Research On Automatic Classification And Diagnosis Of ECG

With the extensive application and development of massive medical data storage technology and artificial intelligence technology in the field of medicine,more and more scholars have made research on automatic diagnosis and classification based on ECG signals and achieved some results.Because ECG data is easy to collect and has a more intuitive regularity,it is one of the earliest biomarkers to be studied and applied in clinical cardiovascular disease diagnosis in the development of medical field.This paper mainly studies the data preprocessing,feature extraction and optimization,model construction and optimization of ECG automatic diagnosis and classification.The main work includes the following parts:1)A modified evolutionary scheme for ECG signal preprocessing is developed.This part mainly includes noise processing,R-peak location and period segmentation.Firstly,a set of ECG noise processing scheme is designed.The combination of unbiased likelihood threshold estimation and improved fixed threshold estimation is used to set the threshold dynamically.According to the improved threshold processing function,the high-frequency components are denoised by the improved wavelet threshold.Then the high frequency component and low frequency component after denoising are reconstructed to obtain the final denoising signal.Secondly,we design and implement a correction algorithm to correct the R-peak position in the read MIT-BIH database,so as to improve the accuracy of R-peak location.Finally,the cycle of ECG signal is intercepted according to the R peak position to complete the preprocessing of ECG signal.2)The scheme of feature extraction and optimization which can accurately describe ECG signal is developed.Firstly,the initial time-domain features are determined,and then the wavelet coefficients after wavelet transform are taken as the initial frequency-domain features according to the energy distribution of ECG signals.Secondly,Then,the scheme of feature optimization is designed and implemented.The principal component analysis and independent component analysis are combined to transform the time-domain and frequency-domain initial features respectively,and the statistical independent time-domain feature matrix and frequency-domain feature matrix after dimension reduction are obtained.Finally,the time-domain and frequencydomain features are fused to obtain the initial feature matrix.3)A better fuzzy decision tree model based on ECG is developed.Firstly,the fuzzy representation method of fuzzy set theory and the comprehensibility of decision tree algorithm are combined to create a dynamic membership function and a fuzzy decision tree model.Then,the optimization scheme of the model is designed according to the relevant parameters of the fuzzy decision tree.the necessary parameters needed to construct the fuzzy decision tree are used as the particles in the particle population for iterative optimization.A dynamic fitness function is created to obtain the optimal parameter combination.Finally,according to the optimized parameters,the optimal fuzzy decision tree model is output to complete the construction and optimization of the classification model.The above scheme is tested on MIT-BIH arrhythmia database to verify the validity of time-frequency characteristics and the fuzzy decision tree model based on Particle Swarm Optimization,and the classification accuracy is 97.64%.Therefore,the research of this paper can be used as the basis of the research of ECG automatic diagnosis and classification system.