Session P76.4

Time Series Calculation of Heart Rate Using Multi Rate FIR Filters

MR Risk*, DF Slezak, P Turjanski, RM Taborda, G Marshall

University of Buenos Aires
Buenos Aires, Argentina

The spectral analysis for heart rate variability, based on the Fourier transform, needs even sampled data. Several methods has been used to transform beat series de heart rate (uneven sampled data), in even sampled time series, including linear and spline based methods. The objectives of this study were to develop an interpolation method based of multi rate FIR filters, and to implement this method for parallel processing machines. A total of three data sets were used: a) simulated heart rate with an IPFM model, b) autonomic blockage database, and 3) long term Holter studies. The simulated heart rate used an IPFM model, the heart rate was modulated with two sines of 0.1 and 0.25 Hz; the autonomic blockage database has data from 12 healthy subjects, with a combination of postural (supine and standing) and pharmacological (atropine and propranolol) conditions; and the long term database used 10 Holter recordings of 24 hours each. A regression analysis, between spectral density of interpolated data using FIR filters and cubic splines, showed for simulated beat series an intercept 2851 ms2/Hz (P=0.001), and a slope 0.963 (P<0.001). Spectral analysis, for the autonomic blockage data set, was processed for both interpolation using FIR filters and cubic splines, the results for Bland and Altman analysis was for low frequency band (> 0.04 and < 0.15 Hz), showed a difference of -47 ms2, and standard deviation of 131 ms2; finally for the high frequency band (> 0.15 and < 0.4 Hz), the difference was 3 ms2, and the standard deviation of 48 ms2. The third data set, long term recordings were used to tests the implementation of parallel processing; the time for processing with FIR filters the 10 recordings was 1) 106 s for serial processing using R language, 2) 0.44 s for serial processing using C++, and 3) 0.06 s for parallel processing using MPI-based implementation using a cluster of 16 dual nodes. The presented method of time series calculation, using FIR filters, probed to be equivalent for both simulated and real data. This method is suitable for parallel programming implementation, such as clusters and Grid-based systems, allowing a very fast processing of long term recordings.

(Abstract Control Number: 178)