Dynamic-AF: a photoplethysmography-based atrial fibrillation burden evolution model in a post-ablation cohort

Three-tier PPG framework: (1) daily snapshot, an MLP with Monte Carlo dropout giving probabilistic AF classification; (2) weekly burden by Bayesian fusion of daily probabilities; (3) trajectory prediction with dilated TCNs. Prospective validation on 400 subjects (200 persistent AF): sensitivity 0.963, specificity 0.958 for daily detection with 5.8 % ambiguous samples excluded; 40 post-ablation patients: AUC 0.903 for 12-week deterioration prediction (ΔAUC 0.114 vs a frequency-based transformer, p<0.001). Device/site and split type not stated in the abstract.

Transitioning from static binary classification to dynamic monitoring, this study addresses the critical need for continuous atrial fibrillation (AF) burden evolution mapping in a post-ablation cohort. We propose Dynamic-AF, a hierarchical photoplethysmography (PPG) framework operating through three integrated tiers: (1) Daily Snapshot employing multi-layer perceptron with Monte Carlo dropout for probabilistic AF classification; (2) Weekly Burden computed via Bayesian fusion of daily probabilities; (3) Trajectory Prediction utilizing dilated temporal convolutional networks for progression modeling. Prospective validation on 400 subjects (200 persistent AF) demonstrated 0.963 sensitivity and 0.958 specificity in daily detection with 5.8% ambiguous samples excluded; longitudinal analysis of 40 post-ablation patients showed 0.903 Area Under Receiver Operating Curve (AUC) for 12-week deterioration prediction, significantly outperforming conventional methods (∆AUC=0.114 vs. frequency-based transformer, p<0.001).

Verbatim from OpenAlex (opens in a new tab) · Abstract via OpenAlex (publisher-deposited); © the authors/publisher

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.