基于鼾声特征的呼吸暂停低通气指数预测  

Prediction of apnea hypopnea index based on snoring sound characteristics

作　　者：叶自强 彭健新 YE Ziqiang;PENG Jianxin(School of Physics and Optoelectronics,South China University of Technology,Guangzhou 510640,Guangdong,China)

机构地区：[1]华南理工大学物理与光电学院,广东广州510640

出　　处：《声学技术》2024年第6期862-866,共5页Technical Acoustics

基　　金：国家自然科学基金(11974121)。

摘　　要：打鼾是阻塞性睡眠呼吸暂停低通气综合征(obstructive sleep apnea hypopnea syndrome, OSAHS)患者最典型的症状,患者鼾声中含有重要诊断信息。文章从OSAHS患者整晚鼾声中提取了梅尔倒谱系数、线性预测系数、谱熵等声学特征参数,通过相关分析研究患者鼾声的特征参数与呼吸暂停低通气指数(apnea hypopnea index, AHI)的相关性,探讨基于梯度提升回归的AHI预测,并与其他模型进行对比。结果表明:谱熵和梅尔倒谱系数的某些维度与AHI具有较强的相关性,其相关系数大于0.6。与线性模型相比,梯度提升回归模型表现出更好的预测效果,其预测的AHI与参考测量值之间的相关性高,相关系数为0.813。结果表明,基于鼾声的声学特征预测OSAHS患者的AHI是可行的,对OSAHS的诊断具有较好的参考价值。Snoring is the most common symptom in the patients with obstructive sleep apnea hypopnea syndrome(OSAHS).It provides important diagnostic information for patients.In this study,acoustic features such as Melfrequency cepstral coefficients(MFCC),linear predictive coefficients(LPC),and spectral entropy(SE)are extracted from the snoring sounds of patients throughout the night.The correlation between snoring characteristics and the apnea hypopnea index(AHI)is evaluated through correlation analysis.An AHI prediction model based on gradient boosting regression is investigated,and its performance is compared with that of other models.The results indicate a strong correlation between some characteristics of SE and MFCC with AHI,which have a correlation coefficient greater than 0.6.Compared to the linear model,the gradient boosting regression model demonstrates better prediction performance with a correlation coefficient of 0.813 between predicted AHI and measured values.These findings manifest that it is feasible to predict AHI based on the acoustic characteristics of snoring,providing valuable reference for diagnosing OSAHS patients.

关 键 词：阻塞性睡眠呼吸暂停低通气综合征 鼾声 呼吸暂停低通气指数 声学特征 梯度提升回归 

分 类 号：O429[理学—声学]