From PhysioNet to Foundation Models -- A history and potential futures
From mailing magnetic tapes to training foundation models, a 56-page paper traces 35 years of medical data sharing.
Gari D. Clifford's arXiv paper 'From PhysioNet to Foundation Models' examines 25+ years of medical data evolution through the PhysioNet Resource. The 56-page analysis tracks cardiology's journey from physical media to massive AI models, identifying challenges like AI's carbon footprint, Tiny-ML potential, and scientific repeatability. It proposes leveraging PhysioNet Challenges and open-access principles to address issues in disseminating massive physiological databases and foundation models.
Why It Matters
Provides crucial historical context for AI in medicine and outlines practical solutions for sustainable, reproducible medical AI development.