Research And Implementation Of ECG Automatic Classification Model Based On Convolutional Neural Network

Cardiovascular disease has been the one of the leading causes of death in the world during the past decade,so it’s of great importance to monitor the patients of cardiovascular disease.While for the limited medical resource to the patients,it’s hard to maintain the quality of medical care for the patients.At present,electrocardiography(ECG)is the most commonly used clinical detection to diagnose cardiac diseases.There are also many studies of using ECG on the automatic arrhythmia detection system have been proposed.However,most of the literature research are based on ECG digital data for ECG filtering,segmentation,classification and diagnosis.Few studies are based on ECG images for automatic diagnosis.Meanwhile,in addition to the online open source ECG database,the data of major manufacturers of ECG monitors is encrypted.Patients and medical staff can only see the printed ECG images,but can not obtain the decrypted ECG digital data.Therefore,this paper intends to combine deep learning with image processing to automatically identify,segment and classify ECG waveforms in the absence of ECG digital data.In this paper,the main research work is divided into the following parts:(1)Automatic extraction of ECG waveforms.The ECG curve needs to be separated from the background grid of ECG before the classification of heartbeat.This paper proposes a comprehensive algorithm combining Gamma transform and OTSU algorithm to separate the curve.Then,the extracted ECG curve is refined through the skeleton refinement algorithm.(2)Constructing automatic heartbeat segmentation and classification model for ventricular premature and normal heartbeat.In this paper,a more intelligent convolutional neural network classification model is proposed.After the extraction of the ECG curve,the heartbeat can be segmented and classified without any further processing.(3)Evaluating and optimizing the performance of the model by applying the model to clinical trials.Most of the experiments are based on standard database such as MIT-BIH database and lack of clinical trials and evaluation.In this paper,the FZU-FPH clinical database and thousands of paper-based ECG obtained from hospital are used to verify and optimize the algorithm,which proves the practicability of the algorithm.From the experimental results,the average accuracy of classification in normal and ventricular premature heartbeat is 98.25% and 99.73% based on MIT-BIH and FZU-FPH,respectively.Besides,the proposed method can extract ECG curve waveform,and accuracy of classification based on clinical scanned ECG can reach 85.57%.Compared with the existing classification methods,the method proposed in this paper is more intelligent,gains more clinical feasibility and has good generalization ability.