Research map / Vision-language and multimodal / CLIP and zero-shot
Knowledge-enhanced Visual-Language Pretraining for Computational Pathology
Xiao Zhou, Xiaoman Zhang, Chaoyi Wu, Ya Zhang, Weidi Xie, Yan-Feng Wang
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 |
|---|---|---|---|
| initialize text encoder with pathology knowledge encoder instead of PubMedBERT (ablation) | KEP-32 · KatherColon zero-shot | weighted F1 median: 0.486 -> 0.530 | Improved |
| add frozen knowledge-encoder distillation branch (ablation) | KEP-32 · KatherColon zero-shot | weighted F1 median: 0.530 -> 0.563 | Improved |
| add frozen PubMedBERT distillation branch instead of knowledge encoder (ablation) | CLIP + PubMedBERT · PathPair disease retrieval | Recall@1: 0.161 -> 0.034 | Hurt |
| add visual projection head (ablation) | CLIP + PubMedBERT · KatherColon zero-shot | weighted F1 median: 0.407 -> 0.486 | Improved |
| initialize visual encoder with BiomedCLIP instead of CLIP (ablation) | KEP · KatherColon zero-shot | weighted F1 median: 0.563 -> 0.580 | Improved |
| use AdaSP loss instead of Triplet loss for knowledge encoding (ablation) | KEP-32 · PathPair disease retrieval | Recall@1: 0.265 -> 0.409 | Improved |
| add knowledge enhancement with CLIP initialization (KEP-32 vs PLIP) | PLIP · Arch-PubMed i2t | Recall@10: 0.067 -> 0.098 | Improved |
| add knowledge enhancement (KEP-32 vs PLIP) | PLIP · Arch-book t2i | Recall@10: 0.165 -> 0.148 | Hurt |
| add knowledge enhancement (KEP-32 vs QuiltNet) | QuiltNet · Arch-PubMed i2t | Recall@10: 0.139 -> 0.140 | No clear effect |
| use knowledge-enhanced KEP-CTP vs MI-Zero without in-house data | TCGA-BRCA subtyping · common breast cancers Top-1 | balanced accuracy: 0.390 -> 0.443 | Improved |
| use KEP-16 vs PLIP on breast subtypes | PLIP · TCGA-BRCA rare Top-1 | balanced accuracy: 0.115 -> 0.217 | 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.
- PLIP — PLIP: Language-Image Pre-training for Person Representation Learning (NeurIPS 2024) · better on 5 of 6 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
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.90)
- PathFLIP: Fine-grained Language-Image Pretraining for Versatile Computational Pathology (AAAI 2026 · similarity 0.90)
- Decomposing Disease Descriptions for Enhanced Pathology Detection: A Multi-Aspect Vision-Language Pre-training Framework (CVPR 2024 · similarity 0.90)
- Multi-Resolution Pathology-Language Pre-training Model with Text-Guided Visual Representation (CVPR 2025 · similarity 0.90)
- PathView-Bench: Can Multimodal Large Language Models Achieve Fine-grained Multiscale Understanding of Pathology Images? (NeurIPS 2026 · similarity 0.90)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.89)
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.
