Please join the next seminar in the series hosted by the Cancer Research UK Cambridge Centre and Cambridge MedAI:
Multimodal AI for Radiology
Stephanie Hyland, Principal Researcher, Microsoft Research Cambridge
When: Thursday 18 April, 15:00—17:00
Where: Rayleigh seminar room, Maxwell Centre, JJ Thomson Avenue, Cambridge, CB3 0HE
The seminar will be followed by drinks and nibbles.
All are welcome
Abstract: Advances in large language models have enabled the development of large multimodal models (LMMs), combining vision understanding with the language generation capabilities of LLMs. In this talk I will describe our ongoing efforts to develop multimodal models specialised for radiology applications, in particular for radiology report generation, starting with chest X-ray reporting. Starting from the general domain models DINOv2 and Vicuna-7B, I will describe how we developed and evaluated "MAIRA-1", our radiology-adapted LMM, and outline limitations and open research questions.
Stephanie Hyland is Principal Researcher in Microsoft Research Health Futures. She works on multimodal machine learning and its use for healthcare, with a special focus on radiology as part of Project MAIRA. Her research to date also includes the modelling of medical time series of the kind found in electronic health records, especially physiological time series from ICU patients. Above all she is interested in making machine learning practically useful; as such she enjoys working closely with clinical collaborators to identify problems where machine learning can support human experts, as well as thinking carefully on how to evaluate and validate models.
Prior to joining Microsoft, I completed my PhD at Cornell University/Weill Cornell Medicine/Sloan Kettering Cancer Center in Computational Biology and Medicine, and studied Theoretical Physics and Applied Mathematics at Trinity College Dublin (BA Mod.) and Cambridge University (Part III; MASt).
