Model-Based Detection of Heart Rate Turbulence Using Mean Shape Information

A generalized likelihood ratio test (GLRT) statistic is proposed for detection of heart rate turbulence (HRT), where a set of Karhunen-LoEve basis functions models HRT. The detector structure is based on the extended integral pulse frequency modulation model that accounts for the presence of ectopic beats and HRT. This new test statistic takes a priori information regarding HRT shape into account, whereas our previously presented GLRT detector relied solely on the energy contained in the signal subspace. The spectral relationship between heart rate variability (HRV) and HRT is investigated for the purpose of modeling HRV "noise" present during the turbulence period, the results suggesting that the white noise assumption is feasible to pursue. The performance was studied for both simulated and real data, leading to results which show that the new GLRT detector is superior to the original one as well as to the commonly used parameter turbulence slope (TS) on both types of data. Averaging ten ventricular ectopic beats, the estimated detection probability of the new detector, the previous detector, and TS were found to be 0.83, 0.35, and 0.41, respectively, when the false alarm probability was held fixed at 0.1.