Classification Of Depressive Disorder Based On Brain MRI

In recent years,the incidence of depression is increasing day by day,but the diagnosis of depression mainly depends on the experience of doctors,lack of objectivity,which leads to high missed diagnosis and misdiagnosis rate.MRI images can provide information about the anatomy of the brain or the state of brain activity.Therefore,using MRI images to classify depression can improve the objectivity of diagnosis and reduce misdiagnosis and missed diagnosis.However,due to the lack of certain biomarkers of depression,the difference in brain MRI images between depressive disorders and normal people is very difficult to distinguish and depression is also characterized by heterogeneity,so that using MRI images to classify depression is a great challenge.Under this background,this paper carried out the study of classification of depressive disorder based on brain MRI.The main contents is as follows:1.Preprocessing algorithm of MRI images.In order to achieve fixed-point quantitative analysis,for the morphological differences of different subjects,the resting state functional MRI(RS-fMRI)of different subjects were mapped into the standard template through head motion correction,segmentation,registration and normalization combined with the structural MRI(sMRI)of the same subjects.For the noise caused by non-brain activity,linear regression,Gaussian filter and band-pass filter were used to remove the noise.2.Feature extraction and feature selection algorithm.This paper studies five common features,which include low frequency fluctuation(ALFF),fractional ALFF(fALFF),degree centrality(DC),functional connection(FC)and regional homogeneity(ReHo),used in univariate statistical analysis methods.In view of the heterogeneity of depression and the characteristics of feature selection algorithm based on different rules,the feature selection method based on the Filter and the feature selection method based on the Embedded are respectively used to search the feature subset from the perspective of comprehensiveness and representation,which screens out valid feature for classifying depression.The experimental results have shown the effectiveness of the method.3.Classifier design and fusion algorithm.Firstly,different single classifiers and five common features were used to achieve depression classification and evaluate the discriminative ability of single features on depression at the individual level.Then,this paper designed classification algorithm based on single modal with multiple features to improve the discriminative ability for depression.From the perspective of feature fusion and multi-classifiers fusion,linear feature fusion and multi-classifier fusion based on decision fusion with voting mechanism are respectively used to fuse the selected feature subsets and single feature classification results.The experimental results have shown that all the five features can effectively distinguish depression with different feature selection algorithms,and ReHo has the best discriminative ability for depression.Both feature fusion and decision fusion can effectively improve the classification performance,but the effect of decision fusion is better than that of feature fusion.Especially,the best classification result will be obtained when the static voting mechanism is adopted.The classification accuracy is 90.22%and the sensitivity is 86.96%,which verifies the effectiveness of the algorithm.