Scholarlink

<p>Sleep Apnea (SA) is an alarming sleep disorder characterized by repeated cessation of breathing during sleep, which frequently results in severe cardiovascular and metabolic impairment. The traditional techniques, such as polysomnography, are time-consuming, costly, and require clinical supervision. To address these challenges, this study proposes a hybrid DL-based framework for automated SA identification with single-lead ECG signals from the PhysioNet Apnea-ECG dataset. The methodology integrates comprehensive signal processing along with feature engineering to improve the morphological, temporal characteristics of ECG signals. The traditional ML models like SVM with Linear, Polynomial, Sigmoid, Radial Basis Kernels, and DL architectures such as 1D-CNN, 1D-CNNN-BiLSTM, and a hybrid 1D-CNN + BiLSTM-Attention mechanism automatically weight temporal features to focus on diagnostically significant ECG segments and improve the hidden feature space representation of prominent patterns. The efficacy of the model is demonstrated in the form of a validated quantitative performance evaluation using 10-fold cross-validation. The model achieved 98.39% Accuracy, 99.02% Precision, 98.29% Sensitivity, 96.53% Specificity, 98.66% F1-Score, 96.67% MCC, 97.78% AUC on PhysioNet-Apnea ECG dataset, indicating strong classification performance with robust generalization across patient’s sub-groups. The strength of the proposed model has been demonstrated using the MIT-BIH Polysomnographic database across various sleep patterns and recording conditions in diagnostic accuracy, efficiency, and generalization across physiological variations. The results have been compared with some of the state-of-the-art methods for establishing the superiority of the proposed model.</p> Graphical abstract <p></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

Attention enhanced hybrid deep learning model with 1D-CNN and BiLSTM for automated sleep apnea detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月17日

查看全文
  • Bipasha Patnaik,
  • Debasish Swapnesh Kumar Nayak,
  • Santanu Sahoo

摘要

Sleep Apnea (SA) is an alarming sleep disorder characterized by repeated cessation of breathing during sleep, which frequently results in severe cardiovascular and metabolic impairment. The traditional techniques, such as polysomnography, are time-consuming, costly, and require clinical supervision. To address these challenges, this study proposes a hybrid DL-based framework for automated SA identification with single-lead ECG signals from the PhysioNet Apnea-ECG dataset. The methodology integrates comprehensive signal processing along with feature engineering to improve the morphological, temporal characteristics of ECG signals. The traditional ML models like SVM with Linear, Polynomial, Sigmoid, Radial Basis Kernels, and DL architectures such as 1D-CNN, 1D-CNNN-BiLSTM, and a hybrid 1D-CNN + BiLSTM-Attention mechanism automatically weight temporal features to focus on diagnostically significant ECG segments and improve the hidden feature space representation of prominent patterns. The efficacy of the model is demonstrated in the form of a validated quantitative performance evaluation using 10-fold cross-validation. The model achieved 98.39% Accuracy, 99.02% Precision, 98.29% Sensitivity, 96.53% Specificity, 98.66% F1-Score, 96.67% MCC, 97.78% AUC on PhysioNet-Apnea ECG dataset, indicating strong classification performance with robust generalization across patient’s sub-groups. The strength of the proposed model has been demonstrated using the MIT-BIH Polysomnographic database across various sleep patterns and recording conditions in diagnostic accuracy, efficiency, and generalization across physiological variations. The results have been compared with some of the state-of-the-art methods for establishing the superiority of the proposed model.

Graphical abstract