Research Of Depression Recognition Method Based On Graph Convolution And Magnetic Source Imaging

Depression has gradually become one of the most common mental diseases.If patients with depression can get accurate diagnosis and timely treatment in the early stage of the disease,they can reduce the damage caused by depression.In recent years,many studies have tried to provide reference for the diagnosis of depression from the perspective of neuroimaging.At present,the research on brain state recognition and classification based on Magnetoencephalogram(MEG)usually focuses on designing various features from signals,and ignores the topological structure information of brain network,resulting in poor classification results.In addition,the individual differences of the subjects also seriously affect the generalization performance of the classification model.In order to solve the above problems,based on MEG,this paper integrates deep learning,complex network and source imaging,and carries out the following two aspects of research:Firstly,aiming at the problem that the topological structure information of brain network is not fully utilized,this paper proposes a brain function network classification method based on graph convolution.In this paper,MEG signals after artifact removal by independent component analysis are used to construct brain functional connections and brain functional networks,and Graph Convolution Network is used to process the topology information of brain networks,so as to realize the recognition of depression patients at the single-trial level.The results show that in the across-subjects verification framework,the proposed method can achieve an accuracy rate of 81% and a recall rate of 92%,which greatly improves the classification performance compared with some current baseline models.This shows that it is feasible to recognize depression in brain function network based on graph convolution.Secondly,aiming at the problem that MEG signal difference among different subjects affects the classification effect,this paper proposes a depression recognition method based on MEG source imaging.In this paper,we use source imaging technology to estimate the real source signals in the brain through MEG on the scalp surface,and use this to build a brain network as the classification basis.The source signal is not distorted by the skull,so it can alleviate the problem of too large individual differences among subjects.In addition,this paper also tries to combine the original signal with the source signal,so as to study the improvement of the recognition effect of depression by fusing internal and external brain features.The results show that the source signal can significantly improve the performance of the classification model compared with the scalp signal,and the method of combining intra brain and extra brain features also achieves better results than the single intra brain or extra brain signal.Based on the interaction between channels,this study established functional connections and brain networks as the basis for depression recognition,and realized the recognition of depression patients at the single trial level.At the same time,source imaging technology is used to alleviate the problem that MEG signals of different subjects are excessively distorted,and the classification effect is improved by combining the characteristics inside and outside the brain.