Research On Heartbeat Classification Technology Based On One Dimensional Convolutional Neural Network

Electrocardiogram reflects the working condition of the human heart and is of great significance for the diagnosis and treatment of cardiovascular diseases.With the development of Holter and wearable ECG monitoring devices,the number of ECGs has increased dramatically.It is time-consuming and labor-intensive for doctors to process.The diagnosis results are different due to human factors such as individual doctors’ differences and experience levels.Automatic ECG signal analysis technology can analyze ECG signals in a timely and accurate manner,and plays a very important role in the diagnosis of cardiovascular disease.In order to make the automatic analysis of ECG signals better assist doctors in the diagnosis and treatment of cardiovascular diseases,a heartbeat classification algorithm based on one-dimensional convolutional neural network is studied.The main research contents are as follows:(1)Expand the ECG data to increase the generalization ability of the model.For many researchers who only use the public MIT-BIH ECG database or other single data sets when training or testing models,there will be problems of incomplete data set samples and uneven data volume of different heartbeat types.Based on the MIT-BIH ECG database,the Holter ECG data obtained by Tianjin Zhongke Alof Medical Technology Co.,Ltd.in the hospital and the ECG data collected by Elephant Suixinbao users are combined to perform one-dimensional convolutional nerve network models are trained,validated and tested.(2)Detection of unsuppressed noise in ECG signals based on threshold.After acquiring the new clinical ECG,the first step is usually to perform the signal quality assessment to find noise and process the detected noise.The detection algorithm based on threshold was researched for the unsuppressed noise in the ECG signal.A fixed-length sliding window was used to slide over the ECG signal segment to find the maximum value sequence,and then the threshold was used to determine whether the ECG signal segment was noise.Noise,discard the ECG segment and record its position in the original ECG data,so as not to affect the analysis of the later ECG signals.(3)QRS complex detection based on one-dimensional convolutional neural network.In order to solve the problems that existing QRS wave group detection algorithms rely on prior knowledge to manually extract features,robustness and poor anti-noise ability,an efficient QRS wave group detection algorithm based on one-dimensional convolutional neural networks is studied.The algorithm uses a one-dimensional convolutional neural network to implement QRS / non-QRS wave group two classification,and then outputs the QRS wave group detection results in real time through a non-maximum suppression method.Experimental results show that the algorithm has the characteristics of high accuracy,good robustness and strong anti-interference performance.(4)Heartbeat classification based on one-dimensional convolutional neural network.Aiming at the problem of the low recognition rate of normal heartbeat and supraventricular ectopic heartbeat when using a single one-dimensional convolutional neural network for multi-classification of heartbeats,two one-dimensional convolutional neural network two-classification models were used to completed the multi-classification of most common heartbeats in clinical practice,such as the normal heartbeat,the supraventricular ectopic beat and the ventricular ectopic heartbeat.The experimental results prove that the classification results of two one-dimensional convolutional neural network two-class classification models on three types of heartbeats have been significantly improved.