Research On ECG Signal State Recognition And Application Based On Nonlinear Analysi

With the accelerated pace of society,increased stress in life and the effects of poor lifestyle habits,various cardiovascular diseases are becoming more common and their incidence is increasing.The electrocardiogram(ECG)signal reflects the changing profile of cardiac information and can provide a strong basis for clinical diagnosis.The current methods of feature extraction for ECG signals are generally based on linear analysis,but ignore the non-linear characteristics of ECG signals.To address the above problems this paper adopts a non-linear topological analysis based method to analyse and apply research on ECG signals,and the main work is as follows:(1)Firstly,the processing flow of the nonlinear topological analysis method is studied,and its basic theory is described.This paper deals with two main aspects of nonlinear topological methods: the spatial embedding of the original ECG signal and the topological analysis of point cloud data.For spatial embedding,this paper investigates the construction method of delayed embedding of ECG signals;for topological dynamic analysis,the process of simple complex shape transformation and topological feature calculation is studied.(2)Based on the ECG signal data on Apnea-ECG dataset for sleep apnea classification detection task,a non-linear dynamics analysis method for extracting Heart Rate Variability(HRV)sequences from the ECG signal is proposed.The non-linear dynamics is studied by embedding the time series into the phase space using a phase space reconstruction method.In this work,the process of topological data analysis in sleep apnea detection is studied,and the topological kinetic features based on Shannon’s entropy are used for symptom detection and classification experiments,and the experimental process and results are analysed.(3)Based on the non-linear topological dynamics method proposed in this paper,it is applied to the state identification task of bradycardia detection,and it is used in the Preterm Infant Cardio-Respiratory Signals(PICS)dataset to conduct classification and identification validation experiments and analyse the results,and to compare with some existing methods.In the sleep apnea recognition task,the best recognition performance was achieved by topological feature learning,with a classification recognition accuracy of 88.9% for detection validation,outperforming the currently commonly used linear analysis methods.The accuracy of the validation results on the bradycardia recognition task was 94%,and for the first time a non-linear topological analysis method was applied to the detection of this condition.The results show that the non-linear topological dynamic analysis method has shown excellent recognition capability in all the classification and identification tasks of the ECG signal.The excellent discrimination of state changes in physiological systems by nonlinear topological analysis was demonstrated,reflecting the effectiveness of topological features in describing heart rate variability.