EEG Classification Of Disturbance Of Consciousness Based On Convolutional Neural Network

Disorders of consciousness is a disease that suppresses or even loses the consciousness of the human brain.Accurate diagnosis of patients with disorders of consciousness can provide great help for subsequent treatment.Electroencephalogram(EEG)is a commonly used diagnostic aid for disorders of consciousness in clinical practice.It can be divided into task-state EEG and resting EEG.The study and acquisition of task-state EEG requires designing tasks and providing stimulation to patients,while resting EEG does not require designing tasks and is more convenient.At present,traditional methods such as power spectrum and entropy value are mostly used in the analysis and research of resting EEG,while deep learning is rarely used,which performs well in many classification tasks.Therefore,the deep learning method was used in this study to classify EEG signals with resting state disorders of consciousness.In cooperation with Zhejiang Provincial Armed Police Corps Hospital,the EEG data of 153 patients with disorders of consciousness were collected.After data preprocessing,we used convolutional neural network and cyclic convolutional neural network to classify the data of disorder of consciousness.The work contents are as follows:(1)This thesis firstly constructs the CNN model by referring to the existing EEG processing models.According to the characteristics of small data set,it constructs the compact convolutional neural network model — — EEGNet model,which has generally good EEG processing effect,and improves the EEGNet model according to the sampling duration and dimension.The classification accuracy of CNN model and EEGNet model are 75.06% and 85.10%,respectively.(2)The cyclic neural network LSTM model is constructed.Due to the general effect,a cyclic convolutional neural network model CRNN was constructed by combining the convolutional neural network and GRU model.Then,the CRNN model is improved,and the classification accuracy of the improved compact connection model based on Inception-CRNN reaches 86.22% and the AUC value reaches 0.9259 on the test set.At the end of the study,KNN and SVM,which are commonly used machine learning algorithms,were used for experiments,and their accuracy rates were 65.45%and 70.94%,respectively.The proposed compact junction model based on InceptionCRNN has the highest classification accuracy and can effectively solve the classification problem of EEG with resting state consciousness disorder,which is expected to become a tool for clinical diagnosis.