Scholarlink

<p>Sleep apnea is a prevalent sleep disorder characterized by interruption or release of air from the lungs for greater than 10&#xa0;s, posing significant risks to health if left undiagnosed. Identification of sleep apnea attacks is crucial for developing powerful therapeutic and management techniques. In this work, a hybrid deep learning model is proposed to detect sleep apnea using 70 single-lead ECG recordings taken from the PhysioNet Apnea ECG v1.0.0 dataset. The ECG signal is pre-processed and analyzed by the deep learning model to recognize sleep apnea. Among deep learning techniques, convolutional networks (MobileNetV1) and the novel hybrid convolutional-recurrent networks: MobileNetV1 + LSTM (Long Short-Term Memory) and MobileNetV1 + GRU (Gated Recurrent Unit) are implemented. The available data is divided into a training set (used to tune model parameters), a validation set (used to tune hyperparameters, prevent overfitting, and improve the generalizability of the models) and a test set (used to evaluate the invisible data). It is found that the hybrid deep model MobileNetV1 + GRU achieved the best detection results with an accuracy of 90.72%. This work can be efficiently integrated with the design of home-based sleep apnea screening devices for remote patient monitoring and early detection of sleep apnea. This work can be extended to detect other sleep disorders. The full implementation code can be found at <a href="https://zenodo.org/records/15363519">https://zenodo.org/records/15363519</a>.</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

Sleep apnea detection using hybrid deep learning model

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • G. Ananthi,
  • M. Siva Nandhini,
  • L. Subha Lakshmi

摘要

Sleep apnea is a prevalent sleep disorder characterized by interruption or release of air from the lungs for greater than 10 s, posing significant risks to health if left undiagnosed. Identification of sleep apnea attacks is crucial for developing powerful therapeutic and management techniques. In this work, a hybrid deep learning model is proposed to detect sleep apnea using 70 single-lead ECG recordings taken from the PhysioNet Apnea ECG v1.0.0 dataset. The ECG signal is pre-processed and analyzed by the deep learning model to recognize sleep apnea. Among deep learning techniques, convolutional networks (MobileNetV1) and the novel hybrid convolutional-recurrent networks: MobileNetV1 + LSTM (Long Short-Term Memory) and MobileNetV1 + GRU (Gated Recurrent Unit) are implemented. The available data is divided into a training set (used to tune model parameters), a validation set (used to tune hyperparameters, prevent overfitting, and improve the generalizability of the models) and a test set (used to evaluate the invisible data). It is found that the hybrid deep model MobileNetV1 + GRU achieved the best detection results with an accuracy of 90.72%. This work can be efficiently integrated with the design of home-based sleep apnea screening devices for remote patient monitoring and early detection of sleep apnea. This work can be extended to detect other sleep disorders. The full implementation code can be found at https://zenodo.org/records/15363519.