The questions we're working on.
MedAI research asks a practical question about clinical and educational AI: does it actually hold up? Start with the studies below — each one is a short, plain-language entry point into the full paper.
Three studies to start with.
Each card opens the published paper. The complete 2026 list is in the archive below.
Can an AI really read an ECG — or is it just recognizing the patient?
When beats from the same patient land in both training and testing, a model can look near-perfect while learning nothing clinical; this study shows how to tell the difference before anyone relies on it.
Read the studyTwo models score the same. Only one is steady enough to trust.
Headline accuracy hides how a model got there, so this work reads the training itself to find which eye-cancer survival predictions are stable enough to be clinically useful.
Read the studyWhat happens when a fracture-reading AI meets a patient it never trained on?
A model tuned on one population can quietly fail on another, which is why this study argues that out-of-distribution testing has to happen before a hospital puts an imaging model to work.
Read the studyView Full 2026 Publication List
2026
Connect with MedAI
Have a question, collaboration idea, project, or interest in getting involved with the Center?