MFB-LANN: A lightweight and updatable myocardial infarction diagnosis system based on convolutional neural networks and active learning. (October 2021)

• Record Type: Journal Article
• Authors: He, Ziyang; Yuan, Zhiyong; An, Panfeng; Zhao, Jianhui; Du, Bo
• Is Part Of: Computer methods and programs in biomedicine. Volume 210(2021)
• Publication Date: 2021-10
• Subjects: Myocardial infarction (MI) -- Convolutional neural networks (CNN) -- Lead attention mechanism (LAM) -- Active learning (AL) -- Real-time diagnosis; 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.2021.106379
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: A lightweight and updatable MFB-LANN model based on neural networks and active learning is employed to diagnose MI. The MFB module and the LAN module can learn the internal features of the heartbeat and ECG lead relationship features. The model can effectively overcome the individual differences between in patients through patient-specific scheme and active learning. The model update process and real-time diagnosis process can be completed on the embedded device. The model achieved satisfactory results for MI multi-category diagnosis under intra-patient and patient-specific schemes. Abstract: Background and objectives: 12 leads electrocardiogram (ECG) are widely used to diagnose myocardial infarction (MI). Generally, the symptoms of MI can be reflected by waveforms in the heartbeat, and the contribution of different ECG leads to different types of MI is different. Therefore, it is significant to use the heartbeat waveform features and the lead relationship features for multi-category MI diagnosis. Moreover, the challenge of individual differences and lightweight algorithms also need to be further resolved and explored in the ECG automatic diagnosis system. Methods: This paper presents a lightweight MI diagnosis system named multi-feature-branch lead attention neural network (MFB-LANN) via 12 leads ECG signals. It is designed based on the characteristics of the ECG lead. Specifically, 12 independent feature branches correspond to different leads, and each branch contains different convolutional layers to extract features in the heartbeat, then a novel attention module is developed named lead attention mechanism (LAM) to assign different weights to each feature branch. Finally all the weighted feature branches are fused for classification. Furthermore, to overcome individual differences, patient-specific scheme and active learning (AL) are used to train and update the model iteratively. Results: Experimental results based on Physikalisch-Technische Bundesanstalt (PTB) database shows that the MFB-LANN achieved satisfactory results with accuracy of 99.63% based on 5-fold cross validation under the intra-patient scheme. The patient-specific experiment yielded an average accuracy of 96.99% compared to the state-of-the-art. By contrast, the model achieved acceptable results on the hybrid database (PTB and PTB-XL), especially achieving 94.19% accuracy after the update. Moreover, the system can complete the update process and real-time diagnosis on the ARM Cortex-A72 platform. Conclusions: Experiments show that the proposed method for MI diagnosis has more obvious advantages compared to other recent methods, and it has great potential to be applied to the mobile medical field.