Session S84.6

Detection and Estimation of T Wave Alternans with Matched Filter and Nonparametric Bootstrap Test

JL Rojo-Alvarez*, O Barquero-Perez, I Mora-Jimenez, R Goya-Esteban,
J Gimeno-Blanes, A Garcia-Alberola

Universidad Rey Juan Carlos
Madrid, Spain

Alternans phenomena in the cardiac repolarization phase has been shown to be related to arrhythmogenicity. However, a definitive method for its estimation and detection from the T wave of ECG recordings is not yet available. We propose a statistical signal processing scheme aiming to overcome two of the major limitations of currently proposed schemes. First, many previous methods look for the most patent difference in amplitude between even and odd beats. However, alternans can be present in any part of the T-wave morphology and therefore, comparison should be made in terms of the differences throughout all the T wave length, in order to increase the signal to noise ratio of the estimation. Second, few hypothesis tests have been proposed to detect the actual presence of alternans, and still they rely on a pre-assumed statistical shape. We used a data set composed of the 100 ECG recordings included in the Challenge, in which the presence of alternans was blinded to us. As a first approach, ECGs from a single lead were selected for each case. Baseline was removed by using a median filter together with spline interpolation, and high frequency noise was filtered out with a 64th order low-pass filter. T waves were segmented by using an automatic method for R wave detection, and then taking the segment between 60 ms after the R wave and 50% of the estimated cycle length of the episode. T wave templates were separately generated for even and odd beats, and the difference between the template and a given beat was obtained by minimizing the absolute error of their windowed circular shift comparison. Increased separability in synthetic examples was clearly obtained when building two separate error series, corresponding to the estimation errors of the even template with the odd beats, and vice versa. The statistic for estimating the alternans of an episode was the absolute averaged difference between both series. In order to have a nonparametric hypothesis test for comparison, a paired bootstrap resampling test was made for deciding whether the estimated averaged T wave alternans was significantly different from the noise level. Results. For the estimation scheme, 78 cases had amplitude lower than 10% of the maximum observed (low level), 12 cases had moderate alternans (between 10% and 30%), and 10 had larger values. Nonparametric hypothesis test showed only 6 significantly different episodes of alternans. Results will be further explored by improving preprocessing and segmentation stages, by jointly considering the available channels, and by reducing the time window of the hypothesis test (shorter than two minutes).

(Abstract Control Number: 361)