Non-Contact Radar Technology and Machine Learning for Automated Sleep Apnea-Hypopnea Syndrome Detection

Sleep Apnea-Hypopnea Syndrome (SAHS) is a prevalent sleep disorder that significantly affecting patients' quality of life, often going undetected due to its appearance during sleep. The current gold standard for SAHS detection, polysomnogram, is costly and inconvenient for long-term monitoring. This paper introduces a novel method using non-contact Impulse-Radio Ultra-Wideband (IR-UWB) radar and machine learning to automatically detect apnea-hypopnea events. Initially, after selecting the appropriate target range bins from each radar data, the Variational Mode Decomposition (VMD) method is applied to reconstruct de-noised respiratory signals. Then, twenty time-frequency domain features are extracted from each signal, and the most optimal features are opted using the automatic Sequential Forward Feature Selection (SFFS) method. Finally, the selected features are fed into three different classifiers to distinguish between three events: normal, apnea, and others. The APNIWAVE database is used to assess the proposed SAHS detection approach. The results demonstrate an accuracy of 99.5% (with a sensitivity of 99.7%, specificity of 99.5%, and F1-score of 99.5%) in per-segment classification using a Random Forest (RF) classifier. Our method can be employed to create an affordable and reliable system for monitoring SAHS in a household setting.