Scholarlink

<p>Electrocardiogram (ECG) analysis is an essential tool for detecting cardiac arrhythmias due to its affordability and simplicity. However, manual review of ECG data is both time consuming and prone to human error, and automation is a critical need. In this work, we introduce MAB-Net, a new deep learning model that combines MobileNet, BiLSTM, and attention mechanisms to enhance the efficiency and accuracy of arrhythmia detection. MAB-Net was trained on the MIT-BIH arrhythmia dataset, which contains 87,594 ECG samples across five classes (non-ectopic beats, supraventricular ectopic beats, ventricular ectopic beats, fusion beats, and unknown beats), and achieved a loss of 0.53 and an accuracy of 98.52%. These results outperform other models such as CNN (96.14%), CNN-LSTM (97.54%), VGG (98.22%), and MobileNet (98.26%). MAB-Net also performed well at identifying less common arrhythmia classes, demonstrating its reliability and scalability. MAB-Net is shown to be able to accurately and automatically diagnose arrhythmias in both clinical and remote healthcare settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MAB-Net: an enhanced MobileNet with attention and BiLSTM model for arrhythmia detection using electrocardiogram signals

  • 期刊论文

  • 发表日期：

    2025年3月13日

查看全文
  • K. R. Febeena,
  • Cini Kurian

摘要

Electrocardiogram (ECG) analysis is an essential tool for detecting cardiac arrhythmias due to its affordability and simplicity. However, manual review of ECG data is both time consuming and prone to human error, and automation is a critical need. In this work, we introduce MAB-Net, a new deep learning model that combines MobileNet, BiLSTM, and attention mechanisms to enhance the efficiency and accuracy of arrhythmia detection. MAB-Net was trained on the MIT-BIH arrhythmia dataset, which contains 87,594 ECG samples across five classes (non-ectopic beats, supraventricular ectopic beats, ventricular ectopic beats, fusion beats, and unknown beats), and achieved a loss of 0.53 and an accuracy of 98.52%. These results outperform other models such as CNN (96.14%), CNN-LSTM (97.54%), VGG (98.22%), and MobileNet (98.26%). MAB-Net also performed well at identifying less common arrhythmia classes, demonstrating its reliability and scalability. MAB-Net is shown to be able to accurately and automatically diagnose arrhythmias in both clinical and remote healthcare settings.