Research On Clinical Application Of EEG Signal Based On Time Series Neural Network

Electroencephalogram(EEG)is a technology used to detect electrical signals in the brain cortex,which can reflect the physiological state of the human brain,and is often used to detect and assist in the diagnosis of brain diseases,such as epilepsy.Due to the highly random nature of EEG waveforms,it is difficult for humans to directly discover its correlation with brain diseases.With the improvement of people’s health needs and the development of artificial intelligence technology,researchers have proposed various EEG signal processing and analysis methods,providing more technical means for clinical medicine and brain research.In this thesis,the time series neural network is used to process and analyze the EEG signal,and the two clinical problems of depression diagnosis and anesthesia depth monitoring are carried out.Major depressive disorder is a serious psychiatric disorder in which patients are often extremely misanthropic.According to statistics,nearly one million people worldwide die by suicide due to depression every year.Therefore,the identification and treatment of depression has become the focus of medical research.At present,the diagnosis of depression is largely dependent on self-assessment scales and psychiatrist assessments,which are highly subjective,resulting in a low accuracy of depression diagnosis.In recent years,researchers have proposed many classification methods of depression based on EEG and machine learning algorithms,but the recognition accuracy still needs to be improved.Anesthesia depth monitoring is one of the important life-sustaining monitoring methods in clinical operations.Insufficient or excessive anesthesia may have serious effects on patients,such as intraoperative awareness and spontaneous breathing disorders.Clinical indicators such as blood pressure,heart rate,and respiratory rate can be used to estimate the depth of anesthesia.However,due to differences in patient physique and anesthetic drugs,the performance of these indicators varies greatly,so it is unreliable to analyze these clinical indicators alone.In view of the above background and problems,this thesis proposed a time-series neural network-based framework for depression classification and anesthesia depth monitoring,and conducts simulation experiments and analysis.The main work of the thesis includes:1.We propose a classification framework for depression based on EEG and gated recurrent units.The original single-channel EEG is denoised by various preprocessing methods,its differential entropy features are extracted,and the GRU network is used for feature classification.In order to verify the performance of the framework,this thesis builds frameworks with different EEG characteristics and different network models for comparative experiments,and evaluates the models.Experiments show that the classification framework proposed in this thesis can improve the accuracy of depression identification under the premise of meeting the actual clinical significance.2.An anesthesia depth monitoring framework based on EEG and Transformer encoder is proposed.First,the original single-channel EEG is preprocessed to extract spectral and differential entropy features,and the two types of features are fused and sent to the Transformer encoder network to complete the prediction of anesthesia depth.Experiments show that the feature fusion of wavelet spectrum and differential entropy combined with Transformer encoder network has the best prediction performance.3.A portable embedded anesthesia depth monitoring system based on single-channel EEG was designed.The construction of the system mainly includes software and hardware integration and visual interface design.The system first collects EEG signals,completes preprocessing operations such as denoising,and finally performs real-time prediction of anesthesia depth in combination with the previous deep learning model.