Key result
Wearable PPG deep-learning model estimates AF burden within ~7% error, outperforming existing algorithms.
Why the study?
Long-term AF burden measurement using wearable PPG is challenged by motion-induced signal corruptions that produce sparse, irregularly sampled observations, and translating these fragmented measurements into a reliable burden estimate remains an open problem.
Does AF-BurdenNet improve atrial fibrillation burden estimation accuracy from intermittently sampled wearable photoplethysmography compared to existing irregular-sequence models in a clinical cohort?
Cohort (n=140)
Does AF-BurdenNet improve atrial fibrillation burden estimation accuracy from intermittently sampled wearable photoplethysmography compared to existing irregular-sequence models in a clinical cohort?
Answers are grounded in this paper and cite their sources.
AF-BurdenNet enables accurate longitudinal atrial fibrillation burden quantification from fragmented wearable PPG signals by incorporating uncertainty-weighted temporal fusion to handle natural signal missingness.
Background: Long-term atrial fibrillation (AF) burden measurement using wearable photoplethysmography (PPG) is challenged by motioninduced signal corruptions that produce sparse, irregularly sampled observations. Translating these fragmented measurements into a reliable burden estimate remains an open problem. Method: We propose AF-BurdenNet, a two-stage framework for intermittently sampled PPG. Stage one evaluates each valid segment using a multi-view discrimination model and quantifies predictive uncertainty via calibrated entropy. Stage two aggregates these discrete measurements through a multi-scale temporal regression network to yield a patient-level aggregate burden estimate, where entropy serves as a dynamic gating signal to down-weight unreliable observations and compensate for temporal discontinuities. Results: On a 140-subject clinical cohort with natural missingness, AF-BurdenNet achieves a mean absolute error of 7.16% and Pearson correlation of 0.91 for AF patients, outperforming existing irregular-sequence models. Controlled experiments confirm the framework does not exploit spurious AF-missingness correlations, and uncertainty-aware gating suppresses overestimation in zero-burden controls (mean estimated burden 0.87%). Significance: By explicitly modeling the missingness patterns inherent to ambulatory PPG and introducing uncertainty-weighted temporal fusion, this framework advances wearable PPG from episodic detection toward robust longitudinal burden quantification under realistic sensing constraints.
No takes yet. Share an insight, caveat, or question.
Xuenan Liu (2026) conducted a cohort in Atrial fibrillation (n=140). AF-BurdenNet (two-stage uncertainty-aware framework for intermittently sampled PPG) vs. Existing irregular-sequence models was evaluated on Atrial fibrillation burden estimation accuracy (mean absolute error and Pearson correlation) (MAE 7.16%, r 0.91). AF-BurdenNet estimated atrial fibrillation burden from intermittent wearable PPG with a mean absolute error of 7.16% and Pearson correlation of 0.91, outperforming existing irregular-sequence models.
Synapse has enriched 10 related papers. Consider them for comparative context: