Support Vector Machine Classification Of Major Depressive Disorder And Bipolar Depression Based On Resting-state Functional Magnetic Resonance Imaging

Objective:To explore the significance of support vector machine(SVM)classification of major depressive disorder and bipolar depression based on resting-state functional magnetic resonance imaging(rsf-MRI),evaluate the performance of the classification model,which seek imaging basis for the classified diagnosis of major depressive disorder(MDD)and bipolar disorder(BD).Materials and methods:79 patients with MDD and BD(including BD-I and BD-II)who met the diagnostic criteria of Diagnostic and Statistical Manual of Mental Disorders(fifth Edition,DSM-V),were recruited for the first time in the Department of Psychiatry,Second Xiangya Hospital of Central South University.and collected demographic information and the score of Beck Depression Inventory(BDI),These patients underwent resting f MRI before treatment.According to the clinical diagnosis,the patients were divided into major depression disorder(MDD)and bipolar depression(BD)groups for analysis.Taking regional homogeneity(ReHo),amplitude of low frequency fluctuation(ALFF)and degree centrality(DC)as characteristic parameters,the MDD and BD patients were classified by using the machine learning toolkit PRo NTo based on MATLAB,SPM function library and libsvm function library,and the effects of different characteristic parameters and parameter combinations on the classification effect were compared,To provide neuroimaging reference for the clinical classification of patients with depression.Results:ReHo classification accuracy reached 89.19%,followed by DC and characteristic combination of ALFF+DC,ReHo+DC,ALFF+ReHo+DC,the classification accuracy is 81.08%.Finally,the classification accuracy of ALFF+ReHo and ALFF were 79.73% and75.68%,respectively.The brain regions with the largest classification weight in the optimal feature ReHo classification model mainly include frontal lobe,occipital lobe and cerebellum.Conclusion:ReHo has the best effect on the classification of unipolar and bipolar depression in rsf-MRI features,and ReHo has the best classification effect among the resting state functional imaging features.The characteristics of frontal lobe,occipital lobe and cerebellum are highly sensitive to the classification and diagnosis of depression.This study suggests that regional brain region ReHo can be further studied as an adjunct diagnostic tool to differentiate MDD from BD in a more efficient manner.