Towards short-term forecasting of ventricular tachyarrhythmias

Abstract

This thesis reports the discovery of spectral patterns in ECG signals that exhibit a temporal behavior correlated with an approaching Ventricular Tachyarrhythmic (VTA) event. A computer experiment is performed where a supervised learning algorithm models the ECG signals with the targeted behavior, applies the models on other signals, and analyzes consistencies in the results. The procedure was successful in discovering patterns that happen before the onset of a VTA in 23 of the 79 ECG signal segments examined. A database with signals from healthy patients was used as a control, and there were no false positives on this database. The patterns discovered by this modeling process, although promising, still require thorough external validation. An important contribution of this work is the experimental procedure itself, which can be easily reproduced and expanded to search for more complicated patterns.