Research Article
ECG Arrhythmia Heartbeat Classification Using Deep Learning Networks
@INPROCEEDINGS{10.1007/978-3-030-69992-5_14, author={Yuxi Yang and Linpeng Jin and Zhigeng Pan}, title={ECG Arrhythmia Heartbeat Classification Using Deep Learning Networks}, proceedings={Cloud Computing. 10th EAI International Conference, CloudComp 2020, Qufu, China, December 11-12, 2020, Proceedings}, proceedings_a={CLOUDCOMP}, year={2021}, month={2}, keywords={ECG heartbeat classification CNN LSTM Attention mechanism ResNet}, doi={10.1007/978-3-030-69992-5_14} }- Yuxi Yang
Linpeng Jin
Zhigeng Pan
Year: 2021
ECG Arrhythmia Heartbeat Classification Using Deep Learning Networks
CLOUDCOMP
Springer
DOI: 10.1007/978-3-030-69992-5_14
Abstract
The electrocardiogram (ECG) records the process of depolarization and repolarization of the heart and contains many important details related to the condition of the human heart. In this paper, we designed four deep learning network structures and three electrocardiogram signal preprocessing methods, under the same dataset, explored the impact and performance of different preprocessing methods and models on the ECG arrhythmia classification work. For a fairer comparison, we used intra-patient and inter-patient evaluation for the final classification evaluation. In the evaluation of the intra-patient, the proposed network structures can achieve an accuracy of more than 95%. In the evaluation of inter-patient, all classification models can achieve an accuracy rate of more than 81.7%. During our research, we found convolutional neural network (CNN) is good at capturing spatial features of ECG. Long short-term memory networks (LSTM) is suitable for processing time-series signals. The combination of the two has a better classification performance than the sole network. Besides, the Attention mechanism can help the model do better on focusing on abnormal heartbeats also improve the interpretability of the model. Residual neural Network (ResNet) has good behavior in intra-patient, but not suitable for the inter-patient classification due to the vanishing gradient problem. Compared to the different preprocessing methods, we recommended using the raw signal in future work.
