Towards Generalizable AI in Medicine via Generalist-Specialist Collaboration

Abstract

Generalist foundation models are powerful and flexible across diverse tasks, while specialist models can provide precise expertise for specific medical settings. This work introduces Generalist-Specialist Collaboration (GSCo), a cooperative framework that combines a medical generalist model with lightweight specialists. In GSCo, the specialist models provide task-specific guidance, such as diagnostic predictions and visually similar clinical cases, as contextual information for the generalist model, which then makes the final prediction or generates the final report.

The authors also develop MedDr, an open-source generalist foundation model tailored for medicine, together with lightweight specialist modules for downstream tasks. Comprehensive experiments across 32 datasets and multiple medical modalities show that MedDr and GSCo improve performance over strong generalist and specialist baselines in medical image diagnosis and medical report generation. The results suggest that combining generalist and specialist models can be an effective and computationally efficient path toward scalable and generalizable clinical AI.

Read paper here

Code available here

Recommended citation: Sunan He, Yuxiang Nie, Hongmei Wang, Shu Yang, Yihui Wang, Zhiyuan Cai, Zhixuan Chen, Yingxue Xu, Linshan Wu, Ngai Shing Cheng, Luyang Luo, Huiling Xiang, Xi Lin, Mingxiang Wu, Yifan Peng, George Shih, Ziyang Xu, Xian Wu, Qiong Wang, Ronald Cheong Kin Chan, Varut Vardhanabhuti, Xiaohui Duan, Winnie Chiu Wing Chu, Yefeng Zheng, Pranav Rajpurkar, Kang Zhang, and Hao Chen. 2026. Towards generalizable AI in medicine via Generalist-Specialist Collaboration. Nature Biomedical Engineering. https://doi.org/10.1038/s41551-026-01653-3.