Session P94.6

An Alternative Decision Rule for Threshold Based T-Wave Measurement Algorithms Based on Second Derivative Extrema

PV Rivera Farina*, P Laguna, J Pérez Turiel, A Herreros, S Wong

Fundación CARTIF
Valladolid, Spain

QT interval is a surface ECG measure which has being subject of great research interest. Its accurate measure is very relevant as a non-invasive index of cardiac risk. This is of utility for drug toxicity quantification, VF risk assessment; Ischemic cardiopathy as in diabetes mellitus because of the increased difficulty the cells have to metabolize glucose and then generating related QT prolongation. Usually a prolongation of the QT interval beyond the normal cause is associated with bad prognosis.
In this paper we revisit the wavelet transform based method, and propose an alternative decision rule for the T wave end marking. Rather than using a threshold related or the highest inflexion point of the derivative, we use the extreme on the second derivative which first appear ahead of the inflexion point. This rule differs from the thresholds one in its simplicity to be applied and its potential for real time analysis.
The algorithm detects the end of the T-wave by using the first and second derivative on the fifth scale. Results for the T wave obtained in simulations give a mean error of -6.73 +/- 14.5 ms (-7.80 +/- 21.04 ms when adding noise in a SNR of approximately 4 DB) whereas results in Physionet's QTDB give (-1.22 +/- 38.85) in the case of T end location. The results for the threshold method were (-1.6 +/- 18.1).
In addition we had proved that the proposed rule is a simplified approximation of another algorithm proposed in
[1] Q. Zhang, A. I. Manriquez, C. Médigue, Y. Papelier, and M. Sorine An Algorithm for Robust and Efficient Location of T-Wave Ends in Electrocardiograms, IEEE Trans. Biomed. Eng. vol. 53, no. 12, december 2006
Algorithm which reports 1.72+/- 41.27 ms results when applied to the physionet database. This implies that the simplified framework here proposed for [1] result in lower dispersion and so more stable measurements are obtained than at the original. The enlarged dispersion compared to the threshold based ones (38 to 21 ms) could in real time application be paid by the much higher simplicity specially in terms of rejecting the add-hoc rules need in the threshold based algorithms.

(Abstract Control Number: 11)