Detection of sleep breathing sound based on artificial neural network analysis
Detection of sleep breathing sound based on artificial neural network analysis
复制标题
DOI：
10.1016/j.bspc.2017.11.005
复制
发表时间：
2018-03-01
影响因子：
5.1
通讯作者：
Kawata, Ikuji
中科院分区：
文献类型：
--
作者：
Emoto, Takahiro;Abeyratne, Udantha R.;Kawata, Ikuji
众所周知，阻塞性睡眠呼吸暂停低通气综合征 (OSAHS) 会导致白天嗜睡，并与 II 型糖尿病、心血管疾病和中风等疾病有关。多导睡眠图 (PSG) 测试是诊断 OSAHS 的传统方法。然而，这种测试昂贵、不方便，并且需要在睡眠期间放置身体接触传感器。最近，在多项研究中，非接触式麦克风采集的打鼾/呼吸发作（SBE）已被用于 OSAHS 诊断。 SBE 的范围可能从几乎听不见到很大。由于信噪比 (SNR) 较低，SBE 检测（尤其是低强度 SBE）在噪声环境中可能具有挑战性。在本文中，我们提出了一种从睡眠期间记录的数据快速检测低强度 SBE 的新方法。我们的方法基于人工神经网络（ANN）技术。当将 ANN 训练为特定主题分类器时，我们表明与之前的方法相比，所提出的方法能够更快地检测低强度 SBE。当使用 ANN 作为独立于受试者的分类器时，我们表明所提出的方法可以对实际睡眠期间可能发生的低强度 SBE 和低强度非 SBE 进行分类，平均准确度为 75.10%。 (C) 2017 Elsevier Ltd. 保留所有权利。
Obstructive sleep apnea-hypopnea syndrome (OSAHS) is known to cause daytime drowsiness and an association with diseases such as Type II diabetes, cardiovascular disease, and stroke. A polysomnography (PSG) test is the traditional method for diagnosing OSAHS. However, this test is expensive, inconvenient, and requires the placement of body contact sensors during sleep. Recently, in several studies, the snoring/breathing episodes (SBEs) acquired by non-contact microphones have been used for OSAHS diagnosis. SBEs may range from barely audible to loud. SBE detection, especially low-intensity SBEs, can be challenging in noisy environments because of the low signal-to-noise ratio (SNR). In this paper, we propose a novel method for the rapid detection of low-intensity SBEs from data recorded during sleep. Our method is based on an artificial neural network (ANN) technique. When an ANN is trained as subject-specific classifier, we show that the proposed method is capable of detecting low-intensity SBEs more rapidly compared to our previous method. When an ANN is used as subject-independent classifier, we show that the proposed method can classify low-intensity SBEs and low-intensity non-SBEs that may occur during actual sleep with an average accuracy of 75.10%. (C) 2017 Elsevier Ltd. All rights reserved.