Automated ECG classification using a non-local convolutional block attention module. (May 2021)

• Record Type: Journal Article
• Authors: Wang, Jikuo; Qiao, Xu; Liu, Changchun; Wang, Xinpei; Liu, YuanYuan; Yao, Lianke; Zhang, Huan
• Is Part Of: Computer methods and programs in biomedicine. Volume 203(2021)
• Publication Date: 2021-05
• Subjects: ECG -- Cardiac arrhythmias, Cardiovascular diseases, Convolutional neural network, Attention mechanism, Non-local convolutional block attention module; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2021.106006
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: A novel convolutional neural network with non-local convolutional block attention module(NCBAM) which focuses on representative features along spatial, temporal and channel was proposed. Without using recurrent neural network architecture, the ncbam can rely fully on attention mechanism to draw global dependencies. The NCBAM increases the interpretability of deep learning-based models from the physiological and pathological point of view. Trained and tested on two public databases. Obtained an average F1 score of 0.9664 in the arrhythmias classification, as well as AUC of 0.9314 and Fmax of 0.8507 in the ECG diagnostic. Abstract: Background and objective: Recent advances in deep learning have been applied to ECG detection and obtained great success. The spatial and temporal information from ECG signals is fused by combining convolutional neural networks (CNN) with recurrent neural network (RNN). However, these networks ignore the different contribution of local and global segments of a feature map extracted from the ECG and the correlation relationship between the above two segments. To address this issue, a novel convolutional neural network with non-local convolutional block attention module(NCBAM) is proposed to automatically classify ECG heartbeats. Methods: Our proposed method consists of a 33-layer CNN architecture followed by a NCBAM module. Initially, preprocessed electrocardiogram (ECG) signals are fed into the CNN architecture to extract the spatial and channel features. Further, long-range dependencies of representative features along spatial and channel axis are captured by non-local attention. Finally, the spatial, channel and temporal information of ECG are fused by a learned matrix. The learned matrix is to mine rich relationship information across the above three types of information to make up for the different contribution. Results and conclusion: The proposed method achieves an average F 1 score of 0.9664 on MIT-BIH arrhythmia database, as well as A U C of 0.9314 and F max of 0.8507 on PTB-XL ECG database. Compared with the state-of-the-art attention mechanism based on the same public database, NCBAM achieves an obvious improvement in classifying ECG heartbeats. The results demonstrate the proposed method is reliable and efficient for ECG beat classification.