Scholarlink

<p>Electrocardiograms (ECG) are essential for monitoring heart functions, playing a critical role in healthcare and research. Machine learning (ML) techniques, including convolutional neural networks (CNNs), increasingly classify ECG data due to their potential for quick and accurate analysis. However, the limited size of ECG datasets and the complexity of CNN algorithms make capturing the temporal and morphological characteristics of ECG signals challenging. This study proposes a two-way approach combining signal-based and image-based methodologies. In the first pathway, ECG signals are analyzed directly using CNN with wavelet packet transform (WPT) to enhance ECG-based heartbeat classification. The integration of WPT aims to extract temporal and frequency-domain features. In the second pathway, WPT converts the signals into time–frequency images, and a separate CNN model designed for image-based analysis subsequently classifies them. An ensemble method then fuses the results from both pathways to produce the final classification. The proposed model achieves an accuracy of 0.9990 with a precision and recall of 0.9990, outperforming existing CNN-based approaches. A comparative analysis validates the model’s performance, displaying the model’s capability to handle ECG signals effectively. Combining CNN with WPT offers a robust solution for automated, rapid, and precise ECG signal classification. This approach addresses key challenges in ECG analysis, providing a promising framework for real-world applications in healthcare and research.</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

CNN–WPT model for the efficient heartbeat classification

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Shivani Saxena,
  • Nikunj Tahilramani,
  • Chirag N. Paunwala,
  • Ahsan Z. Rizvi

摘要

Electrocardiograms (ECG) are essential for monitoring heart functions, playing a critical role in healthcare and research. Machine learning (ML) techniques, including convolutional neural networks (CNNs), increasingly classify ECG data due to their potential for quick and accurate analysis. However, the limited size of ECG datasets and the complexity of CNN algorithms make capturing the temporal and morphological characteristics of ECG signals challenging. This study proposes a two-way approach combining signal-based and image-based methodologies. In the first pathway, ECG signals are analyzed directly using CNN with wavelet packet transform (WPT) to enhance ECG-based heartbeat classification. The integration of WPT aims to extract temporal and frequency-domain features. In the second pathway, WPT converts the signals into time–frequency images, and a separate CNN model designed for image-based analysis subsequently classifies them. An ensemble method then fuses the results from both pathways to produce the final classification. The proposed model achieves an accuracy of 0.9990 with a precision and recall of 0.9990, outperforming existing CNN-based approaches. A comparative analysis validates the model’s performance, displaying the model’s capability to handle ECG signals effectively. Combining CNN with WPT offers a robust solution for automated, rapid, and precise ECG signal classification. This approach addresses key challenges in ECG analysis, providing a promising framework for real-world applications in healthcare and research.