Knowledge-enhanced pretraining for vision-language pathology foundation model on cancer diagnosis - PubMed
9 hours ago
- #cancer diagnosis
- #pathology foundation model
- #knowledge enhancement
- Introduces KnowledgE-Enhanced Pathology (KEEP), a vision-language foundation model for cancer diagnosis.
- Systematically integrates medical knowledge into pretraining using a disease knowledge graph.
- Knowledge graph includes 11,454 diseases and 139,143 attributes, organizing image-text pairs into 143,000 groups.
- Aligns visual and textual representations within hierarchical semantic spaces for better disease understanding.
- Outperforms existing models on 18 public benchmarks and 4 rare cancer datasets, especially for rare subtypes.