Background: Intracardiac electrograms (EGMs) stored in ICDs can be used for pace mapping and ablation of clinical ventricular arrhythmia, which can’t be induced during the procedure. ICD programmer devices do not have any interface to digital EGMs; therefore, automatic computing of similarity scores between EGMs is impossible. When presented as images by programmer devices, EGMs can only be assessed for similarity by subjective eyeballing, highlighting the need for objective measures.

We report a case of ablation of premature ventricular contraction (PVC) triggering ventricular fibrillation (VF) localized by pace mapping and offline conformation using a pre-trained Siamese neural network algorithm for computing EGM image similarity.

Case report: A 72-year-old male with ischemic cardiomyopathy suffered multiple ICD shocks and syncope due to recurrent VF episodes triggered by monomorphic PVC despite therapy with amiodaron (Figure 1A). 3D-Mapping revealed a low-voltage area in the anteroseptal region of the left ventricle (LV). Pace mapping from different locations of LV was performed with an assessment of similarity between EGM of clinical VF-trigger retrieved from ICD and EGMs acquired from ICD during low-output stimulation from mapping catheter. RF ablation was conducted in the area with the most remarkable similarity between pacing-induced EGMs similar to clinical EGM-pattern assessed by eyeballing.

All pacing sites were tagged on the 3D map, and corresponding EGM images were saved for offline analysis. A pre-trained neural network calculated Manhattan similarity scores for all EGM images and clinical EGM-pattern. The region of the highest similarity (>75%) mostly matched with one manually defined by eyeballing and located in the midventricular anteroseptal segment of the LV (Figure 1B). The similarity score decreased gradually in the remote regions.

During the eight-month follow-up, the patient was free from VF episodes.

Conclusion: To our knowledge, it is the first application of automatic ICD-EGMs similarity measurement using a neural network algorithm. This technology opens new possibilities for the utilization of ICD-EGMs during the pace mapping of ventricular tachycardias.