Computational Pathology Foundation Models
Self-supervised encoders trained on whole-slide tissue images for diagnosis and biomarker tasks.
Tier Grades
Tier A: Frontier AI labs
feasible
Big hospital and tech partnerships hold the 100,000-slide archives used for UNI.
Tier B: Indonesian AI labs
not feasible
An Indonesian lab lacks slide archives of that size; using released encoders is possible.
Tier C: Institution units such as ITB
stretch
A university hospital unit can fine-tune released encoders on local slides.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Pilots need hospital data access and clinical oversight, so delivery is a stretch.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataUNI is pretrained on over 100 million tissue patches from over 100,000 H&E whole-slide images across 20 tissue types.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 20 authors on the UNI paper (arXiv 2308.15474).
- Time horizonYear.
- Shelf-life signalsSlide-level multimodal models replace patch encoders, Regulatory approval of a foundation-model-based diagnostic
- Pipeline roleTier A and hospital labs pretrain encoders; Tier C validates them on local slides; Tier D supports hospital pilots under clinical governance.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicComputational Pathology Foundation Models
Models
UNIMahmood Lab, Harvard, vision
Prov-GigaPathMicrosoft AI, vision
VirchowPaige, visionCONCHMahmood Lab, Harvard, vision+8 more
UNIMahmood Lab, Harvard, visionProv-GigaPathMicrosoft AI, visionVirchowPaige, visionCONCHMahmood Lab, Harvard, vision+8 moreWho Runs Itnone recorded
Models 12
CONCHMahmood Lab, Harvard, visionexemplarProv-GigaPathMicrosoft AI, Providence, visionexemplarUNIMahmood Lab, Harvard, visionexemplarUNI2-hMahmood Lab, Harvard, visionproduced byVirchowPaige, visionexemplarVirchow2Paige, visionproduced by
H-optimus-0Bioptimus, visionproduced by
HibouHistAI, visionproduced byTITANMahmood Lab, Harvard, visionproduced by
PhikonOwkin, visionproduced byCTransPathvisionproduced byGigaTIMEvisionproduced byPapers
- Data Efficient and Weakly Supervised Computational Pathology on Whole Slide ImagesMing Y. Lu et al., arXiv preprint, 2020 founding
- Towards a Visual-Language Foundation Model for Computational PathologyMing Y. Lu et al., arXiv preprint, 2023 state of the art
- A General-Purpose Self-Supervised Model for Computational PathologyRichard J. Chen et al., arXiv preprint, 2023 state of the art
- Virchow: A Million-Slide Digital Pathology Foundation ModelEugene Vorontsov et al., arXiv preprint, 2023 state of the art
- A Survey of Pathology Foundation Model: Progress and Future DirectionsConghao Xiong et al., IJCAI 2025 (per arXiv comment), 2025 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2308.15474 https://arxiv.org/abs/2308.15474
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Computational Pathology Foundation Models", https://aurion.semai.id/topics/computational-pathology-foundation-models/, records of October 2026.