Research map / Vision-language and multimodal / Multimodal LLMs
Towards Reliable Medical Large Vision-Language Models via Counterfactual Preference Optimization
Xiaoguang Zhu, NaipengWang NaipengWang, Kartik Patwari, Lianlong Sun, Chen-Nee Chuah, ChengxinPang ChengxinPang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add standard DPO loss L_DPO alongside counterfactual loss (ablation) | CoMedPO (SFT) · VQA-RAD Open | recall: 29.50 -> 42.50 | Improved |
| use counterfactual loss instead of causal-weighted DPO loss (ablation) | CoMedPO (SFT) · VQA-RAD Open | recall: 37.50 -> 42.50 | Improved |
| add counterfactual clinical query for composite pairs (ablation) | CoMedPO (SFT) · VQA-RAD Open | recall: 42.00 -> 42.50 | No clear effect |
| use matched prompt with composite image at inference instead of composite only (ablation) | CoMedPO (SFT) · VQA-RAD Open | recall: 37.50 -> 43.43 | Improved |
| use causally constructed background instead of random patch in composite (ablation) | CoMedPO (SFT) · VQA-RAD Open | recall: 37.00 -> 42.50 | Improved |
| add CoMedPO counterfactual optimization to MMedPO (no SFT) | MMedPO · SLAKE Closed | accuracy: 58.25 -> 71.17 | Improved |
| add CoMedPO to MMedPO (no SFT) | MMedPO · IU-XRay | ROUGE-L: 29.52 -> 32.68 | Improved |
| add CoMedPO to SFT-enhanced MMedPO | MMedPO + SFT · SLAKE Closed | accuracy: 67.82 -> 74.40 | Improved |
| add CoMedPO to SFT-enhanced MMedPO | MMedPO + SFT · VQA-RAD Closed | accuracy: 67.73 -> 74.90 | Improved |
| add counterfactual optimization to mDPO | mDPO · SLAKE Closed | accuracy: 60.29 -> 65.07 | Improved |
| add counterfactual optimization to mDPO | mDPO · IU-XRay | ROUGE-L: 22.38 -> 30.65 | Improved |
| add counterfactual optimization to SPPO | SPPO · VQA-RAD Open | recall: 38.50 -> 42.50 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MMedPO — MMedPO: Aligning Medical Vision-Language Models with Clinical-Aware Multimodal Preference Optimization (ICML 2025) · better on 8 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MMedPO: Aligning Medical Vision-Language Models with Clinical-Aware Multimodal Preference Optimization (ICML 2025 · similarity 0.92)
- CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Models (NeurIPS 2024 · similarity 0.88)
- LLaDA-MedV: Exploring Large Language Diffusion Models for Biomedical Image Understanding (CVPR 2026 · similarity 0.87)
- Enhancing Medical Large Vision-Language Models via Alignment Distillation (AAAI 2026 · similarity 0.87)
- How Do Medical MLLMs Fail? A Study on Visual Grounding in Medical Images (ICLR 2026 · similarity 0.87)
- MedSIGHT: Towards Grounded Visual Comprehension in Medical Large Vision-Language Models (ICML 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
