Fractal analysis of cardiorespiratory signals for sleep stage classification

Abstract

Cardiorespiratory polygraphies do not allow the traditional sleep scoring. Therefore this study evaluated whether fractal dimension (FD) analysis of ECG and respiration (RSP) provides information on sleep stages. We considered two sleep-scored overnight full-polysomnographies. R-R intervals (RRI) from the ECG and RSP were resampled (4 Hz) and normalized to unit variance. FD of a segment of N samples of m signals is log(N-1)/[log(N-1)+log(d/L)], with L length of the trajectory in the m-dimensional space, d its extension. Monovariate (m=1, FDrri and FDrsp) and bivariate (m=2, FDrri, rsp) fractal dimensions were estimated over running windows of 15 s, and averaged over wake, lighter and deeper NREM stages, and REM. Unlike FDrri and FDrsp, the bivariate FDrri, rsp showed the same behavior in both subjects, being lowest in wake, increasing with the depth of NREM sleep and decreasing slightly in REM. This suggests that bivariate FD can provide information for sleep scoring of cardiorespiratory polygraphies.