PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations
The paper introduces PACE-ECG, an ECG pretraining framework that combines contrastive learning across temporally neighboring recordings from the same patient with clinically guided multitask supervision. Frozen representations were evaluated for post-myocardial-infarction mortality and incident heart failure prediction across MIMIC-IV and two external HEEDB cohorts. The model reportedly outperformed DeepECG-SSL in-domain and on external populations while using an encoder 41 times smaller and less pretraining data. A matched ablation indicates that patient-temporal pairing improved external AUROC over generic self-supervision.
It offers a concrete pretraining-design result for ECG foundation models with external-cohort evidence, which is useful for readers tracking transferable clinical AI representations.
Comments & Questions (0)
No comments yet
Be the first to comment!