Data distribution impacts the performance and generalisability of contrastive learning-based foundation models of electrocardiograms

Short summary

Pretraining foundation models on diverse ECG cohorts improves in-distribution accuracy but reduces out-of-distribution generalisation by encoding cohort-specific artifacts; an 'In-Distribution Batch' (IDB) strategy boosts out-of-distribution robustness by 9-40%.

AI-generated from the title and abstract; the full text is not read.