Correlation dimension estimation from electrocardiograms

In this work the correlation dimension (D2) of electrocardiograms (ECG) has been computed on ECG recordings from the MIT-BIH Arrhythmia Data Base: 28 signals have been studied both analysing changes in D2 estimations obtained processing different time-slices within the same ECG signal, and comparing the D2 values computed from different subjects. We observed a greater time variation in the estimated D2 from pathological subjects, while D2 seems to be more repeatable in the case of healthy ones. Instead there is no clear evidence of significative differences in the D2 values computed from normal and sick subjects. Since the presence of noise affecting the signal induces significant errors in D2 estimates, we proposed the computation of the parameter alpha which gives us a rough information on the noise level.