Digital Health -> Digital Health D-AB25 - New Technologies and New Challenges (ID 28) Abstract

D-AB25-06 - An Artificial Intelligence-based Solution To Reduce False Positive Detections Of Atrial Fibrillation By An Implantable Loop Recorder (ID 1461)


Background: Implantable loop recorders (ILRs) are routinely used in patients (pts) with suspected or known atrial fibrillation (AF). Since the positive predictive value of ILR detected AF is low, manual adjudication is required to distinguish true and false positive (TP, FP) AF events.
Objective: To determine if an artificial intelligence (AI)-based ECG analysis solution at the end-user level can reduce the incidence of FP AF detections.
Methods: We identified consecutive pts with an ILR (Reveal LINQ, Medtronic) for either diagnosis (i.e., cryptogenic stroke [CS]) or management of known AF. ECG strips of ILR detected AF were extracted from consecutive monthly monitoring reports in .pdf format. A validation dataset of 1190 ECG strips (591 AF episodes from 148 CS pts; 599 AF episodes from 200 known AF pts) was annotated twice; an expert adjudicated all discrepancies. These ECG strips were then processed by an AI-based algorithm (Cardiologs) using a deep neural network trained with a dataset of more than 1,000,000 ECGs. The performance of the AI-algorithm was assessed.
Results: Overall, 412 (70%) and 287 (48%) of all ILR detected AF episodes in the CS and known AF cohorts, respectively, were manually adjudicated to represent FP AF. The AI-based algorithm reduced the number of FP episodes by 69.7% and 62.0% in the CS and known AF cohorts, respectively, with only a 1.7% and 0.6% corresponding false negative rate (Figure).
Conclusion: A novel AI based ECG analysis solution employed at the end-user level reduced by nearly 2/3 the burden of adjudicating FP ILR detected AF episodes. Routine application of this approach offers significant promise in reducing the clinical burden to manage ILR pts.