Research map / Vision-language and multimodal / Multimodal LLMs
Constructing Ophthalmic MLLM for Positioning-diagnosis Collaboration Through Clinical Cognitive Chain Reasoning
Xinyao Liu, Diping Song
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 multi-turn cognitive chain structure instead of single-turn instructions (ablation) | FundusGen fine-tuning · GMAI-MMBench | accuracy: 63.2% -> 66.7% | Improved |
| add multi-turn cognitive chain structure instead of single-turn instructions (ablation) | FundusGen fine-tuning · Fundus-MMBench | accuracy: 67.1% -> 69.7% | Improved |
| add region-annotated instructions (ablation) | FundusGen fine-tuning · Fundus-MMBench | accuracy: 65.3% -> 68.9% | Improved |
| add region-annotated instructions (ablation) | FundusGen fine-tuning · GMAI-MMBench | accuracy: 59.3% -> 64.7% | Improved |
| add General Report startup data (ablation) | FundusGen fine-tuning · Fundus-MMBench | accuracy: 62.9% -> 66.2% | Improved |
| add General Report startup data (ablation) | FundusGen fine-tuning · GMAI-MMBench | accuracy: 56.4% -> 60.9% | Improved |
| add cognitive chain data for 21 complex diseases (ablation) | FundusGen fine-tuning · GMAI-MMBench (21 complex diseases) | accuracy: 70.4% -> 75.2% | Improved |
| fine-tune on FundusGen | LLaVA-v1.5 · Fundus-MMBench and GMAI-MMBench average | accuracy: 26.9% -> 40.0% | Improved |
| fine-tune on FundusGen | Qwen2-VL · Fundus-MMBench and GMAI-MMBench average | accuracy: 34.7% -> 56.9% | Improved |
| fine-tune InternVL2.5 on FundusGen | InternVL2.5-8B · GMAI-MMBench | accuracy: 36.5% -> 66.7% | Improved |
| use FundusGen with cognitive chains instead of Classification Annotation-Guided Data | InternVL2.5-8B · GMAI-MMBench | accuracy vs data scale | Improved |
| generate synthetic reports with FundusExpert instead of GPT-4o | Qwen2-VL-7B · Fundus-MMBench | accuracy: 35.6% -> 40.1% | Improved |
Closest papers by meaning
- Multi-OphthaLingua: A Multilingual Benchmark for Assessing and Debiasing LLM Ophthalmological QA in LMICs (AAAI 2025 · similarity 0.87)
- M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding (ICLR 2026 · similarity 0.87)
- Chiron-o1: Igniting Multimodal Large Language Models towards Generalizable Medical Reasoning via Mentor-Intern Collaborative Search (NeurIPS 2025 · similarity 0.87)
- MedMO: Grounding and Understanding Multimodal Large Language Model for Medical Images (CVPR 2026 · similarity 0.86)
- OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning (CVPR 2026 · similarity 0.86)
- OpenMedReason: Scientific Reasoning Supervision for Medical Vision–Language Models (NeurIPS 2026 · similarity 0.86)
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.
