Scholarlink

<p>Automated categorization of arrhythmias is increasingly vital for the clinical oversight of heart conditions. Identifying these pathologies typically relies on the examination of Electrocardiogram (ECG) recordings. Nonetheless, the limited availability of specialists capable of evaluating vast quantities of ECG information remains a major hurdle, requiring substantial healthcare investment. Consequently, leveraging machine learning for the detection of ECG characteristics has emerged as a practical solution. This research introduces a unique multichannel learning architecture that integrates wavelet-transformed signal data with Mel-Frequency Cepstral Coefficients (MFCCs) characteristics for ECG classification. Within this framework, the primary signal path utilizes wavelet extraction to isolate precise time-frequency details, followed by One-Dimensional Deep Convolutional Neural Network (1D-DCNN) and Bidirectional Long Short-Term Memory (BiLSTM) layers to analyze temporal patterns. Concurrently, a 2D CNN is employed to derive spectral attributes from the MFCCs features. Once these feature sets are merged via concatenation, they are processed through a series of dense, fully connected layers to determine the final class. The introduced methodology reached exceptional classification accuracies of 99.70% on the MIT-BIH Arrhythmia Database (MITDB) and 99.44% on the MIT-BIH Atrial Fibrillation Database (AFDB), exceeding the performance of current state-of-the-art techniques. These results highlight the significant potential of combining multichannel learning with wavelet-based feature isolation to refine the accuracy of ECG signal analysis.</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

Multichannel Learning Framework for Enhanced ECG Signal Classification Using Wavelet and MFCCs Features

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Tran Anh Vu,
  • Mai Tat Chuyen,
  • Nguyen Thi Diem Anh,
  • Hoang Quang Huy,
  • Pham Thi Viet Huong

摘要

Automated categorization of arrhythmias is increasingly vital for the clinical oversight of heart conditions. Identifying these pathologies typically relies on the examination of Electrocardiogram (ECG) recordings. Nonetheless, the limited availability of specialists capable of evaluating vast quantities of ECG information remains a major hurdle, requiring substantial healthcare investment. Consequently, leveraging machine learning for the detection of ECG characteristics has emerged as a practical solution. This research introduces a unique multichannel learning architecture that integrates wavelet-transformed signal data with Mel-Frequency Cepstral Coefficients (MFCCs) characteristics for ECG classification. Within this framework, the primary signal path utilizes wavelet extraction to isolate precise time-frequency details, followed by One-Dimensional Deep Convolutional Neural Network (1D-DCNN) and Bidirectional Long Short-Term Memory (BiLSTM) layers to analyze temporal patterns. Concurrently, a 2D CNN is employed to derive spectral attributes from the MFCCs features. Once these feature sets are merged via concatenation, they are processed through a series of dense, fully connected layers to determine the final class. The introduced methodology reached exceptional classification accuracies of 99.70% on the MIT-BIH Arrhythmia Database (MITDB) and 99.44% on the MIT-BIH Atrial Fibrillation Database (AFDB), exceeding the performance of current state-of-the-art techniques. These results highlight the significant potential of combining multichannel learning with wavelet-based feature isolation to refine the accuracy of ECG signal analysis.