Multimodal Mild Depression Recognition Based On EEG And EM

Compared with depression,mild depression is more common and may develop into moderate or severe depression over time,which brings greater pain and burden to patients.However,its diagnosis faces challenges such as lack of defined diagnostic criteria and strong subjective effects,and few studies have focused on the effective monitoring of mild depression.This study uses the Electroencephalography(EEG)-Eye Movement(EM)synchronous acquisition network to simultaneously record the EEG and EM signals of 20 mildly depressed and 19 normal control college students during the task of freely viewing emotional face pictures.The mild depression detection system to help doctors diagnose and monitor mild depression.Based on the autoencoder,we completed three parts of research.1.A unimodal mild depression recognition study based on EEG/EM features was performed.Unimodal EEG/EM features can effectively identify mild depression on Neu_block and Emo_block,and can achieve the highest classification accuracy of 77.37% and 74.84%,respectively.2.We carried out the multimodal mild depression recognition research based on EEG and EM features.This study uses two feature fusion strategies(Feature Fusion and Hidden Layer Fusion)to fuse EEG and EM signals,and uses the complementarity of different modes to improve the mild depression recognition performance.The results show that both fusion strategies can improve mild depression recognition accuracy,and the hidden layer fusion strategy performs more prominently.Feature Fusion and Hidden Layer Fusion achieved the highest classification accuracy of 81.74% and 83.93%,respectively.3.We conducted the cross modal mild depression recognition study based on EEG and EM features.We used two strategies(EM training,EEG training)for cross modal research,and explored whether the shared representation learned by the autoencoder has commonality in EEG and EM.The results show that both strategies can achieve effective cross modal mild depression recognition,and EEG training performance is more prominent.EM / EEG training achieves classification accuracy of more than 60%/70% on multiple classification algorithms and multiple EEG bands.The two strategies can achieve the highest cross modal classification accuracy of 70.29% and 75.32%,respectively.The research on cross modal mild depression has strong research value,especially when a certain mode is difficult to collect,expensive or a certain mode is missing,it plays a significant role.In this study,the effective features of EEG / EM and their correlation with depression were analyzed.The results showed that the unimodal/multimodal/cross modal recognition of mild depression based on EEG and EM might be the key to the early diagnosis of mild depression.The related research results provided theoretical basis for the auxiliary diagnosis and treatment of mild depression,and provided technical support for the research and application of recognition of mild depression based on EEG and EM.