Multi-Modal Foundation · google
Path Foundation
Vision transformer developed by Google to extract feature embeddings from histopathology image patches.
What Path Foundation is
Path Foundation is a digital pathology model developed by Google for biomedical and clinical research. It processes microscopic tissue images to produce medical feature representations for computational pathology workflows.
The model addresses the challenge of interpreting complex histological images by learning general representations from large collections of tissue patches, facilitating automated image classification and downstream cellular analysis.
How it works
The model uses a vision transformer architecture implemented within the tf-keras library. It was pre-trained on 60 million image patches derived from The Cancer Genome Atlas (TCGA) to learn general-purpose feature representations from whole-slide tissue images.
Key facts
- Architecture
- Vision Transformer
- Training data
- 60 million patches from TCGA
- Framework
- tf-keras
- Category
- Pathology
What you can use it for
- Histopathology Feature Extraction
- Generating compact numerical embeddings from histological tissue patches for downstream statistical and machine learning analyses.
- Tissue Image Classification
- Categorising microscopic tissue patches into distinct diagnostic or histological classes.
- Digital Pathology Research
- Supporting exploratory oncology and pathology research using representations learned across diverse TCGA cancer cohorts.
Run it here
We list this model but cannot run it yet: gated on the Hub — never run past a gate the user did not accept.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Pre-trained on a diverse corpus of 60 million histopathology patches from The Cancer Genome Atlas.
- Functions both as an image classifier and as a feature extractor for medical embeddings.
Watch out for
- Training is limited to tissue samples sourced from The Cancer Genome Atlas, which may not generalise to all clinical staining preparations or non-oncological conditions.
- Downstream classification requires appropriate task-specific adaptation when using the model purely as an embedding extractor.
Inputs and outputs
You provide
Image
PNG, JPEG or TIFF. Sent to the model as-is.
You get back
Labels with confidence scores, highest first.
Licence
Check the licence before commercial use · other
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use Path Foundation commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
What architecture is Path Foundation built on?
It is built on a vision transformer architecture implemented using tf-keras.
What training dataset was used for this model?
It was trained on 60 million patches extracted from The Cancer Genome Atlas (TCGA).
What are the primary applications of this model?
The primary applications are image feature extraction, generating histopathology embeddings, and histological image classification.
How to cite it
arXiv:2310.13259
Written from Path Foundation's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.