Simulation study and performance evaluation of T-wave alternans detector

Abstract

Several methods for T-wave alternans (TWA) detection have been proposed so far. Here, a simulation study of T-wave alternans is proposed in order to compare the performance of TWA detectors, as well as to quantify the performance characteristics in terms of sensitivity, positive predictivity and beat-to-beat amplitude estimation accuracy. The FFT-based approach is the most extended TWA detection method, although it does not provide any beat-to-beat amplitude estimation. Complex Demodulation (CD) and Correlation Method (CM) were applied and also two new methods, Capon filtering (CF) (which is a variant of CD) and another approach based on Karhunen-Loeve Transform which also includes CF. In this study, ECG signals are simulated repeating a single beat, and adding noise from 4 different noise sources at different levels. TWA episodes with different amplitudes and waveforms were added to the signals. One can conclude from this study that the CM obtained much lower detection rates than the other methods, especially in sensitivity for low amplitude TWA. The amplitude accuracy in the detected episodes was also found to be the worst with CM. The Alter-based approaches (CD, CF and KLT+Capon) have shown a similar behavior. However, the KLT+Capon method needs a much higher computational complexity. In general, alternans are detected with rates near 100% for SNR greater than 10 dB for physiological noise sources. A new TWA information is introduced: TWA waveform along the ST-T complex that could be related to the alternans source localization inside the heart.