Sleep Quality Analysis with Cardiopulmonary Coupling

Abstract

Sleep quality is commonly assessed with subject self-reporting, interviews and psychological variables. However, more precise methods comprise estimation of physiological signals where polysomnography is considered to be the gold standard and can be performed to produce qualitative or quantitative measurements regarding the subjects sleep. However, polysomnography is known to be a complex and expensive process that is inaccessible to a large group of the population. The main objective of this work is to produce an algorithm that is capable to ascertain sleep quality, using only data from a single-lead electrocardiogram, that can be implemented in a small digital signal processor or a small field-programmable gate array. Therefore, providing a tool that can be used for identifying the need of a polysomnography analysis or as a complementary method to improve the detection capability of other pathologies. The employed method first determined both normal-to-normal sinus interbeat interval series and electrocardiogram derived respiration signal, producing the cardiopulmonary coupling with the cross-spectral coherence method. This information was fed to a classifier to determine the non-rapid eye movement (non-REM) and cyclic alternating pattern (CAP) periods. Cyclic alternating pattern rate was then computed to provide a qualitative measure of sleep quality, considering the age-related cyclic alternating pattern rate percentages in healthy subjects as reference threshold. The employed method provided a reasonable non-rapid eye movement sleep classification (70% accuracy) but a poor cyclic alternating pattern detection capability (62% accuracy). From the ten tested classifiers deep stacked autoencoder produced the best results and the average absolute difference of predicted and true cyclic alternating pattern rate was 15%, classifying correctly 67% of the subjects, regarding their sleep quality (either good or bad).