APNEAGUARD AI: AN ADVANCED DEEP LEARNING ECOSYSTEM FOR NON-INVASIVE SLEEP APNEA SCREENING UTILIZING LOW-RESOLUTION WEARABLE PHOTOPLETHYSMOGRAPHY

Abstract

Sleep apnea is a prevalent yet significantly underdiagnosed condition linked to severe cardiovascular and metabolic comorbidities. The clinical gold standard, Polysomnography (PSG), remains inaccessible to most due to high cost, invasiveness, and limited availability. This paper presents ApneaGuard AI, a scalable, non-invasive screening system that leverages low-resolution (1 Hz) heart rate (HR) and blood oxygen saturation (SpO₂) data from consumer smartwatches. We propose a multi-modal deep learning approach based on an InceptionTime ensemble architecture capable of capturing multi-scale temporal patterns in sparse signals.