SCM

Forum: help

Monitor Forum | Start New Thread Start New Thread
RE: Differences between RHRV and KUBIOS [ reply ]
By: Abraham Otero on 2018-10-26 07:34
[forum:46375]
As we explained in the preface to the book on RHRV, one of the main hindrances in the HRV research field is the difficulty in reproducing results from other researchers. This is due to the lack of standardization in the values of many parameters and other implementation details in the HRV algorithms. Some examples are how exactly ectopic beats are filtered, the algorithm used to interpolate the RR intervals to obtain a time series of constant sampling frequency, how to remove the DC component (from all the RR series, from each window, etc.), the window type (Hamming, Welch, etc.), window size, and window overlap used in the Fast fourier transform, or the mother wavelet used. Explicitly or implicitly, in any HRV analysis dozens of decisions on either parameters or implementation details are made. Some of these decisions are difficult to document in a scientific paper. But they are essential for the faithful and accurate reproduction of the results.

RHRV, among other things, attempts to address these problems. By simply posting the RHRV analysis script as supplementary material of a paper, the reproduction of the results over the same, or over new data, will be trivial: just run the script. Being RHRV an open and free package, no one should have any impediment to reproduce the results.

Answering more directly to your question, since Kubios is not open source and we can not see its source code, it is difficult to know exactly what parameters we are before what are the differences with RHRV

Differences between RHRV and KUBIOS [ reply ]
By: Baracat Bruno on 2018-10-25 15:06
[forum:46374]
By doing the same analyzes with RHRV and Kubios we arrive at identical results in the time domain but different in a ratio of 1/3 for the values of VLF, LH and HF computed vith “CalculatePowerBand” (e.g Kubios / RHRV ≈ 3). This with real data cleaned for any artefact, but also with dummy data which we know a priori the values of variability.
With Matlab (Kubios is programmed in Matlab) and the parameters below we arrive at the same results as Kubios.
Where can this underestimation of RHRV come ?

Settings in Matlab:
- Window Hanning
- Spline resampling, frequency 4Hz
- N FFT = 1024
- cw = (1 / N) * sum (wdw. ^ 2) with wdw = hanning (length (nnint)) and NNint = Number of NNs
- PSD = (abs (fft (NNint, N)) ^ 2.) / (N * fs * cw)
- PSD in the VLF, LF and HF bands obtained by the sum of the PSDs
- VLF between 0.00 and 0.05; LF between 0.05 and 0.15; HF between 0.15 and 0.45

Settings in CalculatePowerBand
- Size=256
- Shift =128
- Same band as in Matlab

With Kubios we specify a window of 256 and 50% of overlap.

Tks

Thanks to:
Vienna University of Economics and Business University of Wisconsin - Madison Powered By FusionForge