Research map / Vision-language and multimodal / Multimodal LLMs
Clinical Cognition Alignment for Gastrointestinal Diagnosis with Multimodal LLMs
Huan Zheng, Yucheng Zhou, Tianyi Yan, Dubing Chen, Hongbo Lu, Wenlong Liao, Tao He, Pai Peng, Shen Jianbing
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 hierarchical clinical cognition SFT (ablation) | Qwen3-VL-8B · Average (5 GI benchmarks) | accuracy: 35.30 -> 66.31 | Improved |
| add hierarchical clinical cognition SFT (ablation) | Qwen3-VL-2B · Average (5 GI benchmarks) | accuracy: 16.05 -> 54.49 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-8B (SFT) · Average (5 GI benchmarks) | accuracy: 66.31 -> 67.67 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-4B (SFT) · Average (5 GI benchmarks) | accuracy: 61.98 -> 64.05 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-2B (SFT) · Average (5 GI benchmarks) | accuracy: 54.49 -> 57.40 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-8B (SFT) · Multi-label subset (strict exact match) | accuracy: 10.64 -> 13.62 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-8B (SFT) · CrohnIPI | accuracy: 62.14 -> 63.11 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-8B (SFT) · GastroVision | accuracy: 76.92 -> 77.21 | Improved |
| add counterfactual-driven GRPO after SFT (ablation) | Qwen3-VL-4B (SFT) · HyperKvasir | accuracy: 86.10 -> 89.19 | Improved |
| use Gaussian blur instead of solid white mask for counterfactuals (ablation) | CogAlign · GI benchmarks | accuracy | Improved |
| add clinical cognition reward R_cog (ablation) | CogAlign GRPO · GI benchmarks | accuracy | Improved |
| add diagnostic consistency reward R_diag (ablation) | CogAlign GRPO · GI benchmarks | accuracy | Improved |
Closest papers by meaning
- EndoBench: A Comprehensive Evaluation of Multi-Modal Large Language Models for Endoscopy Analysis (NeurIPS 2025 · similarity 0.88)
- Gastric-X: A Multimodal Multi-Phase Benchmark Dataset for Advancing Vision-Language Models in Gastric Cancer Analysis (CVPR 2026 · similarity 0.87)
- OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning (CVPR 2026 · similarity 0.85)
- Beyond Classification Accuracy: Neural-MedBench and the Need for Deeper Reasoning Benchmarks (ICLR 2026 · similarity 0.84)
- M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding (ICLR 2026 · similarity 0.84)
- MedAtlas: Evaluating LLMs for Multi-Round, Multi-Task Medical Reasoning Across Diverse Imaging Modalities and Clinical Text (AAAI 2026 · similarity 0.84)
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.
