Scholarlink

<p>This paper introduces an innovative method for myocardial infarction detection by leveraging advanced ECG signal processing and deep learning techniques. A robust preprocessing pipeline is applied, including Chebyshev, Butterworth, and Daubechies wavelet filters for noise reduction, normalization, and domain-specific data augmentation, ensuring accurate signal representation. The ECG signals are then transformed into Gramian Angular Field (GAF) images and scalograms, capturing both temporal dependencies and time-frequency features. These representations are integrated through a novel Vector Field-Driven Weighted Fusion technique, dynamically assigning weights to temporal and frequency features to emphasize critical attributes for classification. Ablation studies reveal significant drops in accuracy when any component—GAF, scalogram, or weighted fusion—is omitted, underscoring their essential roles. A deep learning architecture with parallel and residual connections further captures hierarchical and spatial features from the fused images. Our model achieves accuracy up to 98.90%, sensitivity up to 98.89%, precision up to 98.94%, and an F1-score up to 98.92% across multiple ECG leads, illustrating its significant performance. These results highlight the precision and innovation of the Vector Field-Driven Weighted Fusion technique, advancing ECG-based cardiac diagnosis through robust myocardial infarction detection.</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

MI-VFDF: vector field-driven weighted fusion of multi-leads ECG-derived GAF and scalogram images for accurate myocardial infarction detection in combined neural networks

  • 期刊论文

  • 发表日期：

    2025年10月16日

查看全文
  • Zakaria Khatar,
  • Dounia Bentaleb,
  • Omar Bouattane

摘要

This paper introduces an innovative method for myocardial infarction detection by leveraging advanced ECG signal processing and deep learning techniques. A robust preprocessing pipeline is applied, including Chebyshev, Butterworth, and Daubechies wavelet filters for noise reduction, normalization, and domain-specific data augmentation, ensuring accurate signal representation. The ECG signals are then transformed into Gramian Angular Field (GAF) images and scalograms, capturing both temporal dependencies and time-frequency features. These representations are integrated through a novel Vector Field-Driven Weighted Fusion technique, dynamically assigning weights to temporal and frequency features to emphasize critical attributes for classification. Ablation studies reveal significant drops in accuracy when any component—GAF, scalogram, or weighted fusion—is omitted, underscoring their essential roles. A deep learning architecture with parallel and residual connections further captures hierarchical and spatial features from the fused images. Our model achieves accuracy up to 98.90%, sensitivity up to 98.89%, precision up to 98.94%, and an F1-score up to 98.92% across multiple ECG leads, illustrating its significant performance. These results highlight the precision and innovation of the Vector Field-Driven Weighted Fusion technique, advancing ECG-based cardiac diagnosis through robust myocardial infarction detection.