Catheter Ablation -> Atrial Fibrillation & Atrial Flutter: -> Experimental methods D-PO01 - Featured Poster Session (ID 11) Poster

D-PO01-179 - Artificial Intellegence Controlled Local Impedance Drop Guided Catheter Pulmonary Vein Isolation (ID 953)


Background: Insufficient lesion depth and discontinuity of ablation lines are reasons of unsuccessful PVI. Artificial intelligence (AI) controlling changes of local tissue impedance (LI) with other ablation parameters during ablation may allow to define the optimal target of radiofrequency application.
Objective: The aim of our work is to develop a new effective technique for detecting unsuccessful ablation points during catheter LI-guided pulmonary veins isolation.
Methods: We developed an AI - able to analyse ablation points parameters during PVI by two-layer fully connected network for solving the binary classification task using cross-validation training approach. A total of 35 patients with paroxysmal and persistent atrial fibrillation underwent a LI-guided denovo PVI. The quality of catheter-tissue contact was assessed using LI before each ablation, which were stopped as soon as LI-drop plateau was reached. After PVI was achieved, remapping and adenosine infusion were performed to reveal reconnections. Ablation parameters including bipolar voltages, local impedance, system impedance, duration, temperature were registered before and after each energy application and were analysed by AI.
Results: AI analysed 38280 parameters of successful lesions and compared with 2420 of unsuccessful lesions. AI identified unsuccessful lesions with area under curve of 0.93 (95% CI 0.92-0.94), sensitivity of 96.0%(94.6-97.5) and specificity of 72.2%(71.3-73.5). On receiver operating characteristic curve analysis, optimal single LI-drop threshold was 9.6 Ω (sensitivity, 94 %; specificity, 60 %). The median LI-drop for successful lesions was 12 Ω (7.2-18.2 Ω, p<0.005), which was larger than for unsuccessful lesions (5.0 Ω [2.7-8.0 Ω], p<0.005).
Conclusion: The AI-analysing ablation parameters during local impedance drop guided PVI predicted unsuccessful lesions with higher sensitivity and specificity than single LI-drop.