Detection and estimation of T wave alternans with matched filter and nonparametric bootstrap test

Though alternans phenomena in the cardiac repolarization phase has been shown to be related to arrhythmgenesis, a definitive estimation method from the T wave of ECG recordings is not yet available. We propose a statistical signal processing scheme which compares the T-wave morphology of even and odd beats by using a running matched filter, in order to increase the signal to noise ratio of the estimation. Given that previously proposed hypothesis tests for alternans detection rely on the knowledge of noise statistical distribution, we also analyzed the usefulness of a nonparametric bootstrap test. Data set composed of 100 ECG recordings included in the Challenge Database were used. Principal Component Analysis was previously made for multilead recordings. Subsequent preprocessing for each available lead consisted of conventional baseline removing, filtering, R-wave detection, exclusion of too noisy segments, T-wave segmentation, and template generation for even and odd beats. The difference between the template and a given beat was obtained by minimizing the absolute error of their comparison with a windowed circular shift. A paired bootstrap resampling test was made for deciding whether the averaged differences between the template and the T-waves were significant compared to the noise level.