Departments of Basic Medical Sciences Seminar

Presenter

Seminar Title

• AI Alone Won't Cut It: Toward Multimodal Collaborations in Medicine

Details

Medical AI mostly runs on single data streams: while no radiologist reads an image without a history, our models do it every day. Across several medical prediction tasks, I will illustrate how unimodal models can produce confident predictions on cases where the deciding evidence sits in data the model never saw. The obvious fix, fusing modalities, often loses to strong unimodal baselines, undone by the missingness, misalignment, and uneven quality that define real clinical data.

Navigating those pitfalls takes expertise no single team holds, because every modality may involve a different specialty, and a model built inside one discipline inherits its blind spots. I will close with what our results suggest actually works, technically and organizationally, and with the claim at the heart of this talk: the future of medical AI depends less on architecture than on teams as multimodal as the data.

BMS Faculty Hosts

• Nipavan Chiamvimonvat, MD

For additional information, please contact Lisa Dunk, 602-827-2188. This seminar will be hosted via Zoom.

CME credit provided by the University of Arizona College of Medicine – Tucson.