Research On The Diagnosis Method Of Depression Based On Deep Learning And EEG Signal

Depression is an imperceptible mental illness.With the development of society,the pressure of survival increases,and the number of depressed patients increases rapidly.How to quickly and accurately diagnose depression has become an urgent problem to be solved.Manual diagnosis is affected by many factors such as the doctor’s subjective judgment,the patient’s mental state,and the diagnosis environment,and it is not easy to obtain an objective and accurate result.As an electrical signal that can reflect brain activity,EEG signals can monitor brain activity continuously,non-invasively,and relatively cheaply.With the advancement of EEG acquisition devices and the development of artificial intelligence technology,EEG signals are increasingly being applied in the diagnosis of depression.However,the generalization ability and diagnostic accuracy of current methods based on single-channel EEG signals and traditional machine learning still need to be improved.In order to solve this problem,based on multi-channel EEG signals and deep learning,the EEG signal channel selection method and depression diagnosis model were studied,and experiments were carried out using multiple data sets.The results showed that the diagnostic efficiency and accuracy were greatly improved.The main research content of the thesis is divided into the following parts:(1)A conversion method for converting multi-channel EEG signals into two-dimensional image features was studied.The EEG signal recognition and classification problem is transformed into an image recognition and classification problem,and the automatic learning and feature extraction of deep learning are used to quickly diagnose image features.(2)A method for identifying the correlation between EEG signals and depression based on convolutional neural networks was studied.In order to simplify the process of extracting features,we select EEG signal channels that are strongly related to depression,use multi-channel EEG with markers as data,and divide the EEG signals of each channel into data sets for neural network training.In the case of a single channel,the classification performance of each channel on depression is compared,so as to select several EEG signal channels with the highest correlation with depression.Experiments show that the correlation between the signal of each channel of EEG and depression is not consistent,which verifies the effectiveness of the method and reduces the process of feature extraction and analysis.(3)A method for diagnosing depression based on multi-channel EEG signal fusion and cropping enhancement was studied.In order to utilize the multi-channel information of the EEG,the multi-channel EEG data is first fused and transformed into a two-dimensional image,then the multi-scale cropping technique is used to increase the number of samples in the data set,and then it is trained by a convolutional neural network.Experiments show that the combination of multi-channel fusion and multi-scale cropping can make full use of the information contained in multiple sensor records and significantly improve the classification accuracy of depression diagnosis.Moreover,the complexity is low and the robustness is high,which is beneficial to the wide-range application of the detection system.