Loading
Pathology and radiology foundation models are AI networks pretrained without labels on over a million slides or scans (Virchow used 1.5 million), then adapted with small task heads to predict mutations, prognosis or treatment response from routine images. They power the FDA-cleared ArteraAI tools, but validation across hospitals and how regulators treat general-purpose models remain unsettled.
Virchow (Paige/MSK, 1.5M slides), UNI and CONCH (Harvard), Prov-GigaPath (Microsoft/Providence), PLUTO, and radiology models (Merlin, RadFM). They predict molecular alterations, prognosis, and treatment response from routine H&E and CT, and power the FDA-cleared ArteraAI tools. Multimodal patient-level models integrating genomics, imaging, and notes are in development (e.g., CanSim-style efforts, Tempus, Owkin).
Self-supervised pretraining (DINOv2, contrastive) on unlabelled images; frozen encoder plus small task heads.
Nothing in the corpus depends on this yet.
Dependencies are what this technology cannot be delivered without: manufacturing steps, instruments, software, upstream methods. See its full chain on the map.
An FDA-cleared AI test (May 2026) that reads breast cancer slides to estimate recurrence risk in early hormone-positive disease.
The first AI tool cleared by the FDA to predict both prognosis and treatment benefit from a routine biopsy slide, in prostate cancer.
The first AI for reading biopsy slides authorised by the FDA, which points pathologists to prostate cancer they might otherwise miss.
Two technology pages on OnCo cite this paper by its DOI; this record gives the citation a page of its own so a reader can follow it without leaving OnCo. Read the abstract above alongside the citing pages listed under Related; the record was created automatically from the Europe PMC entry and its figures have not been checked by hand.
Two technology pages and one idea page on OnCo cite this paper by its DOI; this record gives the citation a page of its own so a reader can follow it without leaving OnCo. Read the abstract above alongside the citing pages listed under Related; the record was created automatically from the Europe PMC entry and its figures have not been checked by hand.
Query for this technology: (TITLE:"foundation model" OR ABSTRACT:"foundation model") AND (TITLE:"pathology" OR ABSTRACT:"pathology" OR TITLE:"histopathology" OR ABSTRACT:"histopathology" OR TITLE:"radiology" OR ABSTRACT:"radiology") AND (cancer OR tumor OR tumour OR oncology OR carcinoma OR lymphoma OR leukemia OR leukaemia OR myeloma OR sarcoma OR melanoma OR glioma). Results are unfiltered search hits about Pathology & radiology foundation models, not a curated reading list.
Shares Noetik, Pathos AI, AI compute and model platforms for oncology, AI in oncology roadmap: pattern readers → foundation models → agents in the workflow and the tag frontier.
Shares Pathos AI and the tag frontier.
Shares NCT/UCC Dresden, University Hospital Carl Gustav Carus and the tag frontier.
Shares Diagnostics roadmap: stains → gene panels → blood tests that decide treatment and the tag frontier.
Shares AI in oncology roadmap: pattern readers → foundation models → agents in the workflow and the tag frontier.
Shares Diagnostics roadmap: stains → gene panels → blood tests that decide treatment and the tag frontier.
Shares PLUTO (PathAI), AI second reads to stop borderline lesions being upgraded to cancer, Version control and locked reference sets for AI algorithms used as companion diagnostics, A pre-competitive consortium to train a shared multimodal cancer foundation model.
Shares AI-first reading for high-volume common cancer diagnoses, pathologist for the exceptions, Turn the map of immune cells inside a tumour into a standardised test, AI that is built but not validated or deployed, AI in radiology.
Open-source projects that implement or serve this technology, from OnCo's own catalogue: licence and last activity as the repository reported them on the day of the fetch. Listing is not endorsement; check the licence before reuse and the validation before clinical use.
A general-purpose pathology foundation model trained on 100 million tissue patches; weights are gated on Hugging Face under a non-commercial licence.
Clinical Histopathology Imaging Evaluation Foundation model from Harvard for cancer detection, subtyping and outcome prediction.
Microsoft and Providence's whole-slide foundation model trained on 1.3 billion tiles, with released weights.
A toolkit for whole-slide preprocessing and feature extraction with many pathology foundation models.
Hierarchical image pyramid transformer for gigapixel whole-slide images.
A vision-language foundation model for pathology paired with UNI; gated weights, non-commercial.
Pathology Language-Image Pretraining: a CLIP-style model trained on pathology images from Twitter, with weights.
A whole-slide foundation model that produces slide-level embeddings and reports, from the Mahmood lab.
Commercial and regulated products that serve this technology. Each card says what is behind it: a regulator's database, the literature, a public body's list, or only the company's own words. Listing is not endorsement, and a clearance is a regulatory fact, not a clinical one.
An algorithm that flags areas suspicious for prostate cancer on a scanned biopsy slide, the first such pathology algorithm authorised in the United States.
PathAI's image management system for diagnostic laboratories, the platform its algorithms are meant to run inside.
Open-source software, hardware and data projects catalogued by a third party, the Open Medical Registry, that bear on this technology. Listing is not endorsement; check each project's own licence and validation before clinical use.
Evaluation framework for oncology foundation models (FMs)
Official code for Prediction of molecular subtypes for endometrial cancer based on hierarchical foundation model.
A Patient-First Foundation Model for Computational Pathology
From the Open Medical Registry (openmedical.sh), an MIT-licensed catalogue of open-source medicine. Blurbs are one line from each registry record; every project keeps its own licence.