Dynamical differences obtained from different simultaneously recorded leads of ECG signals

In a prelominary study we investigated the relationship between ECG signal morphology and the Correlation Dimension, D₂ ,  in different ECG leads taken from the body surface [3]. In pathological signals (patients with supraventricular arrhythmia, SVA, from the MIT-BIH data base - tapes 203, 209 and 222, sampled at 360 Hz), we first applied a Parks-McClellan FIR low pass (100 taps, 85 Hz), and subsequently decimated by 4. then we computed D₂ from 10000 points (subsampled at 90 Hz) - i.e. for 1 minute and 51 sec. In order to verify stability of the invariant, we computed  D₂ 3 times for each subject (after 1 000, 50 000 and 10 000 samples). The results show that, in some cases, morphological differences of the signal are related with dramatic changes in the correlation dimension estimates. More importantly, D₂ was significantly different in different leads in patients prone to SVA. This was not observed in preliminary analysis of data obtained from normal subjects, neither was the case in subjects from the MIT data base with predominant normal sinus rhythm (112 and 117), and indicates changes in the underlying dynamical system. Such differences were much smaller in normal patients.
 

Using Wavelets to separate dynamical activities embedded in a time series

We further analyzed synthesized ECG signals and signals from the MIT-BIH data base (tapes 207 and 115), preceding Ventricular Flutter [2]. We separated dynamical activity in decorrelated subbands using multiresolution wavelets. By means of the Correlation Dimension  D₂ we were able to show that periods of low dimensional activity precede the actual incidence of tachyarrhythmia [2].
 
 

Dynamic Differences Between Ventricular Fibrillation Types Induced in Human Patients by Different Types of Stimulation

In another study we made an attempt to differentiate between different types of VF, induced in human patients during ICD implantation in order to test the device's detection algorithm and the appropriateness of the therapeutic energy settings [1]. The goal of this preliminary work was to make a rough assessment as to whether the most commonly used induction methods (50 Hz stimulation or T-shock) produced the same type of VF, and if not, then which one would cause VF closer to spontaneous VF. Spontaneous VF (MIT data base) typically has Correlation Dimension  D₂> 2, or a scaling region would be absent. This indicates, that no low-dimensional attractor underlies these time series. VF caused by 50 Hz stimulation and by T-shock often showed  D₂ < 2, i.e. certain periodical dynamic was present. On the whole, T-shock induced VF of higher  D₂ than 50 Hz, often over 2, and sometimes even above 3, thus being closer to spontaneous VF. Although we did not have spontaneous VF from the same patient, our study provided evidence, that T-shock might be the more appropriate method of VF induction [1].

Relevant publications:

1. Casaleggio, A., Gramatikov, B., Thakor N. V.    Dynamic Differences Between Ventricular Fibrillation Types Induced in Human Patients by Different Types of Stimulation.   Computers in Cardiology 1997, Vol. 24, Lund, Sweden, Sept. 7-10, 1997, Publ. IEEE, Inc, pp.89-92 20.

2. Casaleggio, A., Gramatikov, B., Thakor, N.    On the use of Wavelets to separate dynamical activities embedded in a time series.   Computers in Cardiology, Indianapolis, Sept. 8-11, 1996, Publ. IEEE, Inc, 181-184.

3. Gramatikov, B., Casaleggio, A.      On the dynamical difference obtained from different simultaneously recorded leads of ECG signals.   Proceedings of the Bulgarian Seventh National Conference on Biomedical Physics and Engineering (with international participation), Sofia, 17-19 Oct 1996, in English, pp. 110-113.