CI-452770-3 ARTIFICIAL INTELLIGENCE BASED PREMATURE VENTRICULAR CONTRACTION DETECTION USING INSERTABLE CARDIAC MONITOR

Premature ventricular complex (PVC) frequency and duration have important clinical implications. PVC induced cardiomyopathy is a potentially reversible condition in which cardiac function impaired by the occurrence of frequent PVCs. In addition, PVCs may result from acquired cardio toxicity and/or predictive for life-threatening arrhythmias. detection using computationally simple techniques has been recently introduced insertable monitors (ICMs). An ICM device capable monitoring burden over longer period can be useful tool evaluating temporal dynamics leading to appropriate timely interventions patients. The study evaluated an artificial intelligence (AI) based model improve accuracy ICM. We developed deep learning convolution neural network (CNN) was trained validated on ECG data. strips stored implanted real world patients collected deidentified CareLink™ data warehouse. All were manually annotated Features that determined best suited identifying beats including QRS morphology differences, RR interval amplitude slope differences between adjacent extracted provided as input AI model. tested validation dataset performance metrics accuracy, sensitivity specificity computed. CNN ResNet-18 90 patient activated episodes total 14,354 with 2,200 beats. This different morphologies monomorphic, polymorphic, bigeminy trigeminy consisted 4,300 664 true various morphologies. On dataset, obtained 98.9%, 97.3% 99.2% respectively. 3,614 normal 646 correctly identified whereas 28 incorrectly 18 Detection utilizing novel algorithm both feasible exceeding non-AI algorithms.