Edge AI-Enabled Digital Twin Framework for Real-Time Sleep Apnea Detection and Prediction

Sleep apnea is an important but often undiagnosed disorder, which causes frequent pauses in breathing during sleep and can pose significant health risks if left undetected at an early stage. Conventional diagnostic approaches, like polysomnography, are expensive, laborious, and restricted to clinical environments, which made long-term monitoring cumbersome. In this paper, we address these issues by introducing E-DTSA-Net (Edge-based Digital Twin Sleep Apnea Network), which enables real-time detection and prediction via wearable sensors. The proposed system captures physiological signals such as oxygen saturation, heart rate, respiration patterns, and snoring data. The ceramic processor utilizes Edge AI methods to preprocess the data on-device, extract features using 1D CNN, and conduct real-time classification with LightGBM to ensure low latency and drastically reduced data transmission. In a cloud digital twin, the patient’s respiratory behavior is modeled, and it continuously updates from incoming data. A GRU model is used to predict breathing patterns, and an autoencoder system identifies anomalies from it. The system accurately detects apnea events and generates instantaneous alerts for early intervention. With experimental results showing improved performance, lower latency, and enhanced ability for continuous monitoring on the same chip, rendering it attractive for cost-effective and scalable healthcare applications.