Scholarlink

<p>In recent years, the use of deep learning in ECG analysis has become a major research topic. The main reason for the increasing tendency of researchers to use deep learning is to achieve better results than classical methods based on traditional feature extraction. However, the use of these methods is limited due to the impossibility of extracting automatic features based on the relations of different consecutive segmented input data. The proposed hybrid neural network-based model in this paper, along with automatic features extracted from a deep convolutional neural network (DCNN), uses some helpful ECG-based heartbeat classical features that cannot be learned by DCNN because of segmenting ECG record to heartbeats. These classical features are based on simple necessary time-domain parameters for heartbeat record segmentation without any excessive processing time. Also, this approach allows using independent subnetworks, each one properly designed and trained for its input data type. Experiments conducted on the standard MIT-BIH Arrhythmia dataset show the effectiveness of the proposed scheme, in terms of precision 95.04%, recall 92.28%, F1-score 93.64%, and accuracy 99.13% in four-class state and precision 95.99%, recall 94.09%, F1-score 95.02%, and accuracy 99.33% in two-class state for 70%-30% training–test division of dataset.</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

A novel efficient hybrid deep learning framework for ECG-based heartbeat arrhythmia classification

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Reza Bahrami,
  • Ali M. Fotouhi

摘要

In recent years, the use of deep learning in ECG analysis has become a major research topic. The main reason for the increasing tendency of researchers to use deep learning is to achieve better results than classical methods based on traditional feature extraction. However, the use of these methods is limited due to the impossibility of extracting automatic features based on the relations of different consecutive segmented input data. The proposed hybrid neural network-based model in this paper, along with automatic features extracted from a deep convolutional neural network (DCNN), uses some helpful ECG-based heartbeat classical features that cannot be learned by DCNN because of segmenting ECG record to heartbeats. These classical features are based on simple necessary time-domain parameters for heartbeat record segmentation without any excessive processing time. Also, this approach allows using independent subnetworks, each one properly designed and trained for its input data type. Experiments conducted on the standard MIT-BIH Arrhythmia dataset show the effectiveness of the proposed scheme, in terms of precision 95.04%, recall 92.28%, F1-score 93.64%, and accuracy 99.13% in four-class state and precision 95.99%, recall 94.09%, F1-score 95.02%, and accuracy 99.33% in two-class state for 70%-30% training–test division of dataset.