Study On The Key Problems Of Intelligent Detection Of Imbalanced Arrhythmia

Electrocardiogram(ECG)is the most commonly used non-invasive diagnostic tool to record the physiological activity of the heart over a period of time.It helps diagnose many cardiovascular diseases,such as premature atrial contraction(PAC),premature ventricular contraction(PVC),atrial fibrillation(AF),and myocardial infarction(MI).In recent years,with the rapid development of portable ECG monitors in the medi-cal field,such as Holter,and wearable devices in the medical field,such as the Apple Watch,the amount of ECG data has grown rapidly.Therefore,how to automatically and accurately analyze ECG data has become a research hotspot for many years.An effective arrhythmia detection algorithms are proposed,aiming at the current data dis-tribution in the ECG database and the characteristics of the ECG data itself,based on the imbalanced data processing technology and deep learning(DL)technology,accord-ing to the ECG data distribution characteristics of the MIT-BIH Arrhythmia Database and the characteristics of the ECG signals of the China Physiological Signal Challenge2018(CPSC 2018),around the key problems and optimization strategy of the automatic classification of arrhythmia.The main research contents of this paper are summarized as follows.(1)An arrhythmia automatic classification model based on long-short term mem-ory(LSTM)neural network model is proposed.In view of the complex changes and imbalance characteristics of ECG signals,LSTM networks are used to extract the timing characteristics of complex ECG signals,and focal loss function are used to alleviate the gradient optimization problems caused by the imbalance of ECG data.The proposed method improves the recognition rate of a small number of abnormal ECG signals.Fi-nally,the validity of the model is verified using the MIT-BIH Arrhythmia Database?(2)A multi-channel convolutional neural network(MC-CNN)arrhythmia auto-matic classification model is proposed.Aiming at the characteristics of the 12-lead short-term ECG signal with different durations,a 12-channel CNN is designed to pro-cess the ECG data of 12 leads,and each channel is combined with overlapping slice technology to build a CNN network model.The MC-CNN network model realizes the identification of multi-lead variable-length ECG records.Finally,the short-term ECG dataset of the China Physiological Signal Challenge 2018 is used to verify the effec-tiveness of the proposed network.