Efficacy of Spectral Signatures for the Automatic Classification of Abnormal Ventricular Potentials in Substrate-Guided Mapping Procedures

Several peculiar spectral signatures of post-ischaemic ventricular tachycardia (VT) electrograms (EGMs) have been recently published in the scientific literature. However, despite they were claimed as potentially useful for the automatic identification of arrhythmogenic targets for the VT treatment by trans-catheter ablation, their exploitation in machine learning (ML) applications has been not assessed yet. The aim of this work is to investigate the impact of the information retrieved from these frequency-domain signatures in modelling supervised ML tools for the identification of physiological and abnormal ventricular potentials (AVPs). As such, 1504 bipolar intracardiac EGMs from nine electroanatomic mapping procedures of post-ischaemic VT patients were retrospectively labelled as AVPs or physiological by an expert electrophysiologist. In order to assess the efficacy of the proposed spectral features for AVPs recognition, two different classifiers were adopted in a 10-time 10-fold cross-validation scheme. In both classifiers, the adoption of spectral signatures led to recognition accuracy values above 81%, suggesting that the use of the frequency-domain characteristics of these signals can be successfully considered for the computer-aided recognition of AVPs in substrate-guided mapping procedures.