Research On Recognition Method Of Abnormal Heart Rhythm Signal Based On Deep Learning

Cardiovascular and cerebrovascular diseases have always been an important hidden danger that threatens the lives and health of people all over the world,with a very high incidence,and have become a public and social problem that affects the health of the people and hinders economic development.It is for the prevention and treatment of cardiovascular and cerebrovascular diseases.Without delay.Arrhythmia is one of the most common cardiovascular diseases.There are two main manifestations of abnormalities:arrhythmia based on changes in a single heartbeat waveform and arrhythmia based on changes in multiple heart rhythms over a period of time.Have an adverse effect on the human body.In recent years,computer technology has continued to develop in the field of auxiliary diagnosis and treatment,and the study of automatic classification of ECG signals ushered in new opportunities.This paper mainly uses relevant knowledge in the field of machine learning to classify and identify the two manifestations of abnormal heart rhythm signals.It has certain reference value to assist doctors in the diagnosis and prevention of cardiovascular diseases.The research work of this article mainly discusses from the following aspects:(1)ECG signal preprocessing: ECG signal has the characteristics of weak signal,strong randomness,and poor anti-interference ability.Usually the collected ECG signal is accompanied by various noises,which provides a reliable data set for subsequent research.Through analysis The characteristics and sources of ECG signal noise.The median filter is selected to remove low-frequency noise such as baseline drift,and wavelet analysis is used to remove high-frequency noise such as EMG interference.The experimental results show that the two denoising methods can effectively remove the ECG signal.noise.(2)Classification and recognition of combined rhythms and multi-source premature beats based on traditional machine learning: For the automatic classification of abnormal heart rhythms based on changes in multiple heart rhythms over a period of time,there is no official database at present,so first start with the establishment of the database and pass the MIT-BIH arrhythmia database central electrical data R-R interval extraction and segmentation,and the use of sliding window method for data enhancement,to build a database suitable for the classification and identification of syntactic and multi-source premature beats,the database contains two joints of occasional premature beats and premature beats There are five types of abnormalities such as the rhythm,the triple rule of premature beats,two bursts of premature beats,and paroxysmal tachycardia;then three traditional machine learning models of KNN,Decision Tree and Random Forest are used for classification and recognition,and good classification results have been achieved.The Random Forest model with the best effect has achieved 97.39%,98.62%,99.09%,96.87%and 97.73% of the five types of abnormal arrhythmia signals.(3)Arrhythmia classification and recognition based on the 2D-CNN-BilSTM parallel network: For the classification and recognition of arrhythmia based on a single heartbeat waveform change,first map the one-dimensional ECG signal to a two-dimensional space,namely 2D-ECG,in order to convolve the nerve The network can better extract the spatial characteristics of the ECG signal.The experimental results show that 2D-ECG performs better on CNN than the one-dimensional ECG signal;then,combined with the advantages of the Long Short-Term Memory Network for time series processing,it is proposed2D-CNN-BiLSTM parallel network structure model,at the same time,the ECG signal is extracted in space and time sequence.Compared with a single CNN and BiLSTM Network structure,the experimental results show that it is effective against normal,left bundle branch block,and right bundle branch.The recognition accuracy rates of conduction block,ventricular premature beats,atrial premature beats and pacing heart beats reached 99.19%,99.21%,99.46%,99.07%,99.34% and 99.28%,respectively.