Research On Diagnosis Of Major Dispressive Disorder Based On Imaging Feature Learning

As a common worldwide mental illness,depression has become the leading cause of disability around the world and a major factor in the human disease burden.Therefore,the detection and diagnosis of depression has received more and more research attention.However,at present,the diagnosis of depression is mainly based on inquiry,scale evaluation and clinical manifestation diagnosis.There is a lot of professional inspection work,and the authenticity of patient feedback is difficult to be clear.Moreover,this method is subject to subjective influence.These factors make the diagnosis of depression require a lot of professional manual work and testing.In addition,most of the studies on the diagnosis of depression are based on the original image features and relatively simple classification methods,ignoring the noise in the data and the hidden features in different Spaces.Furthermore,while the incidence of depression increases rapidly,the diagnosis rate does not significantly improve.How to use a small number of samples to assist the diagnosis is an urgent problem to be solved.In order to solve the above problems,this paper proposes an auxiliary diagnosis method of depression based on Multi-pass Band Graph Convolution Fusion Network(MBGCFN).Based on non-invasive Functional Magnetic Resonance Imaging(f MRI)data,the semi-supervised diagnosis of depression was performed by extracting multi-pass band features from frequency domain space.Firstly,the quantitative value of the activity intensity of the whole brain region was obtained by using the white matter and gray matter prior brain atlases mapping of f MRI data,and the functional connectivity between brain regions was obtained by using Pearson correlation calculation.Then,the functional connections of the brain regions significantly different from those of the Control(Health Control(HC)and the patient with Major Depressive Disorder(MDD)were screened by Two-Sample t-Test according to P values.Then,the subjects’ multi-band information was extracted through two pass-band branches,low-pass and band-pass.This method can solve the over-smoothing problem of graph convolution.In addition,the adversarial generation module is innovatively introduced to guide the extraction of high discriminative information of samples.At the same time,the residual filter module is used to further remove the high frequency noise.Finally,the result fusion of multi-pass band learning is utilized as the decision basis to achieve the outcome of depression diagnosis.The experimental results show that the proposed method achieves good results in terms of accuracy,sensitivity and specificity.And this method is superior to other related methods.To explore the subjects’ brain significantly functional connection characteristics under the different space of potential information,and deal with the data in different stages and the relationship between the different performance characteristics,a method of depression diagnosis based on MultiStage Graph Convolution Fusion Network(MSGCFN)is proposed.the proposed method uses the theory of subspace clustering to extract features of samples at different stages.Then,based on these features at different stages,this paper innovatively uses data self-expressed attributes to build the affinity connection between samples and samples.Then,the graph convolution method is used to learn the samples features and graph topological structure of each stage.Finally,the information learned at all stages was fused to provide complementary decision information for the final classification of depression.The results show that the proposed MSGCFN in this paper can effectively classify depression and health control,and at the same time,it can significantly improve several indicators of the classification effect of depression.