Title: KFG-MLNet: an ECG-based deep network using knowledge-guided features and multitask learning for arrhythmias classification

Authors: Yanyun Tao; Yuzhen Zhang

Addresses: Intelligent Structure and System, Soochow University, Suzhou, 215137, China; Key Laboratory of System Control and Information Processing, Ministry of Education, Shanghai, 200240, China; Key Laboratory of Digital Signal and Image Processing of Guangdong Province, Shantou, China ' Cardiology Department, The First Affiliated Hospital of Soochow University, Suzhou, 215006, China

Abstract: Deep learning networks performed well in electrocardiogram (ECG)-based arrhythmia classification. However, the classification accuracy and the requirement of lots of training samples are always the issues. In this study, we proposed a model called KFG-MLNet, which integrated expert knowledge-guided feature representation with two lightweight convolutional neural networks (CNNs). KFG-MLNet used a feature representation based on expert knowledge to better express the rhythm and heartbeat shape. This representation fully utilised the inequality of the R-R interval in an abnormal rhythm and the correlation between interheartbeats. Then, KFG-MLNet employed CNNs to realise rhythm and heartbeat arrhythmia classification, respectively. Multitask learning was used to train two CNNs on rhythm and heartbeat arrhythmia classification, and improve each other by sharing the convolutional layer. On the dataset of MIT-BIH, KFG-MLNet improved the accuracy by 2.07%, 10.05% and 3.68% over the other models.

Keywords: arrhythmia; multitask learning; rhythm; knowledge-guided feature; heartbeat recognition; neural network.

DOI: 10.1504/IJSCIP.2024.138677

International Journal of System Control and Information Processing, 2024 Vol.4 No.2, pp.154 - 171

Received: 20 Jul 2023
Accepted: 26 Mar 2024
Published online: 23 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article