MMedAgent-Plus coordinates 18 specialist models across medical tasks
Original title: MMedAgent-Plus: Scaling Multimodal Medical Agents across Modalities, Tasks and Workflows
End-to-end final-answer accuracy is 0.98, matching the dedicated specialist and above a 0.28 baseline.
MMedAgent-Plus pairs a vision-language planner with 18 specialist models covering 12 tasks and 8 modalities. The planner selects tools, runs them in sequence, and synthesizes answers from their outputs. It is instruction-tuned on 208,928 samples from 32 public medical benchmarks. Held-out tests cover tool orchestration, response synthesis, and end-to-end execution without pre-supplied tool outputs.
- The planner is instruction-tuned on 208,928 samples from 32 public benchmarks.
- Tool-call accuracies are 91.0%, 98.1%, and 80.7% across three settings.
- Given correct outputs, synthesis scores 14.4 to 14.8 out of 15.
- It invokes the required specialist in 90.6% of 350 evaluation cases.
From the original article
1 min read at huggingface.co
Medical workflows often require multiple interdependent analyses, but existing specialist and generalist AI models are not designed to coordinate them. We introduce MMedAgent-Plus, a multimodal medical tool-use agent that combines a vision-language planner with 18 leading specialist models spanning 12 tasks and 8 modalities.
By Ruoxin Wang, Yuanting Pan, Binxu Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.