Elwali Ahmed, Meza-Vargas Sonia, Moussavi Zahra
a Biomedical Engineering Program , University of Manitoba , Winnipeg , Canada.
b Respirology Department , University of Manitoba , Winnipeg , Canada.
J Med Eng Technol. 2019 Feb;43(2):111-123. doi: 10.1080/03091902.2019.1617799. Epub 2019 Jun 18.
Obstructive sleep apnoea (OSA) is a common yet underdiagnosed disorder. Undiagnosed OSA significantly increases perioperative morbidity and mortality for OSA patients undergoing surgery, requiring full anaesthesia. Tracheal breathing sounds characteristics during wakefulness have shown a high correlation with the apnoea-hypopnea index (AHI), while they are also affected by the anthropometric parameters, e.g., sex, age, etc. This study investigates the effects of the anthropometric parameters on our new quick objective OSA screening tool during wakefulness. Breathing sounds of 122 individuals (71 with AHI <15 as non-OSA and 51 with AHI > 15 as OSA) were recorded during wakefulness in the supine position. The spectra and bi-spectra of 81 (47 non-OSA) individuals' signals, which were randomly selected, were analysed as a training dataset to extract the most significant features with the lowest sensitivity to the anthropometric parameters. Using a support vector machine (SVM) classifier, these features resulted in 72.1, 64.7 and 77.5% testing classification accuracy, sensitivity and specificity, respectively. We also investigated classifying subjects into subgroups related to each anthropometric parameter and incorporating a voting procedure. This routine resulted in 83.6, 74.5 and 90.1% testing classification accuracy, sensitivity and specificity, respectively. In conclusion, it is possible to positively utilise the anthropometric information to enhance the classification accuracy for a reliable OSA screening procedure during wakefulness.
阻塞性睡眠呼吸暂停（OSA）是一种常见但诊断不足的疾病。未被诊断出的OSA会显著增加接受全身麻醉手术的OSA患者围手术期的发病率和死亡率。清醒时的气管呼吸音特征与呼吸暂停低通气指数（AHI）高度相关，同时它们也受人体测量参数的影响，例如性别、年龄等。本研究调查了人体测量参数对我们新的清醒时快速客观OSA筛查工具的影响。在122名个体清醒且仰卧位时记录其呼吸音（71名AHI<15的个体为非OSA，51名AHI>15的个体为OSA）。随机选择81名（47名非OSA）个体的信号频谱和双谱作为训练数据集进行分析，以提取对人体测量参数敏感度最低的最显著特征。使用支持向量机（SVM）分类器，这些特征的测试分类准确率、敏感度和特异度分别为72.1%、64.7%和77.5%。我们还研究了将受试者分类到与各人体测量参数相关的亚组中，并纳入投票程序。该程序的测试分类准确率、敏感度和特异度分别为83.6%、74.5%和90.1%。总之，在清醒时的可靠OSA筛查程序中，积极利用人体测量信息来提高分类准确率是可行的。
