基于生理信息引导的Superlet-Ridge心肺耦合框架检测阻塞性睡眠呼吸暂停

文献信息

中文摘要

从单导联心电图（ECG）中准确且实用地筛查阻塞性睡眠呼吸暂停（OSA）仍具挑战性，部分原因是传统心肺耦合（CPC）分析未能充分捕捉与呼吸暂停相关的心肺相互作用所对应的频带特异性动态。我们提出一种生理信息引导的Superlet-Ridge框架（PISRF），其中CPC采用生理信息引导的Superlet变换（PISLT）表示，在低频（LF）带采用乘法模式，在高频（HF）带采用加法模式。这种频带特异性设计旨在更好地保留病理性LF调制和生理性HF耦合。在所得CPC表示的基础上，进一步使用基于ridge的分解提取归一化HF ridge能量（HFRn）作为判别特征。在PhysioNet Apnea-ECG数据集上，基于PISLT的CPC优于由连续小波变换和小波同步挤压变换得到的CPC，在各分类器上平均准确率达到84.40%，平均F1分数达到80.17%。通过进一步纳入HFRn ridge特征，PISRF将性能提升至平均准确率85.01%和平均F1分数81.33%。这些结果支持PISRF作为一种可解释且可实际部署的基于ECG的OSA筛查框架。

英文摘要

Accurate and practical screening of obstructive sleep apnea (OSA) from single-lead electrocardiogram (ECG) remains challenging, partly because conventional cardiopulmonary coupling (CPC) analysis does not fully capture the band-specific dynamics associated with apnea-related cardiorespiratory interactions. We propose a Physiology-Informed Superlet-Ridge Framework (PISRF), in which CPC is represented using a Physiology-Informed Superlet Transform (PISLT) with a multiplicative mode in the low-frequency (LF) band and an additive mode in the high-frequency (HF) band. This band-specific design is intended to better preserve pathological LF modulation and physiological HF coupling. Building on the resulting CPC representation, ridge-based decomposition is further used to extract the normalized HF ridge energy (HFRn) as a discriminative feature. On the PhysioNet Apnea-ECG dataset, PISLT-based CPC outperformed CPC derived from the continuous wavelet transform and wavelet synchrosqueezing transform, achieving 84.40% average accuracy and 80.17% average F1-score across classifiers. By further incorporating the HFRn ridge feature, PISRF improved performance to 85.01% average accuracy and 81.33% average F1-score. These results support PISRF as an interpretable and practically deployable ECG-based framework for OSA screening.