Research On Ecg Arrhythmia Classification Model Based On Deep Learning

As the pace of life continues to accelerate,the number of patients with cardiovascular diseases is gradually increasing.Arrhythmia is a significant early symptom of cardiovascular disease,and it is important to classify it intelligently for early diagnosis of cardiovascular disease.Electrocardiogram（ECG）has been widely used as an important basis for the diagnosis of arrhythmia,but the rapid increase of ECG has brought great diagnostic pressure on limited medical resources,so it is imperative to study the automatic classification of arrhythmia based on ECG signals.Traditional arrhythmia classification models rely on manually extracted features,and it is difficult to dig out deeper characteristic information of ECG signals.With the rapid development of artificial intelligence,the research and application of arrhythmia classification based on deep learning technology has become an inevitable trend.This thesis proposed two arrhythmia classification models based on the Deep Residual Shrinkage Network（DRSN）,and implemented arrhythmia classification software.The specific research contents of this thesis are as follows:（1）Aiming at the problem that the existing single network model cannot effectively extract the time-frequency domain features of ECG signal data and the signal still carries the noise feature information after the image is converted,a 2D-Deep Residual Shrinkage Network（2D-DRSN）based arrhythmia classification model is proposed.Firstly,the ECG signal is converted into the ECG time spectrogram by ShortTime Fourier Transform.Secondly,the constructed 2D-DRSN was used to extract the features of the time-frequency spectrum while filtering the noise characteristic information of the time-frequency spectrum,and realized the classification of five arrhythmia types.The results show that this model is better in each classification evaluation index and the the overall evaluation index.（2）To solve the problem of insufficient feature extraction from original ECG signal data with noise,a composite network model based on DRSN and bidirectional gate recurrent unit（DRSN-BiGRU）is proposed.Firstly,preprocessed ECG signal data without denoising is directly applied to the network model.Secondly,based on the advantages of DRSN fusion to extract waveform features and filter noise features,and BiGRU time domain feature extraction and analysis of ECG signal data,the model can achieve the classification of six arrhythmia types.Finally,compared with various arrhythmia classification models before and after noise reduction and other literature studies,the performance of the proposed DRSN-BiGRU model is better in all classification evaluation indexes and overall evaluation indexes.（3）According to the proposed DRSN-BiGRU network model,arrhythmia classification software was designed and implemented.The software has the functions of patient information management module,ECG signal reading module,ECG signal data preprocessing module and arrhythmia classification module.This thesis uses deep learning technology to provide a feasible idea for the design of arrhythmia classification model,and the application of deep residual contraction network to the diagnosis of arrhythmia diseases has broad prospects.