Automatic recognition of arrhythmia based on principal component analysis network and linear support vector machine. (1st October 2018)

• Record Type: Journal Article
• Authors: Yang, Weiyi; Si, Yujuan; Wang, Di; Guo, Buhao
• Is Part Of: Computers in biology and medicine. Volume 101(2018)
• Page Start: 22
• Page End: 32
• Publication Date: 2018-10-01
• Subjects: Principal component analysis network -- Arrhythmia recognition -- Noise robustness -- Deep learning -- Cardiovascular diseases; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2018.08.003
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Electrocardiogram (ECG) classification is an important process in identifying arrhythmia, and neural network models have been widely used in this field. However, these models are often disrupted by heartbeat noise and are negatively affected by skewed data. To address these problems, a novel heartbeat recognition method is presented. The aim of this study is to apply a principal component analysis network (PCANet) for feature extraction based on a noisy ECG signal. To improve the classification speed, a linear support vector machine (SVM) was applied. In our experiments, we identified five types of imbalanced original and noise-free ECGs in the MIT-BIH arrhythmia database to verify the effectiveness of our algorithm and achieved 97.77% and 97.08% accuracy, respectively. The results show that our method has high recognition accuracy in the classification of skewed and noisy heartbeats, indicating that our method is a practical ECG recognition method with suitable noise robustness and skewed data applicability. Highlights: Our method is robust to noise. Our results are better than most other methods. Our method has only a few hyperparameters and does not require iteration. Fiducial point detection is no need. Our approach is applicable for skewed data.