Central sleep apnea detection from ECG-derived respiratory signals. Application of multivariate recurrence plot analysis.

OBJECTIVES:This study examines the suitability of recurrence plot analysis for the problem of central sleep apnea (CSA) detection and delineation from ECG-derived respiratory (EDR) signals. METHODS:A parameter describing the average length of vertical line structures in recurrence plots is calculated at a time resolution of 1 s as 'instantaneous trapping time'. Threshold comparison of this parameter is used to detect ongoing CSA. In data from 26 patients (duration 208 h) we assessed sensitivity for detection of CSA and mixed apnea (MSA) events by comparing the results obtained from 8-channel Holter ECGs to the annotations (860 CSA, 480 MSA) of simultaneously registered polysomnograms. RESULTS:Multivariate combination of the EDR from different ECG leads improved the detection accuracy significantly. When all eight leads were considered, an average instantaneous vertical line length above 5 correctly identified 1126 of the 1340 events (sensitivity 84%) with a total number of 1881 positive detections. CONCLUSIONS:We conclude that recurrence plot analysis is a promising tool for detection and delineation of CSA epochs from EDR signals with high time resolution. Moreover, the approach is likewise applicable to directly measured respiratory signals.