Detection of T wave alternans in non-stationary noise: a GLRT approach

T wave alternans (TWA) has been proposed as a marker for cardiac instability and high risk of malignant ventricular arrhythmias. In this work, we analyze a generalized likelihood ratio test (GLRT) approach to TWA detection. We used several noise models considering Gaussian and Laplacian distributions as well as three stationarity degrees. Another novelty of the proposed detectors is the use of the data from the P wave. The GLRT detectors for all the models were derived and implemented. Their detection performance was evaluated using real ECG signals with simulated TWA, showing that models accounting for the non-stationarity of the noise obtained best results. We also found that Laplacian-based detectors outperformed those based on Gaussian noise assumption.