Application Of Deep Neural Network Optimized Based On ELM-LRF And BLSTM In Ecg Recognition

Electrocardiogram(ECG)is an important tool for clinicians in diagnosis and pathological analysis of cardiovascular diseases.With the development of computer-aided medical technology,many algorithms for electrocardiogram classification have been developed.However,the accuracy of the existing classification algorithms in clinical diagnosis is not high,so it cannot be used as the final basis for judgement.Cardiovascular experts need to make the diagnosis again according to their own experience.Therefore,it is necessary to improve the accuracy of intelligent ECG classification.In this paper,the intelligent classification algorithms of ECG are designed and the main works are as follows:(1)This paper put forward a hybrid algorithm combining ELM-LRF and BLSTM network.Firstly,a one-dimensional ELM-LRF algorithm is designed for ECG.Then,by combining ELM-LRF algorithm with BLSTM,the ELM-LRF-BLSTM algorithm is proposed for ECG classification.Finally,simulation experiments conducted on the Chinese Cardiovascular Disease Database(CCDD)and MIT-BIH-AR database show satisfying results that the accuracy rates reach 84.46% and 99.32% respectively,higher than single ELM-LRF or BLSTM.This indicates that the ELM-LRF-BLSTM can combine the advantages of ELM-LRF and BLSTM network to improve the classification accuracy.(2)In order to further mine the correlation of time sequence,attention mechanism(AM)is introduced to ELM-LRF-BLSTM and proposes LRF-BLSTMAM algorithm.The classification accuracy rates reach 86.12% and 99.57%,respectively.Compared with ELM-LRF-BLSTM,LRF-BLSTM-AM reaches higher accuracy,which indicates that the AM enhances the abilities of feature expression,feature extraction and classification.(3)Random filters are used in ELM-LRF algorithm for feature extraction.Different types of filters can extract diverse features.Based on LRF-BLSTM-AM,this paper proposes AFLRF-BLSTM-AM algorithm based on adaptive filters.This algorithm uses four kinds of filters(random,Patch,PCA and WT filter)to construct a multi-channel feature extractor.Weights of each filter are determined on the importance degree so that to realize the adaptive weighted combination of filters.The algorithm realizes automatic classification of 5 types of arrhythmias on MIT-BIH-AR database and the accuracy rate reaches 99.89%.