Research On Depression Recognition Method Based On Multi-Type EEG Features

With the rapid development of today’s society,people are prone to mental diseases in the face of high-intensity work pressure and fast pace of life.Among them,depression is the most common mental health disorder,which has seriously affected human health and normal life,and the prevalence rate is increasing year by year,but the treatment rate is far from enough.At present,the most widely used diagnostic methods of depression are Beck Depression Scale,patients’ self-report,doctors’ clinical experience or combination.However,the diagnostic results are easily affected by the subjective consciousness of patients and doctors.Early diagnosis can provide timely and effective decision-making for follow-up treatment.Therefore,it is of great significance to study objective and accurate identification methods of depression.Electroencephalogram(EEG)is a non-invasive functional neuroimaging tool,which can reflect human brain activity and working state.With the help of artificial intelligence technology,it can provide important biomarkers for the diagnosis of mental diseases such as depression,and has been widely used in scientific research.At present,the research results of depression recognition based on EEG signals are remarkable,but there are still many challenging problems,such as how to extract the features significantly related to depression from EEG data and provide more comprehensive information representation;the analysis methods of brain functional network are diverse,and there is no unified conclusion in the research results;the accuracy of depression recognition model constructed by machine learning algorithm needs to be improved.Based on the above analysis,this thesis takes EEG as the research object,takes machine learning and complex network theory as the research means,fully excavates the characteristics of multiple types of EEG data,explores the abnormal brain network topology of patients with depression,constructs a stable model through optimization strategies,and improves the recognition accuracy,in order to provide reliable technical methods and EEG physiological indicators for the screening and auxiliary diagnosis of depression.The main work and innovations of this thesis are as follows:1.In view of the redundancy and false connection of high-dimensional brain functional connection features,and the traditional feature selection methods ignore the fuzzy relationship between samples and category markers,a functional connection feature selection method based on fuzzy label is proposed,which makes full use of the semantic information hidden in category markers to guide the feature selection method to obtain discriminant feature subsets.The functional connection matrix is constructed by calculating the association relationship between electrode pairs,initializing the cluster center with the same number of actual categories,calculating the local distance from the sample to the cluster center as the membership degree of a category,taking the membership matrix as the fuzzy label,using the sparse regression model to select the features most related to the fuzzy label as the final feature subset,and inputting it into the classification model for depression recognition.By comparing different feature selection methods,it is proved that this method extracts effective EEG features,so as to improve the recognition accuracy.2.In view of the low spatial resolution of EEG signals and the fact that the coupling method in the construction of brain functional network is vulnerable to the influence of volume conduction and can not accurately quantify the strength of functional connection,a brain functional network construction and analysis method based on common nearest neighbor is proposed.Based on the functional connection matrix,the common nearest neighbors with the highest degree of association with the two electrodes are found,and then the nearest neighbor relationship is fused to form a fuzzy set.Then the lattice closeness of fuzzy set is used to re measure the relationship between electrodes.Further,the functional connection matrix is transformed into a binary matrix,the network attributes are calculated,the functional connection value and network attributes are used as the characteristics for depression identification,and the characteristics of brain functional network with significant differences between the two groups are analyzed by statistical test method.This method can reflect the local information of cooperative changes between electrodes,improve the spatial resolution of EEG signals,reduce the influence of volume conduction,and provide an objective method for the diagnosis of depression.3.Aiming at the instability of EEG signal and individual classification model,a multi classifier fusion method based on differential evolution algorithm is proposed.Firstly,multi type features are extracted from nonlinear and non-stationary EEG signals,and then individual classifiers are constructed by different classification algorithms.The individual classifiers are optimized and weighted by differential evolution algorithm,and the final classification results are obtained by combining the weighted results.Experiments show that the depression recognition model constructed by this method effectively improves the classification accuracy.At the same time,nonparametric displacement test is used to statistically analyze the discrimination ability of different types of characteristics for depression,and electrodes and brain areas that may be more closely related to the abnormalities of patients with depression are found,which provid valuable information for the clinical diagnosis of depression.4.In view of the subjectivity and arbitrariness of binarization methods and the problem that traditional methods rely on manual feature extraction,a depression recognition fusion model based on deep learning is constructed,and a new binarization method based on common nearest neighbor is proposed.Different coupling methods and binarization methods are used to obtain brain network topology,and the input graph convolution neural network automatically extracts potential characterization features.The learned feature matrix is then input into the long-term and short-term memory neural network to store the changes of electrode channels in the time period,extract the time features,and finally use the linear layer to obtain the classification results.This method uses the deep learning fusion model to automatically extract features,makes full use of the topology of EEG signal channel,makes quantitative analysis of adjacency matrix,and looks for a more effective combination method,which is helpful to improve the accuracy of EEG feature-based depression recognition model.In conclusion,based on the neuropathological mechanism of depression and the actual needs of clinical auxiliary diagnosis,this thesis designs a depression recognition model that can effectively solve the existing research problems.In order to verify the effectiveness of the method and further apply it to real cases in hospitals to assist doctors in the diagnosis of depression,this study has practical significance.