Research map / Vision-language and multimodal / Multimodal LLMs
MedRepBench: Benchmarking Structured Understanding of Medical Report Images
Fangxin Shang, Yuan Xia, Dalu Yang, Yahui Wang, BingLinYang BingLinYang
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 |
|---|---|---|---|
| apply GRPO with field-recall reward after SFT | InternVL3-8B (Ours-SFT) · MedRepBench lab reports (no-OCR) | average field recall (%): 73.31 -> 79.45 | Improved |
| apply GRPO with field-recall reward after SFT | InternVL3-8B (Ours-SFT) · MedRepBench lab reports (OCR-assisted) | average field recall (%): 82.37 -> 87.41 | Improved |
| apply GRPO after SFT | InternVL3-8B · MedRepBench subjective (no-OCR) | acceptability rate (%): 56.27 -> 60.64 | Improved |
| apply GRPO after SFT | InternVL3-8B · MedRepBench subjective (no-OCR) | excellence rate (%): 38.86 -> 42.45 | Improved |
| provide OCR text to the VLM | Qwen2.5-VL-32B · MedRepBench lab reports | Rvalue recall (%): 76.32 -> 84.72 | Improved |
| provide OCR text to the VLM | InternVL2.5-38B · MedRepBench lab reports | Rabnormal recall (%): 66.01 -> 81.52 | Improved |
| provide OCR text to the VLM | LLaMA-4-Maverick · MedRepBench lab reports | Rabnormal recall (%): 75.83 -> 86.56 | Improved |
| provide OCR text to the VLM | InternVL2.5-8B · MedRepBench lab reports | Rname recall (%): 71.38 -> 85.64 | Improved |
| scale InternVL3 from 8B to 38B | InternVL3 · MedRepBench subjective (no-OCR) | acceptability rate (%): 48.24 -> 51.15 | No clear effect |
| scale Qwen2.5-VL from 7B to 32B | Qwen2.5-VL · MedRepBench subjective (no-OCR) | excellence rate (%): 5.58 -> 51.40 | Improved |
| scale Qwen3 from 1.7B to 235B-A22B-2507 | Qwen3 (OCR+LLM) · MedRepBench lab reports | Rabnormal recall (%): 69.04 -> 88.03 | Improved |
| scale InternVL3 from 8B to 38B | InternVL3 · MedRepBench lab reports (no-OCR) | Rvalue recall (%): 71.29 -> 64.83 | Hurt |
Closest papers by meaning
- DAMPER: A Dual-Stage Medical Report Generation Framework with Coarse-Grained MeSH Alignment and Fine-Grained Hypergraph Matching (AAAI 2025 · similarity 0.87)
- MedBench: A Large-Scale Chinese Benchmark for Evaluating Medical Large Language Models (AAAI 2024 · similarity 0.86)
- MedHorizon: Towards Long-context Medical Video Understanding in the Wild (NeurIPS 2026 · similarity 0.86)
- Med-CMR: A Fine-Grained Benchmark Integrating Visual Evidence and Clinical Logic for Medical Complex Multimodal Reasoning (CVPR 2026 · similarity 0.85)
- M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding (ICLR 2026 · similarity 0.85)
- Beyond Classification Accuracy: Neural-MedBench and the Need for Deeper Reasoning Benchmarks (ICLR 2026 · similarity 0.85)
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.
