Ventricular tachycardia (VT) is a life-threatening rhythm disorder and a major cause of sudden cardiac death. The clinical procedure of pace-mapping guides catheter ablation by having clinicians pace different sites in the ventricles and interpret electrocardiograms (ECGs) to locate the intervention target. Existing AI-assisted pace-mapping models require retraining for each target and cannot transfer knowledge across multiple VTs within the same patient or across patients. According to the research paper 'cAPM: Continual AI-Assisted Pace-Mapping with Active Learning' on arXiv, a team of researchers has developed a novel system called cAPM that overcomes these limitations through continual learning.
The cAPM Approach
cAPM is built on three core components: a task-agnostic surrogate neural network that learns the mapping from pacing sites to 12-lead ECG morphology; an active-learning strategy that refines this surrogate model by selecting the most informative pacing site for each target; and a continual learning strategy that sequentially processes targets while retaining knowledge from prior ones. This allows cAPM to capture and transfer knowledge accumulated from past pace-mapping data to reduce the number of pacing sites needed for future target VTs.
Evaluation and Results
The system was evaluated on an in-silico testbed consisting of sequentially-presented localization tasks across different physiological conditions and ventricular geometries. The results showed significant improvements over the state-of-the-art active-learning method. The key metrics are summarized in the table below:
| Metric | cAPM (without replay) | cAPM (with replay) | State-of-the-art Active Learning |
|---|---|---|---|
| Probability of localizing within 5 mm tolerance | 81% | 81% | 38% |
| Average number of pacing sites required | 4.5 | 4.5 | 13.7 |
As shown, cAPM achieved an 81% probability of localizing within clinical tolerance (5 mm accuracy) using only 4.5 pace-mapping sites, compared to the state-of-the-art active-learning method achieving 38% probability using 13.7 pacing sites. This represents a 67% reduction in the number of pacing sites needed.
Active Learning and Continual Learning
The active-learning strategy in cAPM selects the most informative pacing site for each target VT, thereby minimizing the number of pacing locations. The continual learning strategy ensures that knowledge from previously mapped targets is retained and applied to new targets, preventing catastrophic forgetting. The authors note that cAPM with and without replay of past data samples performed equally well, suggesting that the surrogate model effectively captures essential patterns from prior tasks.
Implications for Future Clinical Studies
The research team states that these results 'provide a strong basis for preparing cAPM towards in-vivo preclinical and clinical studies where it can be used to guide pace-mapping.' The efficiency gains could reduce procedure time and improve patient outcomes by enabling faster and more accurate target identification during catheter ablation.
Key persons involved in the study include O'Hara, Dylan; Bajracharya, Pradeep; Meisenzahl, Casey; Gillette, Karli; Prassl, Anton J; Plank, Gernot; Nazarian, Saman; Tung, Roderick; Sapp, John L; and Wang, Linwei. The paper is available on arXiv under a Creative Commons Attribution-ShareAlike 4.0 International license.
The continual learning approach demonstrated in cAPM holds potential beyond cardiac electrophysiology. The combination of surrogate models, active learning, and knowledge transfer can be applied to any domain where sequential decision-making with limited data is critical, such as robotics, materials science, or logistics optimization. However, these applications are not yet demonstrated in the current work.
