George B. Moody PhysioNet Challenge

Logo

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 physionetchallenge.org. 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 S41.2

Classifying Simulated And Physiological Heart Rate Variability Signals

N. Wessel, H. Malberg, A. Schirdewan, J. Kurths

University of Potsdam

Potsdam, Germany

Physiological data very often show complex structures which cannot be interpreted immediately. Therefore, the simulation of such time series seems to be extremely sophisticated. However, the classification of physiological and simulated time series should be possible. Hence, the data we are analysing here are the 50 time series from the challenge organised by PhysioNet and Computers in Cardiology 2002. The main intention of this contribution is to sketch our way of discriminating both types of time series. Our approach is rather simple: we exclude time series which show non-physiological behavior. The first decision rule is “The distribution of the RR-intervals is too small”, quantified by the Shannon-entropy of the histogram (exclusion if the entropy is less than 2.8). Next decision rule “No beat-to-beat variability changes day vs. night, rest vs. stress” quantified by the 24h variability of the parameter pNNl10, which gives the percentage of beat-to-beat differences lower than 10 ms (exclusion if pNNl10<0.07). The final decision of the remaining time series was made using the symbolic dynamics approach. The simulated time series showed lower word variability than the physiological, which was quantified by the parameter wsdvar. Using this rules we got an score of 100 that means we could completely discriminate both time series groups. Thus, the intricate interdependencies of variations at different scales in heart rate variability data are still difficult to simulate to mislead an experienced observer.


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

Back