Segmentation of 24-hour cardiovascular activity using ECG-based sleep/sedation and noise metrics

A method to segment cardiovascular time series is proposed using ECG-derived metrics. Segmentation of cardiovascular time series into quasi-stationary and low noise segments is important for the construction of models (based around fixed operational points) and the evaluation of a variety of indices, including cardiovascular (such as HRV) and signal quality-based metrics. Noise and activity-related segments are excluded using beat classification and ECG spectral thresholding. ECG-based cardio-respiratory and cardio-pulmonary coupling (CRC/CPC) metrics are used to determine periods of deep sleep or sedated states, amenable to model fitting and cardiovascular metric evaluation (which require quasi-stationary time series). Performance tests using a realistic ´perfectly sedate/deep sleep´ ECG model over a range of coloured 1/f^beta Gaussian noise sources (0<beta<2) show that the CPC metric is extremely robust to high levels of realistic noise with only a 7% error in classification of deep sleep/sedated states at a signal-to-noise ratio of -20dB (beta=2), 0dB (betales1). In vivo murine tests reveal a correlation between CRC and HR, and an anti-correlation with noise and activity metrics. Tests on human ECGs recorded in an intensive care unit show a similar relationship. The techniques presented in this paper may therefore provide a robust set of metrics for segmenting cardiovascular signals into quiescent and noisy/active states