A Photoplethysmography-Based Machine Learning Algorithm for Early Atrial Fibrillation Detection: A Prospective Validation Study

Atrial fibrillation (AF) and heart failure (HF) frequently coexist and share a bidirectional causal relationship; their concurrence is associated with worse clinical outcomes. Early detection of AF may enable timely intervention and improve outcomes. This study is prospectively validating a machine-learning algorithm for AF detection from PPG signals, intended for integration into the Seerlinq remote monitoring platform. This algorithm builds on the same core PPG signal-processing technology as Seerlinq's HeartCore device (a CE-certified, Class IIb device under the EU MDR that monitors left ventricular filling pressures in heart failure patients). It is a stand-alone algorithm designed specifically to detect clinically relevant (≥ 30s) atrial fibrillation.

Validation of the algorithm will proceed in three stages: (1) internal cross-validation; (2) external validation against an independent cohort with paired PPG-ECG recordings, to confirm generalizability; and (3) validation in a cohort of patients with paroxysmal atrial fibrillation and frequent sinus-AF transitions, to assess performance during clinically challenging rhythm changes.

The study is enrolling toward an estimated 1,000 unique PPG recordings. A 12-lead ECG is used to confirm cardiac rhythm classification (gold standard) as the reference for evaluating algorithm performance.

Inclusion criteria

• Adults ≥18 years with a diagnosis of heart failure (HFrEF, HFmrEF, or HFpEF)
• 12-lead ECG performed to confirm cardiac rhythm classification (AF vs. non-AF)

Exclusion criteria

• Missing a valid PPG recording