Research map / Vision-language and multimodal / Cross-modal learning
KAMP: Knowledge-Anchored Multimodal Pretraining Framework for Medical Image Representation
Feiyu Huang, Jia Li, Zhao Chen, Yang Wu, Caleb Chen Cao, Lei Chen
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 Stage-2 GRPO refinement of the knowledge generator and Stage-3 aligner retraining (ablation) | KAMP (GenPers aligner) · FedBCa (20-shot) | macro-AUC: 79.6 -> 82.6 | Improved |
| use GRPO instead of PPO or DPO for generator refinement (ablation) | KAMP Stage 2 · FedBCa (20-shot) | macro-AUC | Improved |
| add PPO or DPO refinement of the generator (ablation) | KAMP Stage 2 · FedBCa (20-shot) | macro-AUC | Improved |
| inject GPT-5 personalized descriptions (clinical text + metadata) into the image branch instead of no text (ablation) | KAMP Stage-1 aligner · BraTS23-MEN, FedBCa, TG-LIVT (20-shot) | macro-AUC | Improved |
| inject the class label as text instead of no text (ablation) | KAMP Stage-1 aligner · BraTS23-MEN, FedBCa, TG-LIVT (20-shot) | macro-AUC | No clear effect |
| inject raw patient clinical text instead of no text (ablation) | KAMP Stage-1 aligner · BraTS23-MEN, FedBCa, TG-LIVT (20-shot) | macro-AUC | No clear effect |
| use GPT-5 text conditioned only on class label (ablation) | KAMP Stage-1 aligner · BraTS23-MEN, FedBCa, TG-LIVT (20-shot) | macro-AUC | Mixed |
| KAMP versus tri-modality DRIM in the 3-shot regime | KAMP · FedBCa (3-shot) | macro-AUC: 74.3 vs 75.6 | Hurt |
| KAMP versus best baseline at 20-shot on BraTS23-MEN | KAMP · BraTS23-MEN (20-shot) | macro-AUC: 62.1 -> 70.2 | Improved |
Closest papers by meaning
- M-IDoL: Information Decomposition for Modality-Specific and Diverse Representation Learning in Medical Foundation Model (ICML 2026 · similarity 0.91)
- Multi-Granularity Cross-modal Alignment for Generalized Medical Visual Representation Learning (NeurIPS 2022 · similarity 0.89)
- PRIOR: Prototype Representation Joint Learning from Medical Images and Reports (ICCV 2023 · similarity 0.89)
- ProbMED: A Probabilistic Framework for Medical Multimodal Binding (ICCV 2025 · similarity 0.88)
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.88)
- MLIP: Enhancing Medical Visual Representation with Divergence Encoder and Knowledge-guided Contrastive Learning (CVPR 2024 · 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.
