Key result
Machine learning classifies AVNRT ablation lesion outcomes with ~82% accuracy using geometric and biophysical inputs.
Why the study?
The optimal site and biophysical parameters for radiofrequency delivery in AVNRT catheter ablation remain poorly standardized.
What are the geometric and biophysical characteristics of successful versus non-successful ablation sites in AVNRT, and can they predict lesion outcome?
Cohort (n=94)
What are the geometric and biophysical characteristics of successful versus non-successful ablation sites in AVNRT, and can they predict lesion outcome?
Cloud-based spatial normalization and biophysical modeling can identify features of effective AVNRT ablation lesions and predict successful outcomes with over 80% accuracy.
Answers are grounded in this paper and cite their sources.
May assist lesion outcome classification in EP; hypothesis-generating pending prospective validation.
Background Catheter ablation of atrioventricular nodal reentrant tachycardia (AVNRT) is a routine and highly effective therapy. Yet the optimal site and biophysical parameters for radiofrequency (RF) delivery remain poorly standardized. The emergence of cloud-based mapping software enables multicase geometric and biophysical modeling of ablation data to identify reproducible targets. Objective To characterize the geometric distribution and lesion parameters of successful versus non-successful ablation sites in AVNRT, and to develop a predictive model using aggregated CARTONET data. Methods We prospectively analyzed 94 consecutive AVNRT ablation procedures recorded in CARTONET over a 10-year period. Mapping points tagged as "His" defined a patient-specific reference centroid. Ablation points labeled slow junctional rhythm (SJR), fast junctional rhythm (FJR), or AV block were compared for spatial position relative to the His cloud and for biophysical metrics (impedance, power, temperature, duration, stability). Machine-learning modeling (XGBoost classifier) was applied to predict ablation outcomes from geometric and biophysical inputs. Results A total of 310 ablation points were analyzed. Mean distance from the His centroid was 18.3±5.9 mm for SJR sites, 17.9±6.1 mm for FJR, and 15.9±4.8 mm for AV block. Successful lesions (SJR) exhibited higher mean power (40 W), greater impedance drop (~11Ω), and longer duration (28 s) compared with non-successful sites. The machine-learning model achieved an accuracy of 82.3% (95% CI 72.6–90.3) in classifying lesion outcome. Conclusion CARTONET-based spatial normalization allows reproducible modeling of the ideal slow-pathway ablation region and identifies biophysical features associated with effective lesion formation. This approach supports the development of data-driven tools to guide ablation.
No takes yet. Share an insight, caveat, or question.
Mignot et al. (2026) conducted a cohort in Atrioventricular nodal reentrant tachycardia (AVNRT) (n=94). Machine-learning modeling (XGBoost classifier) was evaluated on Classification of lesion outcome (Accuracy 82.3%, 95% CI 72.6-90.3). A machine-learning model using geometric and biophysical inputs achieved an accuracy of 82.3% in classifying AVNRT ablation lesion outcomes.
Synapse has enriched 10 related papers. Consider them for comparative context: