Study On Depression Detection Method Based On Frontal Three-lead EEG

Depression is a common disease that affects people’s mental health in the world.It not only affects the quality of life of patients,but also may cause suicides.Therefore,it is important to diagnose depression in a timely and effective manner.The main diagnosis method of current depression is face-to-face communication between doctors and patients,so as to obtain a subjective diagnosis.However,the patient’s subjective statement is often not accurate enough,which may lead to misdiagnosis,so an objective method is needed to provide doctors with auxiliary diagnosis.EEG can be used for disease diagnosis because it directly reflects the information transmission and status of the human brain.In addition,in order to overcome the shortcomings of complicated and time-consuming operation of multi-lead EEG,and considering that depression can cause abnormal activity of the prefrontal lobe of the brain,this paper designed a classification model of depression based on the frontal three-lead EEG,which can achieve depression quickly and conveniently Detection.Based on the frontal lobe three-lead EEG data,this paper used different methods for the classification of depression based on its time domain,frequency domain and spatial information between leads.Among them,this research mainly used Convolutional Neural Network(CNN)to build a depression classification model and conducted exploratory research.The main work and innovations of this article are as follows:1.In the previous work of EEG classification,features were extracted based on a single lead signal,ignoring the spatial information of different lead EEG.In this study,learning classification model based on CNN was designed,using three-lead EEG data as the input set.And the spatial information features were extracted by using the characteristics of CNN local convolution.Then,six models of different depths were built,and through performance verification,the best nine-layer CNN was selected as the simulation model.In addition,this paper also calculated multiple EEG features,and used support vector machine(SVM)as a control model to compare the performance of the two on multiple indicators such as classification accuracy.The results showed that the CNN model designed in this study can better distinguish the two types of data,and the test accuracy rate reached 81.88%.2.Considering that the EEG waveform has morphological characteristics,and the EEG waveform of patients with depression has more slow wave components than the normal person,this study proposed a method of converting three-lead EEG data into time-domain waveform picture data.Besides,in order to highlight the waveform in the picture,the picture is grayed.Then,a 20-layer CNN suitable for EEG image data was designed,and three classic architecture CNNs were built for simulation analysis.The final test accuracy rate reached 84.60%.3.Because depressive patients have wave power values rising and wave power values falling in the frequency domain compared with normal people,this study constructed spectrum waveform image data as well as waveform image data on time domain.In addition,this study proposed a feature processing module that used CNN to extract features and feature fusion algorithm to fuse the time-domain and frequencydomain features extracted by the two data in three different stages of CNN.Using SVM and KNN classifiers to test the classification effect of the three fusion feature sets,it was found that compared with other feature sets,the feature set after three fusions had a higher correlation with the actual label and better classification performance.In addition,for the feature set 3 after three fusions,this paper used CNN,SVM,KNN three models for experimental simulation and comparison.The final test accuracy rate of CNN reached 86.21%,which is better than the other two methods.Based on the above research,in this article: Firstly,the research method of tri-lead EEG classification based on end-to-end network was explored;Secondly,the time and frequency domain information of EEG was converted into their respective waveform image data,which improved the classification rate of depression and normal control;Thirdly,combined with feature fusion,the research method of using CNN as a feature extraction module and multiple models as classifiers was explored.Through the analysis of the results,the method designed in this study can provide new ideas for the auxiliary diagnosis of depressed patients.