Exercise electrocardiography (ECG) provides clinically important information on rhythm disturbances occurring under physiological stress, but is challenged by motion artefacts, muscle noise, and baseline drift. Premature atrial contractions (PACs) are commonly observed during exercise and may unmask arrhythmic burden, but due to their similarity to normal sinus beats, reliable automated detection remains challenging, particularly from single-lead recordings.
We developed a CNN–BiLSTM U-Net with self- attention to detect PACs in single-lead exercise ECG signals from a UK Biobank exercise cohort. Performance was assessed on a held-out test set using beat-level metrics, and agreement in participant-level PAC counts.
On the held-out test set, the model achieved a beat-level PAC F1 score of 0.74, with precision of 0.80 and recall of 0.70 within a multi-class classification framework (overall macro F1 0.84). Bland–Altman analysis demonstrated good agreement between predicted and ground truth PAC counts per participant, with a mean bias of −0.14 beats and limits of agreement (–3.33, 3.05).
This work demonstrates that automated PAC detection in noisy single-lead exercise ECGs is feasible, enabling scalable, population-level phenotyping of ectopy under cardiac stress.