PO-04-100 ARTIFICIAL INTELLIGENCE TO IMPROVE PROCEDURAL SAFETY OF ATRIOVENTRICULAR NODAL RE-ENTRANT TACHYCARDIA ABLATION

Abstract

During slow pathway modification for atrioventricular nodal reentrant tachycardia, heart block may occur if ablation cannot be stopped in time in response to high risk electrogram features (HREF). To develop an automatic algorithm to monitor for HREF and terminate ablation earlier than human reaction. Digital electrogram data from 327 ablation runs from February 2020 to June 2022 were included. They were divided into test and validation sets which contained 126 and 201 ablation runs respectively. A program was developed with insight gained from the test set. Simulation ablation videos were rendered using the validation set electrogram data. The videos were played to three independent electrophysiologists who each determined when to stop ablation. The timing of ablation termination, sensitivity, and specificity were compared between human and program. Reasons for ablation termination in the test set include short AA time, short VV time, AV block and VA block. Cutoffs for the program were decided as follows: 240ms for short AA time, 220ms for short VV time, 250ms for AV interval and 200ms for VA interval. The sensitivity and specificity for the program in the validation set were 95.2% and 91.1% respectively, which were comparable to that of human performance at 93.5% and 95.4%. If HREF were recognized by both human and program, ablations were terminated earlier by the program 90.2% of times, by a median of 574ms (interquartile range 412-807ms, p<0.001). Program-driven termination of slow pathway modification can supplement human judgement to improve ablation safety.Tabled 1AI compared with human in validation sets (PO-04-100)Operator 1Operator 2Operator 3Total human armAISensitivity0.9700.9840.8510.9350.952Specificity0.9610.9170.9850.9550.911AI lead time in milliseconds- Median (IQR)561 (405.5)542 (320.75)602 (473)574 (396)N/A- p value for median<0.001<0.001<0.001<0.001N/A- Mean +/- SD613.0 +/- 844.6609.8 +/- 1012.9824.1 +/- 1057.8673.7 +/- 969.9N/A- p value for mean<0.001<0.001<0.001<0.001N/A Open table in a new tab