Scholarlink

<p>Cardiac arrhythmia is one of the leading causes of mortality and cardiovascular diseases worldwide, highlighting the need for accurate and automated diagnostic systems. Timely and accurate detection of these cardiac rhythm disorders through electrocardiogram (ECG) signal analysis is crucial. However, conventional diagnostic methods rely heavily on physician expertise and are prone to human error. This study explores the evolutionary path from feature-based machine learning to advanced deep learning paradigms, including CNNs, RNNs, Transformers, and GNNs, and conducts an integrated experimental evaluation of all investigated methods on the widely used MIT-BIH arrhythmia database under identical preprocessing, training, and evaluation conditions. Based on this comprehensive comparative analysis, a novel hybrid SAGEConv-Transformer architecture is proposed to jointly capture structural relationships and temporal dependencies in ECG signals, thereby improving classification performance. Experimental results demonstrate that the proposed SAGEConv-Transformer hybrid model consistently outperforms all evaluated classical machine learning, single-architecture deep learning, and hybrid baseline models, achieving 98.37% accuracy, 98.35% precision, 98.36% recall, and 98.34% F1-score. The robustness and generalizability of the proposed framework were further validated through fivefold cross-validation, external evaluation on the PTB Diagnostic ECG Database, statistical significance testing, and Grad-CAM-based interpretability analysis. These findings confirm that the proposed framework provides an accurate, robust, interpretable, and generalizable solution for automated ECG-based cardiac arrhythmia classification.</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

Benchmarking machine learning and deep learning models with a novel graph-attention SAGEConv-transformer for ECG arrhythmia detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Motahareh Akbari Poodineh,
  • Fatemeh Zare Mehrjardi,
  • Mohsen Sardari Zarchi

摘要

Cardiac arrhythmia is one of the leading causes of mortality and cardiovascular diseases worldwide, highlighting the need for accurate and automated diagnostic systems. Timely and accurate detection of these cardiac rhythm disorders through electrocardiogram (ECG) signal analysis is crucial. However, conventional diagnostic methods rely heavily on physician expertise and are prone to human error. This study explores the evolutionary path from feature-based machine learning to advanced deep learning paradigms, including CNNs, RNNs, Transformers, and GNNs, and conducts an integrated experimental evaluation of all investigated methods on the widely used MIT-BIH arrhythmia database under identical preprocessing, training, and evaluation conditions. Based on this comprehensive comparative analysis, a novel hybrid SAGEConv-Transformer architecture is proposed to jointly capture structural relationships and temporal dependencies in ECG signals, thereby improving classification performance. Experimental results demonstrate that the proposed SAGEConv-Transformer hybrid model consistently outperforms all evaluated classical machine learning, single-architecture deep learning, and hybrid baseline models, achieving 98.37% accuracy, 98.35% precision, 98.36% recall, and 98.34% F1-score. The robustness and generalizability of the proposed framework were further validated through fivefold cross-validation, external evaluation on the PTB Diagnostic ECG Database, statistical significance testing, and Grad-CAM-based interpretability analysis. These findings confirm that the proposed framework provides an accurate, robust, interpretable, and generalizable solution for automated ECG-based cardiac arrhythmia classification.