Foundation Models for Hemodynamic Time Series: A New Paradigm in Cardiovascular Data Modeling

Hemodynamic waveforms encode rich physiological signals essential for cardiovascular assessment, but scalable interpretation has been constrained by the need for labeled data and expensive imaging. Leveraging $\sim$34,000 hours of finger-cuff and arterial blood pressure waveforms from $\sim$12,000 subjects—collected with Edwards Lifesciences ClearSight and FloTrac devices—we develop a transformer-based foundation model that learns robust representations of cardiovascular dynamics. Trained with self-supervised learning, the model delivers sample-efficient performance, matching state-of-the-art benchmarks using only 30\% of labeled data, in detecting aortic stenosis and reduced left ventricular ejection fraction. To our knowledge, this is the first foundation model trained solely on blood pressure waveforms for screening cardiovascular diseases.