Session P73.5

Comparison of Interpolation Methods on Heart Rate Variability (HRV) Spectral Analysis of Intermittent RR-Interval Data

KK Kim*, KS Park

Seoul National University
Seoul, Korea

Heart rate data are obtained generally in process of measuring the bio-signal related heart beat such as ECG, BCG, PPG, and so on. However, these signals can be polluted due to external artifacts, and the heart rate (or RR-interval) data from the signals can become intermittent. In non-invasive and unconstrained bio-signal measurement, the artifacts can be more sensitive. For heart rate variability analysis in frequency domain, these heart rate data should be re-sampled because the heart beats irregularly and the heart rate becomes unevenly sampled data. In re-sampling process, some interpolation methods can be used. In this study, the effect of these interpolation methods on HRV spectral analysis of intermittent heart rate data through the simulation of artificial intermittent heart rate data.
Artificial heart rate data used in this study were based on McSharry’s non-stationary model. For the simulation of intermittent data, the continuous RR-intervals were removed randomly and the duration of removal was increased from 0 to 150 seconds with 5-sec interval after 1000 Monte Carlo runnings in each missing duration. In this simulation, LF/HF ratios were calculated by the HRV spectral analysis in which nearest-neighbor re-sampling (NNR), linear, cubic spline (Spline), piecewise cubic hermite (PCH) interpolation methods are used for FFT periodogram, modified periodogram, Burg algorithm, and covariance method as the spectral estimation. For comparison among the interpolation methods and spectral estimation methods, the squared errors were computed from theoretical LF/HF values of artificial heart rate model.
As the result, the squared errors for analysis with Spline interpolation are almost zero in all spectral methods without missing RR-interval data. Increasing missing duration, the errors with Spline are dramatically increased. With missing duration, the bias of squared error is smallest when the PCH method is used. For the spectral estimation methods, the covariance method with PCH shows the smallest error.

(Abstract Control Number: 176)