected by antipsychotic use
Pinaya, 2016^Citation18	sMRI 1.5T	258 subjects: -143 SCZ -32 FEP -83 HC	DBN-DNN SVM	DBN-DNN, 73.6% ±6.84	Sn 76.4% ±0.1 Sp 70.7% ±12.2 Error rate 26.1±6.5	Sixty-eight brain regions (thirty-four in each hemisphere) and volumes of the forty-five anatomical structures	Frontal, temporal, parietal and insular cortices, corpus callosum, putamen, cerebellum.	They classified HC and SCZ; FEP as third classification on a continuum between HC and SCZ
Xiao, 2017^Citation16	sMRI 3T	326 subjects: -163 Drug-naïve FES -163 HC	SVM	surface area 85,0% cortical thickness 81,8%	Sn 83,0% Sp 87,0% Sn 76.9% Sp 85.0%	The cortical thickness and surface area features of 68 cortical regions from the Desikan atlas were extracted: frontal, temporal, parietal, and occipital lobe regions	Left fusiform, lingual, posterior cingulate, supramarginal, insula, inferior parietal, rostral anterior cingulate, rostral middle frontal AND right isthmus cingulate, lateral occipital, lingual, caudal middle frontal, inferior parietal, frontal and temporal pole cortex	Most of interested areas belong to the DMN, salience network or visual system
Salvador, 2017^Citation13	sMRI 1.5T	383 subjects: -128 SCZ -128 BD -127 HC	Ridge LASSO SVC RF GPC Elastic Net L0	74–77% (SCZ vs HC)	N.A.	Cortical thickness of left and right hemispheres; Cortical volume of left and right hemispheres; Grey and White matter voxel based morphometry (VBM) images; Grey and White matter wavelet based morphometry (WBM) images	Cortical thickness, Cortical volume, Grey and White matter	Multi-class classifications considering all groups at the same time have made high predictive power for the SCZ group
Amin, 2018^Citation47	sMRI 3T T1-weighted	298 subjects: -144 SCZ -154 HC	Translation-based multimodal fusion approach	N.A.	N.A.	dFNC as the functional features and ICA-based sources from gray matter densities as the structural features	Putamen, insular, precuneus, posterior cingulate cortex and temporal cortex	The deep learning approach has a potential for learning dynamic features from the fMRI data, and thus can offer a favorable framework for multimodal fusion in the brain imaging research.
Pinaya, 2019^Citation19	sMRI	75 subjects: -40 HC -35 SCZ	Deep autoencoder	N.A.	Mean deviation metric of 1,14±0.28	Cortical cortical thickness of the 68 cortical subregions (34 per hemisphere) and the anatomical structural volume of 36 structures; total number: 104	Left ventral diencephalon, left choroid plexus, left lateral ventricle, right cuneus, right superior temporal, left putamen, right lateral ventricle, left cerebellum cortex, left precentral.	The model was also able to detect distinct patterns of neuroanatomical deviations in SCZ. The deep autoencoder can be used to measure the overall deviation metric of an individual and elucidate which regions are the most different compared to healthy group.

Note: For this systematic review the inclusion criteria was a Jadad score >3.

Abbreviations: AOS, adolescent-onset schizophrenia; ASD, autism spectrum disorder; AUC, area under ROC curve; DBN, deep belief network; dFNC, dynamic functional connectivity; DLPFC, dorsolateral prefrontal cortex; DNN, deep neural network; eMIC, extended maximal information coefficient; FCs, functional connectivities; FEP, first episode psychosis; fMRI, functional magnetic resonance imaging; GM, grey matter; GPC, Gaussian process classifiers; GSM, generalized sparse model; HC, healthy control; LDA, linear discriminant analysis; LIBSVM, leave-one-out SVM; LOOCV, leave-one-out cross validation; MDD, major depressive disorder; ML, machine learning; MVPA, multivariate pattern analysis; L_0, L₀ norm regularization; LASSO, Least Absolute Shrinkage and Selection Operator; N.A., not applicable; PCC, Pearson correlation coefficient; RF, random forest; RGMV, regional grey matter volume; RFE, recursive feature elimination; ROC, receiver-operating characteristic curve analysis; ROI, region of interest; rsMRI, resting state magnetic resonance imaging; RWMV, regional white matter volume; SCZ, schizophrenia; sMRI, structural magnetic resonance imaging; Sn, sensitivity; SNPs, single-nucleotide polymorphisms; Sp, specificity; SRVS, sparse-representation-based variable selection; SVC, support vector classifier; SVM, support vector machine; T, tesla; VBM, voxel-based morphometry; v-ELM, voting-ELM; ν-MKL, multiple kernel learning; VMHC, voxel-mirrored homotopic connectivity; WBM, wavelet-based morphometry; WM, white matter.

Table 2 Summary of included studies of Machine Learning techniques classifying Schizophrenia using functional neuroimaging

First author, Year	Data utilized	Sample size and diagnosis	Machine learning model	Best accuracy (%)	Other measures (sensibility, specificity, AUC, precision, error rate)	Data features	Brain regions and networks involved	Commentary
Yoon, 2008^Citation24	fMRI 1.5T	34 subjects: −19 SCZ -15 HC	MVPA	93%	N.A.	Images were acquired with a echo-planar imaging (EPI) in the anterior commissure- posterior commissure (AC-PC) aligned axial plane. Twenty-seven interleaved slices for whole-brain coverage were collected with a 22-cm field of view	Prefrontal Cortex, sensory cortex, visual cortex	Accuracy of MVPA was higher for HC group
Yang, 2010^Citation33	fMRI 3T SNPs Combined fMRI-SNPs	40 subjects: −20 SCZ -20 HC	SVM-E/C Hybrid ML method	SVM-C 67, 5% Hybrid ML 87, 3%	SVM-C Sn 65% Sp 70% Hybrid ML Sn 85, 8% Sp 88.8%	150 SNPs from a database; auditory stimuli were used and found to be effective in eliciting fMRI BOLD patterns differentiating HC from SZ subjects	Post-/pre-central gyrus, paracentral lobule, cingulate gyrus, superior and inferior parietal lobule, precuneus	Hybrid ML method using fMRI and SNP data
Su, 2013^Citation39	fMRI 1.5T	64 subjects: -32 SCZ -32 HC	SVM	MIC 76, 6% PCC 81, 2% eMIC 82, 8%	MIC Sn 71, 9% Sp 81, 2% PCC Sn 81, 2% Sp 81, 2% eMIC Sn 81, 2% Sp 84, 4%	180 registered fMRI volumes with the MNI template than further divided into 116 regions according to the automatic anatomical labelling atlas. The atlas divides the cerebrum into 90 regions (45 in each hemisphere) and divides the cerebellum into 26 regions (nine in each cerebellar hemisphere and eight in the vermis)	Default mode network, cerebellum, visual network, sensorimotor network, fronto-parietal network, cingulo-opercular network	The study does not include definiton of ROIs and it is confined to connectivity analyses
Zhu, 2014^Citation31	rs-fMRI 3T	45 subjects: -27 SCZ -28 HC	LOOCV	83.6%	Sn 81.5% Sp 85.7%	Blood oxygen level–dependent (BOLD) time courses were generated for 23 regions of interest (ROIs)	Wernicke’s area, inferior parietal, Broca’s area,pars triangularis, middle frontal, pars opercularis, orbitalis, inferior temporal, superior frontal, caudate, putamen, ventral thalamus, cerebellum crus, striate, extrastriate, posterior, superior parietal, superior temporale, cingulate.	Authors propose an adaptive learning algorithm to distinguish HC and SCZ patients using resting-state functional language network
Castro, 2014^Citation43	fMRI	52 subjects: -31 SCZ -21 HC	ν-MKL	85%	Sn 90% Sp 76%	Authors used the simulation toolbox for fMRI data (SimTB) mimics the BOLD response of subjects. The experimental design is characterized by the absence of task events. Among the 30 components available by default on SimTB, they did not include in the simulation those associated with the visual cortex, the precentral and postcentral gyri, the subcortical nuclei and the hippocampus	Right Caudate Nucleus, precuneus, cingulated gyrus, occipital, gyrus, parietal lobe and left precuneus, temporal gyrus, parahippocampal gyrus, paracentral lobule	All patients were on stable medication prior to the scan session
Arbabshirani, 2014^Citation34	fMRI 3T	370 subjects: −195 SCZ -175 HC	SVM	88, 2%	Sn 86, 7% Sp 89, 5%	47 ICNs were selected, resulting in 1081 FNC features for each subjects: in total we extracted 1128 features for each subject	Subcortical, auditory, visual, somatomotor regions and control processes, default-mode, cerebellar networks	The study evaluates functional network connectivity and autoconnectivity improving the classification performance significantly
Watanabe, 2014^Citation38	rs-fMRI	91 subjects: -54 SCZ -67 HC	SVM LASSO Graph Net Elastic Net ROC	77–88.2%	N.A.	Authors produced a whole-brain resting state functional connectome. 347 non-overlapping spherical nodes are placed throughout the entire brain in a regularly-spaced grid pattern; each of these nodes represents a pseudo-spherical ROI. For each of these nodes, a single representative time-series is assigned by spatially averaging the BOLD signals falling within the ROI	Intra-frontoparietal, frontoparietal-default, intracerebellum networks, lateral prefrontal cortex	Authors introduced empirical risk minimization framework that accounts for 6-D spatial structure in the connectome via the fused LASSOand Graph Net regularizer
Cao, 2014^Citation32	fMRI 1.5-3T SNPs	208 subjects: -92 SCZ -116 HC	SRVS GSM	SNPs 83.1%±1.3 fMRI 63.1%±0.7	N.A.	SNPs/fMRI voxels (n=759,075/153,594. For each method, authors carried out 100 runs and the average of the classification ratios was used as the final identification accuracy	Temporal lobe, lateral frontal lobe, occipital lobe, motor cortex.	Different nature of the data used for the ML analysis
Matsuraba, 2015^Citation36	rs-fMRI	211 subjects: -48 SCZ -46 BD -117 HC	DGM	76.6%	Sn 58.5% Sp 84.9%	Authors parcellated each fMRI image into 116 ROIs using an automated anatomical labeling template. The data of subjects whose fMRI images did not match the template even after the spatial normalization were also discarded	Left and right thalamus, left fusiform, right rectus, right middle cingulum, right supramarginal, left mild temporal, left superior orbital frontal, left superior temporal, right caudate	The DGM was a generative model implemented using deep neural networks
Koch, 2015^Citation23	fMRI 1.5T	98 subjects: -44 SCZ -54 HC	MVPA SVM ROC LIBSVM	69.3–93.2%	Sn 70.5–100% Sp 40.9–93.2%	Gradient-echo echo-planar imaging was used to produce eighteen slices approximately parallel to the AC-PC plane, covering the inferior part of the frontal lobe (superior border above the caudate nucleus), the entire temporal lobe, and large parts of the occipital region. Six fMRI volumes were acquired per trial, resulting in 450 volumes per run	Ventral striatum, right pallidum, putamen, right inferior frontal gyrus, nucleus accumbens, amygdala, insula, thalamus, inferior temporal gyrus	Seven patients taking atypical AP. 21 typical AP and 16 not receiving any medications
Chyzhyk, 2015^Citation30	rs-fMRI	147 subjects: -72 SCZ -75 HC	V-ELM SVM RF	VMHC ≈90%	N.A.	rs-fMRI data was collected with single-shot full k-space EPI with ramp sampling correction using the AC-PC as a reference. The first 6 volumes were discarded for scanner calibration leaving 144 time volumes	Inferior temporal gyrus, parahippocampalgyrus, planumpolare, temporal fusiform cortex, thalamus	Application of SVM and RF to the features extracted from cross-validation as the ensembles of ELM
Cabral, 2016^Citation21	rs-fMRI sMRI	145 subjects: -71 SCZ -74 HC	MVPA	rs-fMRI 70.5% sMRI 69.7% s/rs-fMRI 75%	GM, Sn 82.4% Sp 63.4% acc. 69.7% RS, Sn 71.2% Sp 69.7% rs-fMRI, Sn 100% Sp 62.5–87.5% acc. 76.9–92.3% MRI, Sn 75% Sp 78.8% acc. 71.2%	sMRI data were preprocessed to segment the brain into WM, GM, and cerebral spinal fluid. Blood oxygenation level dependent images of the whole brain using an EPI sequence were acquired in 32 axial slices using the AC-PC plane as a reference. RS scans resulted in 304 seconds duration (152 volumes). After discarding the first 10 images and 2 dummy scans, the remaining 140 images were unwarped	Fronto-occipital, fronto-parietal, fronto.temporal, cortico-thalamic regions, left inferior temporal gyrus, parahippocampal gyrus	The AP medication at MRI scan was converted to chlorpromazine and olanzapine equivalents; chronic effects of AP could contribute to alterate FC with longer duration of illness
Kim, 2016^Citation37	rs-fMRI 3T	100 subjects: -50 SCZ -50 HC	DNN		Error rate of DNN: 14.2% vs error rate of SVM: 22.3%.	Authors acquired 150 volumes. The first 5 volumes were removed to allow equilibration of the T1-related signal. Each of the 116 AAL regions was readily available in each voxel of the normalized EPI volumes	Whole-brain FC patterns	The study successfully demonstrated the feasibility of the DNN classifier toward the automated diagnosis of SZ patients
Liu, 2017^Citation29	rs-fMRI 3T	79 subjects: -48 Drug-naïve SCZ -31 HC	SVM	89.9%	Sn 91.67% Sp 87.10%	For each subject, the fMRI scan lasted for 480s, and 240 volumes were obtained. The first 10 volumes of each subject were discarded to certain steady-state conditions	Left postcentral gyrus, left superior temporal gyrus, left paracentral lobule, right precentral gyrus, right inferior parietal lobule, right middle frontal gyrus, bilateral precuneus	Study includes only AOS without SCZ patients
Guo, 2017^Citation27	fMRI short/long-range FCs	96 subjects: -28 SCZ -40 HC -28 relatives	SVM ROC	94.6%	Sn 92.9% Sp 96.4%	Long-range and short-range FCs. rs-fMRI images were obtained with a gradient-echo echo-planar imaging sequence using 250 volumes	Default-mode network, sensorimotor circuits, right superior parietal lobule, left fusiform gyrus, cerebellum	The results may be confounded by acute positive symptoms since the patients sample was unmedicated and recent onset
Orban, 2017^Citation45	fMRI	382 subjects: -191 SCZ -191 HC	SVM	84%	N.A.	Functional brain connectomes included 2016 functional connections between 64 brain parcels	Whole brain connectivity	Brain imaging data from 6 independent studies and datasets
Wang, 2017^Citation28	rs-fMRI 3T	79 subjects: -48 AOS -31 HC	SVM	90.1%	Sn 88.2% Sp 91.9%	Brain regions with significantly different ReHo values between patients with AOS and healthy controls	Bilateral superior medial prefrontal cortex, left superior temporal gyrus, right precentral lobule, right inferior parietal lobule, left paracentral lobule	Different number of participants in the two groups. with AOS (with less confounding factors) instead of SCZ
Chen, 2017^Citation22	rs-fMRI 3T	60 subjects: -20 SCZ -20 MDD -20 HC	MVPA	85% SCZ vs MDD (average 79%)	Sn 98% Sp 60%	The scan used a gradient-echo echo-planar imaging sequence. A total of 255 volumes were collected for each subject. Each functional volume contained 35 slices. To minimize the effects of scanner signal stabilization, the first five volumes of each subject were excluded from all analyses	Orbitofrontal cortex	The systematically different medications for SCZ and MDD may have different effects on functional connectivity; the medicated patients were in stable condition and this may have an impact on the functional connectivity of cortical networks
Wang, 2017^Citation26	rs-fMRI 3T	79 subjects: -48 drug-naïve AOS -31 HC	SVM FC analysis LIBSVM ROC	92.4%	Sn 89.6% Sp 96.8%	Authors used regional homogeneity (ReHo), a measurement that reflects brain local functional connectivity or synchronization and indicates regional integration of information processing	Right middle frontal gyrus, right superior medial prefrontal cortex, left superior temporal gyrus	SVM analysis applied to an independent dataset
Bae, 2017^Citation40	fMRI 3T	75 subjects: -21 SCZ -54 HC	SVM	92.1%±10.5	Sn 92.0±15.8% Sp 92.1±8.1% Precision 94±6.3%	Authors created anatomic labels for 90 ROIs from the image database. Any subject with fewer than 85 ROIs automatically labeled was excluded	Anterior right cingulate cortex, superior right temporal region, inferior left parietal region	Possible influence of pharmacological treatment and disease stage on the investigated functional connections, they used only n-back tests without rs-fMRI
Qureshi, 2017^Citation44	rs-fMRI 3T	144 subjects: -72 SCZ -72 HC	ELM SVM LDA RF	99.3%	Sn 100% Sp 98, 6%	Authors used measures including the mean cortical thickness, cortical thickness standard deviation, surface area, volume, mean curvature, white matter volume, subcortical segment volume, subcortical intensity, and overall brain volume and intensity as the structural features	Cortical thickness, Surface area, WM/subcortical/overall volume, curvature, global average functional connectivity	Authors developed an ELM, whose effectiveness was compared with that of more known ML methods
Reavis, 2017^Citation25	fMRI 3T	148 subjects: -50 SCZ -51 BD -47 HC	MVPA	41%	N.A.	Structural data were processed and parcellated into anatomical regions used to constrain ROI definitions	Lateral occipital lobe	The paper shows MVPA can be used successfully to classify individual perceptual stimuli in SCZ and BD. The results do not provide groups differences with utilizeted stimuli
Chen, 2017^Citation35	rs-fMRI	109 subjects: -35 SCZ -31 HC -22 ASD -21 HC	MVPA LOOCV	Schizophrenia, 83% ASD, 80.4%	SCZ, Sn 80% Sp 87.1% AUC 0.83 ASD, Sn 77.27% Sp 83.23% AUC 0.76	A total of 255 volumes were collected for each subject. Each functional volume contained 35 slices. The first five volumes of each subject were excluded from all analyses. Only the brain areas within the areas identified in the MVPA were included in the subsequent analyses. They were regarded as meaningful brain areas and used as ROIs in the functional connectivity analysis	Default-mode network, salience network	The two dataset were from two different imaging platforms
Pläschke, 2017^Citation41	rs-fMRI	170 subjects: -86 SCZ -84 HC	SVM	61–72%	Sn 65–77% Sp 46–69% AUC 0.61–0.79	Authors investigated 12 functional networks. Only meta-analytic networks with a minimum of 10 nodes were included, since a lower number of features are uninformative for robust classification	Emotion-processing, empathy and cognitive action control networks	Young-old classification was based on all networks and outperformed clinical classification
Liu, 2018^Citation42	rs-fMRI	79 subjects: −48 Drug-Naïve FES AOS -31 HC	SVM VMHC	94.93%	Sn 100% Sp 87.09%	For each subject, the fMRI scan lasted for 480 s, and 240 volumes were obtained. The first 10 volumes of each subject were discarded to certain steady-state conditions and for participants to acclimatize to a scanning environment during the analyzed portion of the data	Fusiform gyrus, superior temporal gyrus, insula, precentral gyrus and precuneus	authors also used a battery of neurocognitive tests and they demonstrated deficits in multiple cognitive functions in patients
Zeng, 2018^Citation46	rs-fMRI 1.5-3T	734 subjects: -357 SCZ -377 HC	RFE-SVM RFE-LDA SAN DANS	81–85%	Sn 75–83% Sp 81–86%	Authors used multi-atlas based whole-brain fcMRI in the MVPA, which measures functional connectivity of the same image in different spaces. The three atlases used included 176, 160 and 116 ROIs respectively	Cortical-striatal-cerebellar circuit (default, salience, frontoparietal control, ventral attention, dorsal attention and somatomotor, visual).	This paper provides for a multi-variate based whole-brain fcMRI pattern analysis to ensure the optimal use of the wealth of information present in fcMRI scans.
Amin, 2018^Citation47	fMRI 3T T2-weighted	298 subjects: -144 SCZ -154 HC	Translation-based multimodal fusion approach	N.A.	N.A.	dFNC as the functional features and ICA-based sources from grey matter densities as the structural features	Putamen, insular, precuneus, posterior cingulate cortex and temporal cortex	The deep learning approach has a potential for learning dynamic features from the fMRI data, and thus can offer a favorable framework for multimodal fusion in the brain imaging research.

Note: For this systematic review the inclusion criteria was a Jadad score >3.

Abbreviations: AC-PC plane, bicommissural plane; AOS, adolescent-onset schizophrenia; ASD, autism spectrum disorder; AUC, area under ROC curve, Cohe-ReHo, regional homogeneity based on coherence; DANS, discriminant autoencoder network with sparsity constraint; DBN, Deep Belief Network; DNN, deep neural network; dFNC, dynamic functional connectivity; DGM, deep neural generative model; eMIC, extended maximal information coefficient; EPI, echo-planar imaging; FC, functional connectivity; FEP, first episode psychosis; fMRI, functional magnetic resonance imaging; GM, grey matter; GPC, Gaussian process classifiers; GSM, generalized sparse model; HC, healthy control; ICNs, intrinsic connectivity networks; LDA, linear discriminant analysis; LIBSVM, leave-one-out SVM; LOOCV, leave-one-out cross validation; MDD, major depressive disorder; ML, machine learning; MVPA, multivariate pattern analysis; L0, L0 norm regularization; LASSO, Least Absolute Shrinkage and Selection Operator; N.A., not applicable; PCC, Pearson correlation coefficient; ReHo, regional homogeneity; RF, random forest; RFE, recursive feature elimination; ROC, receiver-operating characteristic curve analysis; ROI, region of interest; rsMRI, resting state magnetic resonance imaging; SAN, sparse autoencoder network; SCZ, schizophrenia; sMRI, structural magnetic resonance imaging; Sn, sensitivity; SimTB, simulation toolbox for fMRI data; SNP, single-nucleotide polymorphisms; Sp, specificity; SRVS, sparse-representation-based variable selection; SVC, support vector classifier; SVM, support vector machine; T, tesla; v-ELM, voting-ELM; ν-MKL, multiple kernel learning; VMHC, voxel-mirrored homotopic connectivity; WM, white matter.

Greenstein D, Malley JD, Weisinger B, Clasen L, Gogtay N. Using multivariate machine learning methods and structural MRI to classify childhood onset schizophrenia and healthy controls. Front Psychiatry. 2012;3. doi:10.3389/fpsyt.2012.00053

 Google Scholar

Castellani U, Rossato E, Murino V, et al. Classification of schizophrenia using feature-based morphometry. J Neural Transm. 2012;119(3):395–404. doi:10.1007/s00702-011-0693-7

 Web of Science ®Google Scholar

Iwabuchi SJ, Liddle PF, Palaniyappan L. Clinical utility of machine-learning approaches in schizophrenia: improving diagnostic confidence for translational neuroimaging. Front Psychiatry. 2013;4:95. doi:10.3389/fpsyt.2013.00095

 PubMedGoogle Scholar

Lu X, Yang Y, Wu F, et al. Discriminative analysis of schizophrenia using support vector machine and recursive feature elimination on structural MRI images. Medicine (Baltimore). 2016;95(30):e3973. doi:10.1097/MD.0000000000003973

 Web of Science ®Google Scholar

Pinaya WHL, Gadelha A, Doyle OM, et al. Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia. Sci Rep. 2016;6(1):38897. doi:10.1038/srep38897

 Web of Science ®Google Scholar

Xiao Y, Yan Z, Zhao Y, et al. Support vector machine-based classification of first episode drug-naïve schizophrenia patients and healthy controls using structural MRI. Schizophr Res. 2017. doi:10.1016/j.schres.2017.11.037

 Web of Science ®Google Scholar

Salvador R, Radua J, Canales-Rodríguez EJ, et al. Evaluation of machine learning algorithms and structural features for optimal MRI-based diagnostic prediction in psychosis. Hu D, ed. PLoS One. 2017;12(4):e0175683. doi:10.1371/journal.pone.0175683

 PubMed Web of Science ®Google Scholar

Amin MF, Plis SM, Chekroud A, et al. Reading the (functional) writing on the (structural) wall: multimodal fusion of brain structure and function via a deep neural network based translation approach reveals novel impairments in schizophrenia. Neuroimage. 2018;181:734–747. doi:10.1016/j.neuroimage.2018.07.047

 Web of Science ®Google Scholar

Pinaya WHL, Mechelli A, Sato JR. Using deep autoencoders to identify abnormal brain structural patterns in neuropsychiatric disorders: a large-scale multi-sample study. Hum Brain Mapp. 2019;40(3):944–954. doi:10.1002/hbm.24423

 Web of Science ®Google Scholar

Yoon JH, Tamir D, Minzenberg MJ, Ragland JD, Ursu S, Carter CS. Multivariate pattern analysis of functional magnetic resonance imaging data reveals deficits in distributed representations in schizophrenia. Biol Psychiatry. 2008;64(12):1035–1041. doi:10.1016/j.biopsych.2008.07.025

 Web of Science ®Google Scholar

Yang H, Liu J, Sui J, Pearlson G, Calhoun VD. A hybrid machine learning method for fusing fMRI and genetic data: combining both improves classification of schizophrenia. Front Hum Neurosci. 2010;4:192. doi:10.3389/fnhum.2010.00192

 Web of Science ®Google Scholar

Su L, Wang L, Shen H, Feng G, Hu D. Discriminative analysis of non-linear brain connectivity in schizophrenia: an fMRI study. Front Hum Neurosci. 2013;7:702. doi:10.3389/fnhum.2013.00702

 Web of Science ®Google Scholar

Zhu M, Jie N, Jiang T. Automatic classification of schizophrenia using resting-state functional language network via an adaptive learning algorithm. In: aylward S, Hadjiiski LM, eds. Int Soc Opt Photonics. 2014;9035:903522. doi:10.1117/12.2043240

 Google Scholar

Castro E, Gómez-Verdejo V, Martínez-Ramón M, Kiehl KA, Calhoun VD. A multiple kernel learning approach to perform classification of groups from complex-valued fMRI data analysis: application to schizophrenia. Neuroimage. 2014;87:1–17. doi:10.1016/j.neuroimage.2013.10.065

 PubMed Web of Science ®Google Scholar

Arbabshirani MR, Castro E, Calhoun VD. Accurate classification of schizophrenia patients based on novel resting-state fMRI features. Conf Proc Annu Int Conf IEEE Eng Med Biol Soc Annu Conf. 2014;2014:6691–6694. doi:10.1109/EMBC.2014.6945163

 Google Scholar

Watanabe T, Kessler D, Scott C, Angstadt M, Sripada C. Disease prediction based on functional connectomes using a scalable and spatially-informed support vector machine. Neuroimage. 2014;96:183–202. doi:10.1016/j.neuroimage.2014.03.067

 Web of Science ®Google Scholar

Cao H, Duan J, Lin D, Shugart YY, Calhoun V, Wang Y-P. Sparse representation based biomarker selection for schizophrenia with integrated analysis of fMRI and SNPs. Neuroimage. 2014;102(Pt 1):220–228. doi:10.1016/j.neuroimage.2014.01.021

 Web of Science ®Google Scholar

Matsubara T, Tashiro T, Uehara K. Deep neural generative model of functional MRI images for psychiatric disorder diagnosis. IEEE Trans Biomed Eng. 2015; 1-1. doi:10.1109/TBME.2019.2895663

 Web of Science ®Google Scholar

Koch SP, Hägele C, Haynes J-D, Heinz A, Schlagenhauf F, Sterzer P. Diagnostic classification of schizophrenia patients on the basis of regional reward-related FMRI signal patterns. Schwarz AJ, ed. PLoS One. 2015;10(3):e0119089. doi:10.1371/journal.pone.0119089

 Web of Science ®Google Scholar

Chyzhyk D, Savio A, Graña M. Computer aided diagnosis of schizophrenia on resting state fMRI data by ensembles of ELM. Neural Netw. 2015;68:23–33. doi:10.1016/j.neunet.2015.04.002

 Web of Science ®Google Scholar

Cabral C, Kambeitz-Ilankovic L, Kambeitz J, et al. Classifying schizophrenia using multimodal multivariate pattern recognition analysis: evaluating the impact of individual clinical profiles on the neurodiagnostic performance. Schizophr Bull. 2016;42(suppl 1):S110–S117. doi:10.1093/schbul/sbw053

 Web of Science ®Google Scholar

Kim J, Calhoun VD, Shim E, Lee J-H. Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: evidence from whole-brain resting-state functional connectivity patterns of schizophrenia. Neuroimage. 2016;124(Pt A):127–146. doi:10.1016/j.neuroimage.2015.05.018

 Web of Science ®Google Scholar

Liu Y, Zhang Y, Lv L, Wu R, Zhao J, Guo W. Abnormal neural activity as a potential biomarker for drug-naive first-episode adolescent-onset schizophrenia with coherence regional homogeneity and support vector machine analyses. Schizophr Res. 2017. doi:10.1016/j.schres.2017.04.028

 Web of Science ®Google Scholar

Guo W, Liu F, Chen J, et al. Using short-range and long-range functional connectivity to identify schizophrenia with a family-based case-control design. Psychiatry Res. 2017;264:60–67. doi:10.1016/j.pscychresns.2017.04.010

 PubMed Web of Science ®Google Scholar

Orban P, Dansereau C, Desbois L, et al. Multisite generalizability of schizophrenia diagnosis classification based on functional brain connectivity. Schizophr Res. 2017. doi:10.1016/j.schres.2017.05.027

 Web of Science ®Google Scholar

Wang S, Zhang Y, Lv L, et al. Abnormal regional homogeneity as a potential imaging biomarker for adolescent-onset schizophrenia: a resting-state fMRI study and support vector machine analysis. Schizophr Res. 2017. doi:10.1016/j.schres.2017.05.038

 Web of Science ®Google Scholar

Chen X, Liu C, He H, et al. Transdiagnostic differences in the resting-state functional connectivity of the prefrontal cortex in depression and schizophrenia. J Affect Disord. 2017;217:118–124. doi:10.1016/j.jad.2017.04.001

 PubMed Web of Science ®Google Scholar

Wang S, Zhan Y, Zhang Y, et al. Abnormal long- and short-range functional connectivity in adolescent-onset schizophrenia patients: a resting-state fMRI study. Prog Neuropsychopharmacol Biol Psychiatry. 2017;81:445–451. doi:10.1016/j.pnpbp.2017.08.012

 Web of Science ®Google Scholar

Bae Y, Kumarasamy K, Ali IM, Korfiatis P, Akkus Z, Erickson BJ. Differences between schizophrenic and normal subjects using network properties from fMRI. J Digit Imaging. 2017. doi:10.1007/s10278-017-0020-4

 Web of Science ®Google Scholar

Qureshi MNI, Oh J, Cho D, Jo HJ, Lee B. Multimodal discrimination of schizophrenia using hybrid weighted feature concatenation of brain functional connectivity and anatomical features with an extreme learning machine. Front Neuroinform. 2017;11:59. doi:10.3389/fninf.2017.00059

 Web of Science ®Google Scholar

Reavis EA, Lee J, Wynn JK, et al. Assessing neural tuning for object perception in schizophrenia and bipolar disorder with multivariate pattern analysis of fMRI data. Neuro Image Clin. 2017;16:491–497. doi:10.1016/j.nicl.2017.08.023

 Web of Science ®Google Scholar

Chen H, Uddin LQ, Duan X, et al. Shared atypical default mode and salience network functional connectivity between autism and schizophrenia. Autism Res. 2017;10(11):1776–1786. doi:10.1002/aur.1834

 Web of Science ®Google Scholar

Pläschke RN, Cieslik EC, Müller VI, et al. On the integrity of functional brain networks in schizophrenia, Parkinson’s disease, and advanced age: evidence from connectivity-based single-subject classification. Hum Brain Mapp. 2017;38(12):5845–5858. doi:10.1002/hbm.23763

 PubMed Web of Science ®Google Scholar

Liu Y, Guo W, Zhang Y, et al. Decreased resting-state interhemispheric functional connectivity correlated with neurocognitive deficits in drug-naive first-episode adolescent-onset schizophrenia. Int J Neuropsychopharmacol. 2018;21(1):33–41. doi:10.1093/ijnp/pyx095

 Web of Science ®Google Scholar

Zeng -L-L, Wang H, Hu P, et al. Multi-site diagnostic classification of schizophrenia using discriminant deep learning with functional connectivity MRI. EBio Med. 2018;30:74–85. doi:10.1016/j.ebiom.2018.03.017

 PubMed Web of Science ®Google Scholar