MedVLThinker: Simple Baselines for Multimodal Medical Reasoning

2025-08-04

AI Summary

Current large medical foundation models lack an open, reproducible paradigm for developing robust reasoning capabilities. Method: We propose RLVR—a novel reinforcement learning framework grounded in answer verifiability—specifically designed for multimodal medical reasoning. RLVR integrates difficulty-aware reasoning data curation, supervised fine-tuning, and chain-of-thought generation, systematically enhancing reasoning on the Qwen2.5-VL series. Contribution/Results: Our analysis reveals that pure-text reasoning data substantially outperforms multimodal data—a finding that challenges prevailing assumptions. The 7B variant achieves new open-source state-of-the-art performance across six medical QA benchmarks; the 32B variant matches GPT-4o’s accuracy. To foster reproducibility and community advancement, we fully open-source all training data, model weights, and implementation code.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Large Reasoning Models (LRMs) have introduced a new paradigm in AI by enabling models to ``think before responding" via chain-of-thought reasoning. However, the absence of open and reproducible recipes for building reasoning-centric medical LMMs hinders community-wide research, analysis, and comparison. In this paper, we present MedVLThinker, a suite of simple yet strong baselines. Our fully open recipe consists of: (1) systematic data curation for both text-only and image-text medical data, filtered according to varying levels of reasoning difficulty, and (2) two training paradigms: Supervised Fine-Tuning (SFT) on distilled reasoning traces and Reinforcement Learning with Verifiable Rewards (RLVR) based on final answer correctness. Across extensive experiments on the Qwen2.5-VL model family (3B, 7B) and six medical QA benchmarks, we find that RLVR consistently and significantly outperforms SFT. Additionally, under the RLVR framework, a key, counter-intuitive finding is that training on our curated text-only reasoning data provides a more substantial performance boost than training on multimodal image-text data. Our best open 7B model, trained using the RLVR recipe on text-only data, establishes a new state-of-the-art on existing public VQA benchmarks, surpassing all previous open-source medical LMMs. Furthermore, scaling our model to 32B achieves performance on par with the proprietary GPT-4o. We release all curated data, models, and code to provide the community with a strong, open foundation for future research in multimodal medical reasoning.

Problem

• Lack of open recipes for medical reasoning models
• Need for systematic data curation in medical AI
• Performance gap between text-only and multimodal training

Innovation

• Systematic data curation for medical reasoning
• Reinforcement Learning with Verifiable Rewards
• Training on text-only data boosts performance