An ECG Intelligent Diagnosis System Based On Cloud-edge Computing

As people pay more and more attention to their own health,home medical care has become a development trend.Heart diseases’prevention and diagnosis as an indispensable part of home medical care,its first condition is to achieve fast and accurate ECG diagnosis.At present,it is impossible to diagnose a large number of ECGs within a limited time by manual diagnosis,and the establishment of complex mathematical models through machine learning technology to achieve ECG intelligent diagnosis requires high equipment performance,which is difficult to meet the requirements of different medical fields.In view of the above problems,this paper proposes an ECG intelligent diagnosis system based on cloud-edge computing which combines the complementary advantages of edge device（close to data sources and having certain computing power）and cloud server（low cost,powerful computing power and flexible resources）,and then carries out ECG signal’s preprocessing and wave group detection,construction of ECG classification model and system.The main content of this paper is divided into the following parts:1.A 50Hz notch filter,wavelet threshold method based on db4 wavelet function and morphological filter combined with open/close operation are used to remove three kinds of noises in ECG signal,including power line interference,myopotential interference and baseline wander.The R-wave detection algorithm based on continuous wavelet transform with improving its initial threshold calculation method is utilized to achieve an average detection accuracy rate of more than99%on partial ECGs of the MIT-BIH arrhythmia database.2.Based on the Cin C Challenge ECG database,a total of 314-dimensional mixed features are extracted as the dataset of features,and an ECG two-category classification model which can classify normal/abnormal is designed with the Light GBM algorithm.A 5-fold cross validation is performed on the dataset of features,and the average1 score of model reaches 0.878,which is better than the ECG two-category classification model designed by CART and Cat Boost algorithms.3.An ECG multi-category classification model based on 1D-CNN（1 Dimensional Convolutional Neural Network）is designed to classify normal rhythm,atrial fibrillation,other arrhythmias and noise.And a 5-fold cross validation is carried out on the Cin C Challenge ECG database,the model’s performance parameter1is 0.842,which is superior to the algorithms used in the other papers.4.To build an ECG intelligent diagnosis system based on cloud-edge computing,the ECG two-category classification model based on Light GBM with lower algorithm complexity is used on the edge device（Raspberry Pi 3B+）,the ECG multi-category classification model based on 1D-CNN with higher algorithm complexity is utilized on the cloud server（Huawei Cloud Server）,and then the data communication,information feedback and timed cycle diagnosis of the system are realized.In order to reduce the missed detection of abnormal ECGs and improve the diagnostic speed of the system,the ECG two-category classification model based on Light GBM is improved.Experiments have proved that the system can realize the timed cycle diagnosis at an interval of 1 second;compared with the ECG diagnosis system based on the cloud server,it can make full use of the computing power of the edge device and reduce the computing burden of the cloud server;compared with the ECG diagnosis system based on the edge device,it has faster diagnostic speed.