Comparison of heart rhythm and morphological ECG features in recognition of sleep apnea from the ECG

This study addresses the problem of sleep apnea recognition on a minute-by-minute basis from single-lead ECGs recorded overnight. Analysis of heart rate fluctuations, quantified by the series of RR-intervals, is compared to analysis of ECG morphology variations, assessed using signal vectors from the QRS- and the T-wave region and projecting them onto their first principal component. The resulting series of scalar Karhunen-Loeve coefficients (KLCs) were used as descriptors of morphology. From the derived series, we calculated a measure of similarity and a spectral index in temporal segments of 5 min, and assessed their diagnostic accuracy by ROC-analysis. Although the performance for the RR-series and the similarity feature was 81%/84% sens ./spec., better results up to 87%/87% were obtained from the T-wave KLCs. It is concluded that the effects of sleep apnea on the ECG are reflected more uniformly in morphology variations of the ECG compared to heart rhythm.