Knowledge-enhanced pretraining for vision-language pathology foundation model on cancer diagnosis

Researchers led by Zhou et al. have developed KEEP, a knowledge-enhanced foundation model that integrates structured disease knowledge into vision-language pretraining for computational pathology. Published in the prestigious journal Cancer Cell, the approach combines curated medical knowledge graphs with large-scale pathology image-text datasets to improve AI-driven cancer diagnosis. The model demonstrates notably improved diagnostic accuracy and generalization, with particular gains for rare cancer subtypes where training data is traditionally scarce. This work highlights a promising direction for AI in oncology, showing that embedding domain expertise into foundation models can meaningfully advance automated cancer detection beyond what data-driven methods alone achieve.