Token-Sparse Medical Multimodal Reasoning via Dual-Stream Reinforcement Learning
Lists
Only you see this
Abstract
Vision-language models (VLMs) combining reinforcement learning (RL) ignite remarkable progress in multimodal reasoning, yet still struggle with medical images, which typically exhibit extremely sparse visual evidence to inform clinical decision-making. We recognize that pruning visual tokens outside the grounding region greatly enhances medical reasoning. However, a united RL framework for active visual token pruning (VTP) and medical multimodal reasoning remains unestablished. Here, we propose a dual-stream RL framework, **ViToS**, to fulfill token pruning and question answering. ViToS trains one policy model with two task branches, where one focuses on grounding while the other conducts token-sparse reasoning after VTP. Furthermore, we solve the coupled policy learning problem by introducing the cross-feedback sequential optimization, avoiding gradient conflict and facilitating convergence of the shared policy model. Evaluated on seven medical benchmarks, our method reduces visual tokens to 77% of the original sequence length while achieving a 108.27% relative performance on Lingshu-7B and 104.16% relative performance on HuatuoGPT-Vision-7B. Overall, ViToS delivers superior performance and inference speedup, establishing an efficient paradigm for medical multimodal reasoning.
BibTeX
@inproceedings{chen2026token,
author = {Kaitao Chen and Weiqian Zhao and Jiamin Wu and Qihao Zheng and Shangquan Sun and Chunfeng Song and Xiaosong Wang and Mu Zhou and Mianxin Liu},
title = {{Token-Sparse Medical Multimodal Reasoning via Dual-Stream Reinforcement Learning}},
booktitle = {International Conference on Machine Learning},
year = {2026},
url = {https://openreview.net/forum?id=elrnTLDiT3},
}Loading related content...