Arrhythmogenic sites mapping in post-ischemic ventricular tachycardia using a Siamese neural network

Ventricular tachycardia (VT) is a life-threatening arrhythmia commonly treated by catheter ablation guided by substrate mapping. This procedure relies on the cardiologist's visual inspection of intracardiac electrograms (EGMs) to identify arrhythmogenic sites. This task is challenging due to the huge amount of data to interpret. To address this issue, we proposed a method for discriminating between physiological and anomalous bipolar EGMs based on siamese neural networks (SNN), able to deal with small datasets, for the automatic labeling of the EGMs onto the electroanatomic map. On a balanced dataset of 1504 physiological and anomalous EGMs, from nine post-ischemic VT patients, we demonstrated that a SNN trained to distinguish between the two types of EGMs is able to achieve a high degree of specificity (91±3 percent) and sensitivity (93±3 percent). Potentially, the proposed approach could also be exploited to map the similarity of the EGMs, resulting in a novel electroanatomic map for the identification of areas of abnormal conduction.