Research Of Depression Recognition Algorithm Based On Physiological Signal

Objective: Depression is a common mental disorder with high disability rate,high suicide rate and high relapse rate.It is listed as the fourth major disease in the world,which seriously endangers people’s physical and mental health.At present,the diagnosis of depression mainly relies on systematic consultation with professional psychiatrists combined with a series of psychiatric examinations.However,this method is highly subjective and has a high rate of missed diagnosis and misdiagnosis.It is urgent to find a convenient,objective and efficient method to assist the diagnosis of depression.With the continuous development of physics,biomedicine,computer science and other disciplines,the deep learning method to assist diagnosis of depression based on physiological signals has become a research hotspot.Physiological signals collected in a standardized manner can reflect the physiological and pathological changes of patients,and can be used as a reference basis for disease diagnosis,which can be quantified as evaluation indicators,and can realize the transformation of disease diagnosis from subjective evaluation to objective evaluation.Electroencephalogram(EEG)is a spontaneous bioelectric signal generated by brain neural activity,which contains abundant brain activity information,and is an important basis for brain function research and brain disease diagnosis.Speech signal is a kind of non-electrophysiological signal,carrying the speaker’s own physiological and psychological characteristics.Based on the above theoretical basis,this thesis uses EEG and speech signals to study the auxiliary diagnosis algorithm of depression based on deep learning methods,in order to provide effective auxiliary methods for clinical depression diagnosis.Methods and results: Aiming at the problems of unreasonable modeling and low overall performance of the traditional EEG diagnosis algorithm for depression.A novel depression recognition algorithm based on graph attention network is proposed.Firstly,in order to construct the correlation between brain regions more reasonably and explore the deep connections between brain regions,the electroencephalogram was constructed by calculating the Euclidean distance between electrodes.Secondly,the graph attention mechanism is used to build the graph representation learning model.Finally,the recognition and classification task of EEG signals in patients with depression was realized through the fully connected layer.The experimental results on the data set show that the accuracy,sensitivity,specificity and F1-Score indexes of the proposed method are 78.29%,76.12%,82.78% and 77.23%,respectively.Compared with the comparison method,this method can improve the recognition rate and classification performance of EEG signals in patients with depression.Ablation experiments show that the EEG pattern construction method and the feature extraction method based on the graph attention mechanism can effectively improve the recognition performance of the algorithm.In order to solve the problem of low performance of depression recognition algorithm based on single mode signal,this paper studies depression recognition algorithm based on multi-mode physiological signal.The advantage of depression recognition algorithm based on multi-modal physiological signals is that it can better combine the characteristics of different modal data to complement each other and give play to the advantages of different modal data.In addition,this study makes the sample data better reflect the time and frequency characteristics of the signal through the time-frequency transformation of the signal.This study designs a feature extraction module of speech signal time-frequency graph based on Swin Transformer and a multi-scale feature pyramid module for feature extraction of speech signal time-frequency graph.Image attention network was used to extract EEG features.Then the feature fusion module was used for feature fusion,and finally the depression was identified through the full connection layer.The experimental results on the data set show that the accuracy,sensitivity,specificity and F1-Score of the proposed method are 81.24%,80.24%,83.91% and 80.63%,respectively.The comparative experimental results effectively prove the superior performance of depression recognition algorithm based on multi-modal physiological signals.Ablation experiment and additional validation experiment results demonstrate the superior performance of the proposed Swin Transformer based feature extraction module and multi-scale feature pyramid module in feature extraction and fusion.Conclusion: In order to assist the diagnosis of depression,this thesis proposes an EEG depression recognition algorithm based on graph attention network and a depression recognition algorithm based on multi-modal physiological signals.Among them,EEG depression recognition algorithm based on graph attention network provides a new idea for the construction of the relationship between different brain regions.The depression recognition algorithm based on multi-modal physiological signals plays a certain role in promoting the research of depression auxiliary diagnosis from the perspective of multi-modal signals and data.Moreover,the proposed algorithm has a good classification accuracy,and realizes the optimization and improvement of the overall performance compared with the existing methods,which has certain reference value for the auxiliary diagnosis of clinical depression.