Case Study: NVIDIA achieves scalable medical imaging AI—processing 1M images/day and 94% diagnostic accuracy—with Flywheel

Advancing Imaging AI Development With NVIDIA and Flywheel

NVIDIA faced the challenge of advancing imaging AI development across fragmented, sensitive medical datasets that are often siloed, unstandardized, and time-consuming to curate — with data scientists spending up to 80% of their time on manual preparation. Flywheel stepped in as the vendor, offering the Flywheel platform/Flywheel Enterprise with MONAI Enterprise integration, plus Flywheel Exchange and the Gear Exchange, to provide a centralized, secure solution for data discovery, aggregation, curation, annotation, training, and collaboration.

Flywheel implemented automated workflows, AI-assisted annotation (MONAI Label), federated learning (NVIDIA FLARE), and a library of ready-to-use gears, deployed on NVIDIA DGX infrastructure to scale processing and production. The Flywheel solution enabled measurable impact: UW–Madison teams achieved 94% diagnostic accuracy for a COVID-19 chest X‑ray model (versus 85% for thoracic radiologists), processed more than 1 million images and 10,000 cases per day, condensed eight months of work into a single day, and realized a reported 9% boost in accurate diagnoses.

NVIDIA