Scholarlink

<p>According to the World Health Organization, cardiovascular disease is the leading cause of death worldwide. Researchers are interested in diagnosing cardiovascular disease (CVD) using electrocardiography (ECG) pictures due to their ease of use, explainability, visualization, and representational potential. This paper presents CVD detection based on a deep learning-based two-way feature representation (DLTWFR) utilizing a 2-D Deep Convolutional Neural Network (2-D DCNN) and a 1-D DCNN. The 2-D DCNN accepts raw ECG images as input to provide connectivity and correlation within the original ECG pattern. The 1-D DCNN accepts the modified local ternary pattern (MLTP), and Histogram of Oriented Gradient (HOG) features to provide the correlation between different texture and shape features of the ECG images to characterize the local and global changes in the ECG pattern due to CVD. The proposed DLTWFR helps to improve the feature distinctiveness by combining the deep attributes from the raw ECF images, texture, and shape features of the ECG. The proposed DLTWFR offers improved overall accuracy of 98.20%, recall of 98.20%, precision of 98.10%, and F1-score of 98.05% than the existing state of arts.</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

Deep Learning-Based Two-Way Feature Representation of the ECG Images for Cardiovascular Disease Detection

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • S. G. Bagul,
  • S. B. Bagal,
  • B. S. Agarkar,
  • S. V. Chaudhari

摘要

According to the World Health Organization, cardiovascular disease is the leading cause of death worldwide. Researchers are interested in diagnosing cardiovascular disease (CVD) using electrocardiography (ECG) pictures due to their ease of use, explainability, visualization, and representational potential. This paper presents CVD detection based on a deep learning-based two-way feature representation (DLTWFR) utilizing a 2-D Deep Convolutional Neural Network (2-D DCNN) and a 1-D DCNN. The 2-D DCNN accepts raw ECG images as input to provide connectivity and correlation within the original ECG pattern. The 1-D DCNN accepts the modified local ternary pattern (MLTP), and Histogram of Oriented Gradient (HOG) features to provide the correlation between different texture and shape features of the ECG images to characterize the local and global changes in the ECG pattern due to CVD. The proposed DLTWFR helps to improve the feature distinctiveness by combining the deep attributes from the raw ECF images, texture, and shape features of the ECG. The proposed DLTWFR offers improved overall accuracy of 98.20%, recall of 98.20%, precision of 98.10%, and F1-score of 98.05% than the existing state of arts.