Session S73.2

Comparison of Countershock Prediction Features Based on Autoregressive and Fourier Transformed Spectral Analysis

CN Nowak*, G Fischer, A Neurauter, L Wieser,
B Tilg, HU Strohmenger

UMIT
Hall, Austria

Background: Electrical defibrillation is the treatment of choice of ventricular fibrillation (VF), whereas the timing of the therapy in emergency case is an important factor for increasing the probability of a return of spontaneous circulation. Recent findings indicate that spectral analysis of the VFECG might guide countershock therapy, where the Fast Fourier Transform (FFT) is the standard spectral estimator. From a computational point of view spectral estimators based on autoregressive (AR) modeling compute the spectrum with less computation time and less memory requirement, which play an important role when implemented in a commercial defibrillator. Thus, this study compares the predictive power and the calculation time of parameters obtained by FFT and AR methods.
Methods: The evaluation was carried out in an animal model of VF. In total, 41 shocks were delivered in 25 swine. For the calculation time comparison, a commercial microcontroller was implemented with speed improving algorithms. Two new AR based prediction features, called spectral pole power (SPP) and spectral pole power with dominant frequency weighing (SPPDF), are developed in this study. For validation purposes, the centroid frequency (CF) and the amplitude spectrum area (AMSA) that are two previously studied prediction features calculated with FFT are used.
Results: Calculating the area under the receiver operating characteristic (ROC) curve (AUC), the prediction power of the features based on AR modeling yields better results compared to the established parameters using FFT for their spectral calculation. While the ROC AUC values of CF and AMSA amount 72.1 % and 78.0 %, the predictive power of the AR based parameters has values of 85.9 % and 89.4 %, respectively. Moreover, the calculation time of the new parameters is also nearly 2.5 times faster than parameters with FFT based techniques.
Conclusion: Summing up, AR spectral estimators are a further option of calculating countershock prediction features from the spectra of VFECG compared to the FFT due to the higher predictive power. The benefits of reduced computational load of the AR estimation can also be investigated, when implementing those prediction features on a microprocessor.

(Abstract Control Number: 98)