Automatic Recognition of Ventricular Late Potentials in Intracardiac Electrograms

Giulia Baldazzi1, Marco OrrĂ¹2, Mirko Matraxia3, Graziana Viola4, Danilo Pani2
1DIBRIS, University of Genova; DIEE, University of Cagliari, 2DIEE - University of Cagliari, 3Medical Concept Lab, 4Department of Cardiology, San Francesco Hospital


Aims: Ventricular late potentials (VLPs) are low-amplitude electrical signals that appear in intracardiac electrograms with an unpredictable delay with respect to the QRS complexes. They are thought to be associated to delayed conduction pathways in damaged myocardium tissue and thus to reentry ventricular arrhythmias. As such, spatial localization of VLPs can be exploited by cardiologists for the identification of the ablation areas. Even though some electro-anatomical mapping systems provide automatic recognition features based on the amplitude of the signal after the QRS, more advanced techniques looking at the time-frequency characteristics of the VLPs compared to the normal electrogram could improve the performance of such tools. In this work, a novel automatic approach for a reliable detection of VLPs in intracardiac electrograms is proposed. Methods: 46 signals from five patients, affected by post-myocardial infarction ventricular tachycardia, were used. Intracardiac electrograms from routine procedures were acquired by the CARTO3 System at the San Francesco Hospital (Nuoro, Italy). Radiofrequency ablation followed the usual clinical protocols. Retrospective analysis of such signals led to the annotation of VLPs by an expert cardiologist, to be used for a supervised classifier training and test. The automatic VLPs detection is based on an initial denoising procedure, followed by a time-scale approach based on the continuous wavelet transform (CWT). CWT was used to decompose at multiple scales each segment of the electrograms. A feature signal was created from such a decomposition, by estimating the average power of CWT coefficients. Different morphological features were then extracted from this signal and used to feed an Ensemble Bagged Tree classifier to discern between physiological QRSs and VLPs. Results: By using the proposed automatic approach, the VLPs recognition accuracy exceeded 90%, with a false negative rate under 5%. This study paves the way to deeper analyses on larger datasets.