Med-RewardBench: Benchmarking Reward Models and Judges for Medical Multimodal Large Language Models

2025-08-29

AI Summary

Medical multimodal large language models (MLLMs) hold significant promise for clinical decision support, yet suffer from a critical gap: the absence of diagnosis-accurate and clinically aligned reward models and evaluation benchmarks. To address this, we introduce Med-RewardBench—the first dedicated multimodal reward modeling benchmark for healthcare—spanning 13 organ systems and 8 clinical specialties, with 1,026 rigorously curated cases annotated by domain experts across three stages and evaluated along six clinically grounded dimensions. Our work establishes the first fine-grained, clinical-domain-specific reward modeling benchmark, enabling systematic assessment of MLLM alignment with expert judgment. Comprehensive evaluation of 32 state-of-the-art MLLMs reveals pervasive misalignment with clinical reasoning. Leveraging our expert-annotated data, we further fine-tune multimodal reward models, achieving substantial improvements in clinical alignment. This benchmark and methodology advance rigorous, clinically meaningful evaluation and optimization of medical MLLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Multimodal large language models (MLLMs) hold significant potential in medical applications, including disease diagnosis and clinical decision-making. However, these tasks require highly accurate, context-sensitive, and professionally aligned responses, making reliable reward models and judges critical. Despite their importance, medical reward models (MRMs) and judges remain underexplored, with no dedicated benchmarks addressing clinical requirements. Existing benchmarks focus on general MLLM capabilities or evaluate models as solvers, neglecting essential evaluation dimensions like diagnostic accuracy and clinical relevance. To address this, we introduce Med-RewardBench, the first benchmark specifically designed to evaluate MRMs and judges in medical scenarios. Med-RewardBench features a multimodal dataset spanning 13 organ systems and 8 clinical departments, with 1,026 expert-annotated cases. A rigorous three-step process ensures high-quality evaluation data across six clinically critical dimensions. We evaluate 32 state-of-the-art MLLMs, including open-source, proprietary, and medical-specific models, revealing substantial challenges in aligning outputs with expert judgment. Additionally, we develop baseline models that demonstrate substantial performance improvements through fine-tuning.

Problem

• Benchmarking medical reward models and judges for MLLMs
• Addressing lack of clinical evaluation dimensions in existing benchmarks
• Ensuring diagnostic accuracy and clinical relevance in medical responses

Innovation

• Med-RewardBench benchmark for medical reward models
• Multimodal dataset spanning 13 organ systems
• Baseline models with fine-tuning performance improvements