Abstract

This study examines the utility of recurrence analysis in the field of sleep apnea detection from heart rate series. In 38 patients, ECGs were recorded during sleep in parallel to polysomnography (PSG). Parameters extracted from distance matrices and from recurrence plots (RPs) of the time-delay embedded RR series were tested for their suitability to detect minute-by-minute phases of sleep apnea by means of receiver operating curve (ROC) analysis against the findings of the PSG. Features derived from the continuous distance matrix yielded better results (sensitivity 77%; specificity 78%) than those quantifying the structure of the binary RP (67% and 74%, respectively), accentuating the significance of information on the magnitude of heart rate variations for sleep apnea detection. However, simpler and computationally less demanding spectral techniques yield largely comparable results (77% and 72%, respectively). This raises the question as to whether recurrence analysis of RR series yields additional insight into sleep apnea recognition from heart rate variations.