George B. Moody PhysioNet Challenge


Quick links for this year's Challenge:

Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address.

Session S41.4

CiC Challenge 2002 Undertaken By Non-Stationary And Fractal Techniques

J.C. Echeverra, M.S. Woolfson, J.A. Crowe, B.R. Hayes-Gill

University of Nottingham

Nottingham, UK

We have proposed an improvement to the detrended fluctuation analysis (DFA) by using a recursive least squares method (the alpha-beta filter) to quantify the scaling patterns of Heart Rate Variability (HRV) data. This approach was used to address the second event of the Computers in Cardiology challenge 2002. We have achieved a score of 88 points indicating that by analysing these scaling patterns it was possible to classify correctly the majority (46/50) of the data set as either real or synthetic.

Additionally, the challenge specified that each generator of the synthetic HRV data was used to produce two sequences. Notwithstanding the differences between both outputs, we have been able to identify 11 pairs of HRV sequences created by 11 different generators. Hence, the scaling patterns of synthetic data exhibited by our technique seem to reveal the intrinsic statistical properties of each HRV generator.

Moreover, the scaling patterns found for the real data sets promote further familiarisation with the scaling behaviour presented by subjects with no cardiac abnormalities. It is opportune to report here that these scaling patterns reinforce the likelihood that a uniform power-law behaviour for the long-term range, as has been suggested before, is not always manifested.

Currently, we are investigating the use of empirical mode decomposition, another non-stationary time-frequency method, that we recently evaluated for HRV analysis, to assist in the proper classification of the entire data set.

Supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH grant number R01EB030362.