Improved non-invasive detection of sleep stages when combining skin sympathetic nerve activity and heart rate variability analysis with AI.

Sleep is a cyclic physiological process that goes into different stages, and every stage has its' importance in the construction or recovery of physiological function. Sleep scoring is performed from polysomnography recordings which requires signals from multiple sensors in a specially equipped Sleep Lab. This is expensive, and normal sleep behavior may be affected due to the new environment and extensive testing equipment. Skin sympathetic nerve activity (SKNA) has been shown to vary between sleep stages, and it can be recorded simultaneously with the ECG on the skin using conventional ECG patch electrodes. In this study, we propose that sleep stages can be classified using features derived from the ECG and SKNA recordings. The study was performed on 21 subjects, and we initially extracted 14 heart rate variability (HRV) and 11 SKNA features, and then selected the 17 most relevant (12 HRV, 5 SKNA) features out of 25. We evaluated both individual and combined performance of HRV and SKNA for classification of 5 sleep stages, 3 stages and binary stages. Our study showed that the addition of SKNA information with HRV provides an improved recognition accuracy of 91.57%, 95.36%, and 96.14%, respectively for 5, 3, and binary sleep stages. The SKNA shows similar performance to the HRV for 2 and 3 sleep stages recognition. This AI-powered sleep classification system might provide an advancement in the development of a real-time sleep monitoring device with low computing power that can be used for screening sleep disorders in a large population.