Member-only story
Facilitating Radiology and Healthcare with Medical AI. How far are we with Clinical Large Language Models as Med-Gemini?
The idea of solving healthcare issue with artificial intelligence is actually really old, drawing back to the early ages of AI in the 50'. Just few years ago IBM introduced Watson for several health applications, such as Watson for Oncology and Watson for Genomics. However, it didn’t really hold its promises and was never really adopted in daily use by the medical community.
With the raise of large language models (LLMs) it seems we are all once again in this race to help(?) physicians, especially for what concerns automating report and relating medical imaging to potential diagnosis. In this context, a clinical evaluation is not just a text. It necessitates a model that can comprehend complex multimodal data, such as photos, movies, and obviously the considerable length and breadth of electronic health records (EHRs).
Google research lab already surprised us in 2023 with with Med-Palm, a large language model able to reach 86% of accuracy in the MedQA medical exam benchmark. Now, they top this with a model showing a significant advancement in multimodal and long-context reasoning, with importance to the aforementioned medical promise based on Gemini. Indeed, Med-Gemini is based upon the research coming up from…
