Research Of Depression Detection Method Based On EEG Signals

Depression is a common mental illness,which usually leads to a series of symptoms such as depression,reduced self-worth,decreased interest and happiness,and sleep disorders.Severe depression can even lead to self-harm and suicide.With the acceleration of the pace of life and the increase of work pressure,the incidence of depression is increasing year by year,and it is expected to become the main factor of social disease burden in 2030.Because the pathogenesis of depression is not clear,clinical practice is mainly based on mental health self-rating scale and other means,relying on the clinical experience of doctors to achieve subjective qualitative diagnosis and treatment.At present,the diagnostic methods of depression are relatively backward,and it is difficult to make early diagnosis and quantitative and accurate diagnosis and treatment.In addition,due to people ’s aversion to medical treatment for mental illness,depression cannot be detected and treated as soon as possible.Based on modern scientific and technological means such as artificial intelligence and the latest research results of physiological psychology,it is of great clinical significance and social value to explore the objective and effective diagnostic methods of depression and realize the quantitative and accurate detection of depression.Studies have shown that depression can lead to lesions in the structure and function of brain regions.As the most direct embodiment of brain neural activity,EEG can quantitatively describe the brain state of patients with depression,and the characteristics of portability and easy access make it easier to carry out depression-related research.At present,researchers have established a depression detection model based on machine learning methods,and evaluated depression status after extracting EEG signal features.However,such methods require manual extraction of relevant features,and often require feature importance assessment and screening.In addition,most of the current research is based on the detection model to carry out cross-subject experiments in the data set,and there is a lack of cross-data set classification research based on EEG.Based on this,this paper studies the above problems,collects the resting EEG data of 80 subjects,integrates different deep learning models to model and analyze EEG signals,and conducts cross-dataset analysis with the public data set MODMA to analyze the model ’s ability to eliminate data set differences.Specifically,the main research work of this paper is as follows:(1)Design experiments to obtain EEG data and process them.In this study,the EEG acquisition scheme was optimized according to the physiological and pathological characteristics of depression.The resting EEG data of 40 patients with depression and 40 healthy controls were obtained for 5.5 minutes,and the research data set RMEEG was established after preprocessing such as denoising and resampling.At the same time,in order to explore the robustness of the depression detection model and analyze the impact of data set differences on the research results,this study used the three-lead EEG signals of the public data set MODMA as the basic data for comparative research.(2)The EEG-based depression detection model LSDD-EEGNet and its improved form LSDD-GCNNet are proposed.The LSDD-EEGNe model combines the advantages of convolutional neural network,long short-term memory network and domain discriminator.The convolutional neural network module is used to learn the local features of EEG signals,the long short-term memory network extracts the temporal information,and the domain classifier solves the distribution difference between different sample data.In order to quantify the degree of connection between multiple EEG channels,this paper introduces the graph convolution network module into LSDD-EEGNet,and constructs the LSDD-GCNNet model.The size of the model parameter adjacency matrix represents the degree of association of each electrode,so as to quantify the functional connection of EEG channels.(3)Based on the depression detection model,EEG recognition research was carried out on RMEEG and MODMA datasets.First,this study analyzes the performance of the depression detection model in each frequency band of the two datasets.The experimental results show that the proposed model not only outperforms the comparison model in all frequency bands of the RMEEG dataset,but also performs well in all frequency bands of MODMA.After that,this study quantified the degree of connection between EEG channels based on the LSDD-GCNNet model.The results showed that the connection between Fpl and Fz(Fpz)was stronger than that between Fp2 and Fz(Fpz),indicating that there was asymmetry in EEG signals in the prefrontal lobe.(4)Based on the depression detection model,the cross-dataset EEG recognition of RMEEG and MODMA was studied.Firstly,this study uses the depression detection model to carry out cross-dataset experiments.The results show that the classification performance of this model is better than other models,indicating that the model has a certain ability to eliminate data set differences.After that,this study also designed ablation experiments to analyze the effectiveness of each module of the LSDD-GCNNet model.The experimental results show that although the introduction of the graph convolution model will increase the model analysis ability to a certain extent,it will also make the model unstable,while the three sub-modules of CNN,LSTM and domain discriminator can steadily improve the depression recognition ability of the model.(5)The classification performance of the depression detection model is verified by visualization.After proposing the depression detection model,this study not only carried out cross-subject research in two datasets,but also carried out cross-dataset research on two datasets to further verify the effectiveness of the model.In these two parts of the study,this paper uses the tSNE visualization tool to analyze the characteristics of different frequency bands extracted from the model.The results show that the research method in this paper can extract good EEG features under full-frequency EEG,and can provide technical support for the auxiliary diagnosis of depression to a certain extent.