Model-Based Detection of Heart Rate Turbulence

In this study, the integral pulse frequency modulation model is extended to account for the presence of ectopic beats and heart rate turbulence (HRT). Based on this model, a new statistical approach to the detection and characterization of HRT is presented. The detector structure involves a set of Karhunen–LoÈve basis functions and a generalized likelihood ratio test statistic $T({\bf x})$ . The three most significant basis functions reflect the difference in heart rate prior to a ventricular ectopic beat (VEB) compared to after HRT, the “average” HRT, and a delayed contribution to HRT, respectively. Detector performance was studied on both simulated and ECG signals. Three different simulations were performed for the purpose of studying the influence of SNR, QRS jitter, and ECG sampling rate. The results show that the HRT test statistic $T({\bf x})$ performs better in all simulations than do the commonly used parameters known as turbulence onset (TO) and turbulence slope (TS). In order to attain the same performance as $T({\bf x})$, TS needs at least twice the amount of VEBs for averaging, and TO at least four times. The detector performance was also studied on ECGs acquired from eight patients who underwent hemodialysis treatment with the goal to discriminate between patients considered to be hypotension-resistant (HtR) and hypotension-prone (HtP). The results show that $T({\bf x})$ exhibits larger mean values in HtR patients than in HtP, suggesting that HRT is mostly present in HtR patients. The overlap between the two groups was larger for TO and TS than for $T({\bf x})$.