mlsanityData to 14 Oct 2026

Med-OPD: Improving Medical Vision-Language Models via Evidence-Aware On-Policy Distillation

arXiv 2607.16303 ↗First posted 14 Jul 2026cs.CV cs.AIcited by 2

On the map: LLM reasoning → Clinical and biomedical applications

also in 2 more
Abstract (arXiv)

Medical Vision-Language Models (Med-VLMs) require reliable reasoning from fine-grained visual evidence, yet existing models can produce plausible clinical answers by relying on language priors or medical templates rather than truly attending to diagnosis-critical regions. On-Policy Distillation (OPD) offers dense token-level supervision on student-generated trajectories and provides a privacy-compatible means of capability transfer without requiring the redistribution of raw patient data. However, standard OPD uniformly distills all tokens, causing sparse evidence-dependent tokens to be diluted by abundant clinical narrative tokens. Inspired by the success of OPD in the large language model community, we propose Med-OPD, to our knowledge the first unified post-training framework that integrates on-policy distillation with medical evidence-aware supervision for Med-VLMs. We introduce Medical Evidence Advantage (MEA), a teacher-grounded counterfactual signal that uses an answer-aware hint to focus teacher scoring on evidence supporting the target diagnosis, and measures each token's dependence on medical visual evidence by comparing teacher likelihoods under the original and evidence-degraded imaging modalities. Based on MEA, Med-OPD redistributes the distillation signal at both the token and trajectory levels, emphasizing diagnosis-critical tokens and evidence-reliant rollouts. Experiments on OmniMedVQA subsets show that Med-OPD consistently outperforms SFT and standard OPD across CT, MRI, Disease Diagnosis, and Lesion Grading. These results demonstrate that evidence-aware distillation can better strengthen medical VLMs' reliance on key visual evidence and improve reliable multimodal medical reasoning. The source code and data is publicly available at: https://github.com/yunhang8658/MedOPD.git

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Clinical and biomedical applications, one of 17 directions in LLM reasoning. Builds on 31 of its 36 references that are in the map. First posted on arXiv 14 Jul 2026.

Builds on

31 of 36 references.

Closest ancestors · 5

Same direction, newest first

  1. 2026Joint Optimization of Reasoning and Dual-Memory for Self-Learning Diagnostic Agent8 Apr
  2. 2026MMedExpert-R1: Strengthening Multimodal Medical Reasoning via Domain-Specific Adaptation and Clinical Guideline Reinforcement16 Jan
  3. 2025Med-CMR: A Fine-Grained Benchmark Integrating Visual Evidence and Clinical Logic for Medical Complex Multimodal Reasoning30 Nov
  4. 2025Med-R1: Reinforcement Learning for Generalizable Medical Reasoning in Vision-Language Models18 Mar
  5. 2025MedVLM-R1: Incentivizing Medical Reasoning Capability of Vision-Language Models (VLMs) via Reinforcement Learning26 Feb

Other references · 26

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  3. 2023Visual Instruction Tuningcited by 7,869
  4. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,137
  5. 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
  6. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,729
  7. 2023LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Daycited by 1,305
  8. 2023On-Policy Distillation of Language Models: Learning from Self-Generated Mistakescited by 939
  9. 2023Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decodingcited by 688
  10. 2024A Survey on Hallucination in Large Vision-Language Modelscited by 357
  11. 2026Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipecited by 299
  12. 2024Multi-Modal Hallucination Control by Visual Information Groundingcited by 208
14 more
  1. 2024OmniMedVQA: A New Large-Scale Comprehensive Evaluation Benchmark for Medical LVLMcited by 184
  2. 2024DistiLLM: Towards Streamlined Distillation for Large Language Modelscited by 163
  3. 2025Perception-Aware Policy Optimization for Multimodal Reasoningcited by 99
  4. 2025Self-Rewarding Vision-Language Model via Reasoning Decompositioncited by 97
  5. 2024Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Samplingcited by 90
  6. 2026OpenClaw-RL: Train Any Agent Simply by Talkingcited by 90
  7. 2026Vision-OPD: Learning to See Fine Details for Multimodal LLMs via On-Policy Self-Distillationcited by 77
  8. 2024V-DPO: Mitigating Hallucination in Large Vision Language Models via Vision-Guided Direct Preference Optimizationcited by 67
  9. 2024Visual Description Grounding Reduces Hallucinations and Boosts Reasoning in LVLMscited by 33
  10. 2026Visual-Advantage On-Policy Distillation for Vision-Language Modelscited by 24
  11. 2026DOPD: Dual On-policy Distillationcited by 12
  12. 2026MedMASLab: A Unified Orchestration Framework for Benchmarking Multimodal Medical Multi-Agent Systemscited by 3
  13. 2026UAV-MARL: Multi-Agent Reinforcement Learning for Time-Critical and Dynamic Medical Supply Deliverycited by 2
  14. 2026EmambaIR: Efficient Visual State Space Model for Event-guided Image Reconstructioncited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillationcited by 7
  2. 2026Evidence-Aligned Multimodal On-Policy Self-Distillation for Fine-Grained Visual Understandingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026OpenMedReason: Scientific Reasoning Supervision for Medical Vision-Language Models
  2. 2025TemMed-Bench: Evaluating Temporal Medical Image Reasoning in Vision-Language Models
  3. 2025Benchmarking and Mitigating Sycophancy in Medical Vision Language Models
  4. 2026CARE: Towards Clinical Accountability in Multi-Modal Medical Reasoning with an Evidence-Grounded Agentic Framework
  5. 2026When Models Learn to Ask Why: Adaptive Causal Reasoning for Trustworthy Medical Vision-Language Models
  6. 2026ClinCoT: Clinical-Aware Visual Chain-of-Thought for Medical Vision Language Models
  7. 2025OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning
  8. 2026MEDVISTAGYM: A Scalable Training Environment for Thinking with Medical Images via Tool-Integrated Reinforcement Learning