School of Biomedical Engineering (Suzhou), Division of Life Sciences and Medicine, University of Science and Technology of China, China; Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Science, China.
Suzhou Guoke Medical Technology Development (Group) Co, China.
Sleep Med. 2023 Jul;107:187-195. doi: 10.1016/j.sleep.2023.04.030. Epub 2023 May 9.
Obstructive sleep apnea (OSA) is a chronic sleep disorder characterized by frequent cessations or reductions of breathing during sleep. Polysomnography (PSG) is a definitive diagnostic tool for OSA. The costly and obtrusive nature of PSG and poor access to sleep clinics have created a demand for accurate home-based screening devices.
This paper proposes a novel OSA screening method based solely on breathing vibration signals with a modified U-Net, allowing patients to be tested at home. Sleep recordings over a whole night are collected in a contactless manner, and sleep apnea-hypopnea events are labeled by a deep neural network. The apnea-hypopnea index (AHI) calculated from events estimation is then used to screen for the apnea. The performance of the model is tested by event-based analysis and comparing the estimated AHI with the manually obtained values.
The accuracy and sensitivity of sleep apnea events detection are 97.5% and 76.4%, respectively. The mean absolute error of AHI estimation for the patients is 3.0 events/hour. The correlation between the ground truth AHI and predicted AHI shows an R of 0.95. In addition, 88.9% of all participants are classified into correct AHI categories.
The proposed scheme has great potential as a simple screening tool for sleep apnea. It can accurately detect potential OSA and help the patients to be referred for differential diagnosis of home sleep apnea test (HSAT) or polysomnographic evaluation.
阻塞性睡眠呼吸暂停（OSA）是一种慢性睡眠障碍，其特征是睡眠期间频繁出现呼吸暂停或减少。多导睡眠图（PSG）是 OSA 的明确诊断工具。PSG 的昂贵和干扰性质以及睡眠诊所的获取不足，催生了对准确的基于家庭的筛查设备的需求。
本文提出了一种基于呼吸振动信号的新型 OSA 筛查方法，该方法使用改进的 U-Net，允许患者在家中进行测试。以非接触方式采集整晚的睡眠记录，并使用深度神经网络对睡眠呼吸暂停低通气事件进行标记。然后，根据事件估计计算出的呼吸暂停低通气指数（AHI）用于筛查呼吸暂停。通过事件分析和比较手动获得的 AHI 值来测试模型的性能。
睡眠呼吸暂停事件检测的准确性和灵敏度分别为 97.5%和 76.4%。对于患者的 AHI 估计的平均绝对误差为 3.0 次/小时。实际 AHI 和预测 AHI 之间的相关性显示 R 为 0.95。此外，88.9%的参与者被归入正确的 AHI 类别。
该方案作为一种简单的睡眠呼吸暂停筛查工具具有很大的潜力。它可以准确地检测出潜在的 OSA，并帮助患者进行家庭睡眠呼吸暂停测试（HSAT）或多导睡眠图评估的鉴别诊断。
