A robust T wave alternans detector based on the GLRT for Laplacian noise distribution

T wave alternans (TWA) has been proposed as a marker of cardiac instability and high risk for malignant ventricular arrhythmias and sudden cardiac death. Several algorithms have been used to detect and estimate TWA, such as the spectral method (SM), complex demodulation (CD) or correlation method. In this work, we show that SM and CD methods can be understood as a Generalized Likelihood Ratio Test (GLRT) for TWA episodes in Gaussian noise. However, the noise distribution in ECG recordings is more 'heavy-tailed' than Gaussian noise due to outliers (eg artifacts, impulsive noise, baseline wandering, or ectopic beats). In this paper, we derive a similar GLRT TWA detector and amplitude estimator for Laplacian noise. The resulting estimator is, in fact, a "median filtered complex demodulation". Simulation results suggest that this new approach is more robust than CD or SM when physiological noise is present. The effect is clearer when the noise distribution has heavier tails, as in the case of muscular noise.