Recognition of antegrade and retrograde atrial activation patterns using hybrid wavelet neural network schemes

Currently used arrhythmia recognition algorithms in implantable cardioverter-defibrillators often fail in the discrimination of ventricular tachycardia with I:I retrograde conduction from sinus tachycardia. As new approach to solve this problem, the authors have developed a hybrid, wavelet-neural network scheme for a recognition of antegrade atrial activation (AA) and retrograde atrial activation (RA) patterns in endocardial electrograms (EEs). Bipolar EEs representing AA and RA were obtained during an electrophysiological examination. Consecutive beats within data segments of 10 s duration were selected. Adapted wavelet packet decompositions were applied to extract discriminating scale features in selected beats representing AA and RA. A feed forward neural network was utilized for classifying the activation patterns based on the extracted feature vectors. With the developed hybrid wavelet-neural network scheme a recognition of all analyzed AA and RA episodes was achieved