Loading
Pathology foundation models are image encoders pretrained without labels on millions of slide tiles; UNI and CONCH come from the Mahmood Lab at Harvard, Virchow from Paige, and CTransPath was an early transformer version.
UNI, from Chen and colleagues, is a general-purpose self-supervised pathology encoder trained on over a hundred million tiles and evaluated across dozens of tasks; UNI2-h is its larger successor released on Hugging Face under a gated non-commercial licence. Virchow, from Vorontsov and colleagues at Paige, was trained on about 1.5 million slides for clinical-grade and rare cancer detection, with Virchow2 released the same way. CTransPath (Wang and colleagues, 2022) showed contrastive pretraining of a transformer for histology, and CONCH aligns tiles with pathology text; TITAN extends the family to whole slides and reports. Gated weights are open for research but not for products, which is why OnCo records their openness as gated.
Showing the technology this term belongs to: Pathology & radiology foundation models.
Shares Attention-based multiple-instance learning (ABMIL, CLAM), Foundation model, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Open weights, open code and gated models, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Tile and patch encoding of slides, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Tile and patch encoding of slides, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Open weights, open code and gated models, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Open weights, open code and gated models, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares Foundation model, Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.
Shares TITAN (whole-slide multimodal model), Cancer AI vocabulary (CanSim terms map) and the tag cansim-terms.