Intelligent Diagnosis Algorithm For Non-uniform ECG Data Based On Deep Learning

Artificial intelligence technology has made breakthrough progress in the fields of computer vision and natural language processing in recent years,Deep learning is the key technology of artificial intelligence.At the same time,the detection and classification of electrocardiogram(ECG)is a hot topic in the medical field at present.The ECG classification method based on deep learning has reached a good accuracy rate,but these well-classification categories often depend on the amount of data.The model classification effect will be unbalanced if the model trained by Non-Uniform data.Accuracy of the recognition will be high on some categories with a large amount of data,but,will be low on some categories with a small amount of data.A single-label data samples including 21 types of arrhythmia is used in this paper,after a large number of experiments,a way for recognizing single-lead Non-Uniform ECG data is proposed,will optimized the part cannot identify arrhythmia categories without making lower the recognition of other arrhythmia categories.The main research work of this paper can be divided into the following three parts:Firstly:Based on the MXNet open source deep learning framework,an ECG classification model of residual neural network is designed and implemented.The residual neural network is an improvement on the traditional convolutional neural network.It solves the problems of vanishing gradient and degenerate in CNN by adding shortcut connections.Therefore,there is no need to worry about the network layer being too deep but the classification accuracy rate will not rise instead.Nearly 100,000 single-label ECG data will be used for the experiments in this paper,the preliminary work is mainly preprocessing ECG data,including data filtering,data cleaning,data normalization,and so on.Then,based on the DGX-1 server,eight V100 GPUs were used to train the single-lead ECG model and the eight-lead ECG model.Secondly:Analyzing the performance of the above model,we can see that the Non-Uniform data has a great impact on the performance of the model.For example,"atrial fibrillation" has a well classification because of the large amount of data,but"boundary escape" can hardly be identified because of the small amount of data.In view of this problem,a heartbeat feature extraction model is being designed using a large amount of heartbeat data in the open source MIT-BIH database and uses this model to extract heartbeat feature data for some arrhythmia disease data.the heartbeat feature can be fully extracted with a small amount of data by using this method and the recognition accuracy of the model can be improved.Thirdly:Build an online ECG recognition system based on the Django Rest Framework and Vue framework.Users can diagnose single-lead ECG data and eight-lead ECG data with this online system.At the same time,user visits and the number of various heart rate diseases can be counted by administrator users through the front-end system.