Clinical phenotyping of sleep apnea in heart failure patients using a dual-modal wearable system and deep learning
J Wang, T Zou, Q W Xiong, D H Liu, Y Zou, S H ShenAbstract
Background
Sleep apnea (SA) is prevalent in heart failure (HF) patients and is a key driver of poor prognosis. Accurate differentiation between obstructive (OSA) and central sleep apnea (CSA) is vital, as their management (e.g., CPAP vs. ASV) differs significantly. We evaluated a novel dual-modal wearable system designed for long-term SA monitoring and phenotypic classification in an HF-related cohort.
Methods
A wrist-worn device integrating a piezoelectric nanogenerator (PENG) and photoplethysmography (PPG) was used. We developed a two-stage diagnostic strategy: continuous pulse pressure wave (PPW) screening via PENG, followed by PPG-based oxygen/hemodynamic validation. To address the challenge of phenotyping, a Vision Transformer (ViT) model was trained to identify "respiratory effort" signatures from PPW micro-vibrations, enabling the distinction between airway obstruction (OSA) and cessation of respiratory drive (CSA).
Results
In clinical validation against polysomnography (PSG), the system achieved a robust detection accuracy of 94.2% for sleep apnea events. For the specific task of phenotypic classification, the ViT-based model demonstrated a sensitivity of 88.6% for CSA and 91.3% for OSA, with an overall phenotyping accuracy of 89.5%. The two-stage approach reduced system power consumption by 23%, facilitating multi-night monitoring. Unlike traditional single-modal wearables, the PENG sensor's high mechanical sensitivity captured subtle thoracic-equivalent vibrations crucial for CSA identification.
Conclusion
This dual-modal wearable system offers a clinically viable, low-power solution for differentiating SA phenotypes in heart failure. By providing high-fidelity data on both autonomic response and respiratory patterns, it enables personalized sleep therapy and long-term home-based management for HF patients.System ArchitectureFor image description, please refer to the figure legend and surrounding text.Clinical PerformanceFor image description, please refer to the figure legend and surrounding text.