Disease Diagnosis Research Based On Cardio-brain Electrical Signal Characteristics

Human physiological signals are a window for understanding the state of human life and health information.The detection of physiological signals and research on information mining are of important value for understanding the basic laws of life,exploring the pathogenesis of diseases,timely diagnosing and preventing diseases.The two most widely used physiological signals in clinical practice are Electrocardiogram(ECG)signal and electroencephalogram(EEG)signal.As intuitive reflections of cardiac and brain electrical signals,ECG and EEG are important tools for diagnosing heart and brain diseases.In the past,the methods of visual ECG and EEG relied heavily on the experience of clinicians,and there was a possibility of misdiagnosis and missed diagnosis.Therefore,it is significance to develop a classification method that can identify abnormal ECG and EEG signals in time and accurately.This thesis mainly focuses on the construction of high-performance automated auxiliary detection methods for heart diseases represented by arrhythmia and brain nervous system diseases represented by epilepsy,in order to improve detection efficiency and reduce the workload of doctors.The main research contents of this thesis are as follows:(1)A deep learning method(SEN-Bi LSTM)with the characteristics of heartbeat and continuous heartbeat interval as dual inputs is proposed.Firstly,wavelet decomposition and moving average filter are used to denoise.Secondly,heartbeat segmentation is carried out,and the interval characteristics of heartbeat are extracted.After that,convolution neural network is used to extract spatial features,channel attention network is used to dynamically weight features,and the long short-term memory network captures the time features.On this basis,the characteristics of heartbeat interval and the extracted characteristics of heartbeat are fused,so that the disease information contained in the signal can be obtained from multiple dimensions.The proposed method is verified by 50% cross-validation on MIT-BIH arrhythmia dataset,achieving 99.68% accuracy,94.46% sensitivity,99.43% specificity and 94.98% positive value.(2)An epileptic seizure detection method based on specific EEG rhythm signal decomposition and chaotic dynamic feature extraction is proposed.Firstly,the fixed frequency range empirical wavelet transform filter bank(EREWT)based on EEG rhythm is introduced into the sub-band decomposition of EEG signals.Then,according to the chaotic characteristics of the epileptic brain,the sub-band signal trajectory is drawn by three-dimensional phase space reconstruction(3DPSR).According to Euclidean distance of 3DPSR,the line length,quartile deviation,logarithmic energy entropy and norm entropy are calculated.Finally,an integrated model based on different machine learning algorithms is used for classification.The performance of the proposed method is evaluated by using the epilepsy database of Bonn University.The results show that the overall average classification accuracy of this method for two types and three types of classification problems reaches 99.06% and 98.35% respectively.The system has good classification performance and low running time,and it has a wide application prospect in the automatic detection of epileptic seizures in clinical and portable epileptic devices.