A sample of past research projects that are public.
MAIRA (Multimodal AI Radiology Assistant) is a system for generating radiology reports and answering clinical questions about chest X-rays. It combines large language models with radiology-specific vision encoders, and has been evaluated in rigorous clinical settings at the Mayo Clinic.
Rad-DINO is a self-supervised vision foundation model pre-trained on large collections of chest X-rays. It learns rich visual representations without manual labels and achieves strong performance on a range of downstream radiological tasks.
AI-model to prioritize patients for Barrett’s esophagus screening based on routine H&E whole-slide images. Ongoing trial with CancerUK. The work shows how pathology AI can reduce dependence on costly staining and pathology review time while maintaining strong diagnostic performance in clinical workflows.
Projects leveraging data from NASA's Solar Dynamics Observatory (SDO) to study how AI can enhance remote sensing instruments in space missions, including reducing the need for costly telemetry and enabling virtual instruments.