cAPM: Continual AI-Assisted Pace-Mapping with Active Learning

Researchers introduce cAPM, an AI-assisted system that uses continual learning and active learning to improve cardiac pace-mapping procedures for treating ventricular tachycardia. The system demonstrates 81% localization accuracy using only 4.5 pacing sites compared to 38% accuracy with 13.7 sites for existing methods, potentially reducing procedure time and patient risk.

cAPM represents a significant advancement in clinical AI by addressing a critical limitation in cardiac care: the time-intensive nature of pace-mapping procedures used to identify ablation targets for life-threatening arrhythmias. The innovation lies not in a single breakthrough but in combining three established techniques—surrogate neural networks, active learning, and continual learning—to create a system capable of learning from cumulative patient data rather than requiring retraining for each case. This architectural choice directly impacts clinical utility by enabling knowledge transfer across multiple targets within individual patients and across the patient population.

The healthcare sector has long struggled with procedural efficiency in electrophysiology labs, where clinician expertise and manual pacing decisions drive outcomes. Traditional AI approaches require substantial retraining for new cases, creating implementation friction. cAPM's continual learning framework sidesteps this bottleneck by building institutional knowledge progressively, meaning earlier patients in a deployment sequence essentially contribute to optimizing care for later patients. The in-silico validation results are compelling—a 3x improvement in success rate using roughly one-third the pacing sites represents clinically meaningful progress that could reduce procedure duration, fluoroscopy exposure, and procedural complications.

For healthcare technology investors and medtech companies, this research validates the market opportunity in AI-guided interventional procedures. The pathway to clinical adoption requires preclinical validation and regulatory clearance, but the architecture demonstrates scalability across different physiological conditions and ventricular geometries. Hospitals facing increasing catheter ablation volumes and rising healthcare costs have economic incentives to adopt such technologies. Success in clinical trials could establish a template for similar AI applications in other guided interventional procedures, expanding addressable markets beyond cardiology into oncology, neurosurgery, and orthopedics.

• →cAPM achieves 81% clinical-accuracy localization with 4.5 pacing sites versus 38% with 13.7 sites using existing methods
• →Continual learning enables knowledge transfer across multiple VT targets and patients, eliminating need to retrain for each case
• →The system uses a task-agnostic surrogate neural network to map pacing sites to ECG morphology patterns for active site selection
• →In-silico validation demonstrates robustness across different physiological conditions and ventricular geometries
• →Clinical implementation could reduce procedure duration, radiation exposure, and complications in cardiac catheter ablation