cAPM Improves Cardiac Pace-Mapping Efficiency with Continual AI

Ventricular tachycardia (VT) is a severe heart rhythm disorder, and pace-mapping is a crucial clinical procedure for identifying the precise target for catheter ablation. This process traditionally involves clinicians manually pacing various ventricular sites and interpreting electrocardiograms, which is time-consuming and requires rapid decision-making. While active learning AI models have shown promise in guiding pacing, they typically require complete retraining for each new target VT, limiting their efficiency and scalability. This research introduces cAPM (continual AI-assisted pace-mapping), a novel framework designed to overcome these limitations by enabling knowledge transfer across multiple VTs within the same patient and even across different patients. cAPM integrates a task-agnostic surrogate neural network that learns the mapping from pacing sites to ECG morphology, an active learning strategy to select the most informative pacing sites, and a continual learning component to sequentially retain and transfer knowledge. Evaluations on an in-silico testbed demonstrated cAPM's superior performance. It achieved an 81% probability of localizing within clinical tolerance (5 mm accuracy) using an average of only 4.5 pacing sites. This significantly outperforms state-of-the-art active learning methods, which achieved only 38% probability with 13.7 pacing sites. These results highlight cAPM's potential to revolutionize pace-mapping by making it substantially more efficient and accurate.