A Novel Two-lead Arrhythmia Classification System Based On CNN And LSTM

Arrhythmia classification is a critical and meaningful task for assisting in diagnosis of heart diseases.And it has achieved great achievement in intra-patient paradigm.However,arrhythmia classification is still a tough problem when we consider it from the inter-patient point of view.The majority of previous works merely focus on the intrapatient condition and do not follow the Association for the Advancement of Medical Instrumentation(AAMI)standards.Most of the existing research methods for arrhythmia classification are at intra-patients,and have achieved relatively good results.But in this study of intra-patient,the same patient's data appeared both in the training set and in the test set most of the time.While in the inter-patient,patient data in training set could not appear in the test set.So inter-patient is more realistic.At present,most interpatient researches adopt the traditional method of feature extraction and there is still much room for improvement.Since ECG signal acquisition device in the most hospital can simultaneously collect multiple lead ECG signals,and the commonly used MITBIH database is composed of two lead ECG signals,this paper mainly studies based on the two-lead ECG signal.Since the two-lead ECG signal has spatial correlation and is time series,this paper designs a system of arrhythmia classification for the characteristics of the two-lead ECG signal.In order to compare with existing high-level papers,the research in this paper follows the AAMI standard for classification.Aiming at the existing problems of Intra-patient,a novel classification system of arrhythmia based on multi-lead electrocardiogram(ECG)signal is presented in this paper.The core of the system design is to fuse the two deep learning features with some common traditional features,and select the features through the binary particle swarm optimization algorithm(BPSO)to select significant distinguishing features.Then,the feature vector is classified by SVM with class-weight.For better generation of model and fair comparison,we carried out experiments with an inter-patient paradigm and followed the AAMI standards.Experimental results show that no matter with macro-metrics or micro-metrics,our system has the outstanding superiority than most of the state-of-the-art methods.