Session PB9.6

Detection of Shockable and Non-Shockable Rhythms in Presence of CPR Artifacts by Time-Frequency ECG Analysis

JP Didon*, I Dotsinsky, I Jekova, V Krasteva

Schiller Medical SAS
Wissembourg, France

Continuous cardiopulmonary resuscitation (CPR) is advised to improve the survival rate in out-of-hospital cardiac arrests (OHCA). The shock advisory system of automatic external defibrillators (AEDs) should be designed to work adequately in presence of CPR artifacts. This paper aims to investigate time-frequency domain features of CPR artifacts and cardiac arrhythmias. An automatic classification system for AEDs is built to detect shockable and non-shockable rhythms in CPR-contaminated electrocardiogram (ECG). The accuracy of the algorithm is tested with human ECG signals. In the first part of the study, FFT spectra of pure CPR artifacts (CPR on asystole) were compared to the spectra of noise-free ECG arrhythmias (shockable and non-shockable). For the best separation in the frequency domain we suggested three band-pass filters: one preserving the frequency components within the range of 2-3 Hz (own to the basic CPR wave), the second within the range 4-7 Hz (own to the VF waves) and the third for the range of 10-20 Hz (own to the QRS complexes). Similarity of the consecutive compression-to-compression periods (CPR patterns) was analysed at the output of 2-3 Hz filter. In contrast, VF waves irregularity was enhanced by 4-7 Hz filter, while the presence of the QRS complexes was related to high-amplitude peaks at 10-20 Hz filter output. In the second part of the study, a shock-advisory system was designed to analyse CPR-contaminated ECG episodes of 10 seconds. Several criteria, related to the morphology of the signal, were calculated within the above defined frequency bands. Moreover, the underlying ECG signal was restored using the CPR patterns. A decision tree based on fuzzy logic was built to recognize shockable and non-shockable rhythms under CPR. The shock-advisory system was tested with a CPR dataset, taken from OHCA interventions with AEDs (FredEasy, Schiller Medical SAS, France) on 168 patients (July-December 2006). Sensitivity is: 94.2 % (155 VFs), 100 % (6 shockable VTs), 73 % (37 fine VFs). Specificity is: 87 % (284 non-shockable rhythms), 83.7 % (386 asystoles), 42.1 % (19 non-shockable VTs). The sensitivity of the method meets the AHA recommendations. Non-shockable rhythm recognition shows results equivalent to what is reported in literature.

(Abstract Control Number: 80)