Providence, Microsoft, and the University of Washington have developed an open-weight, AI-powered pathology model called Prov-GigaPath.
Prov-GigaPath, according to a May 22 news release from Providence, is designed to improve patient care and transform cancer diagnostics by capturing global patterns across whole pathology slides.
The model, trained on 1.3 billion pathology image tiles from 171,189 digital whole slides provided by Renton, Wash.-based Providence, is one of the first foundational models for digital pathology that has been pretrained using real-world data.
Additionally, unlike prior models that analyze small portions of slides, Prov-GigaPath captures both local and global patterns across entire slides using the GigaPath vision transformer architecture for pre-training, according to the release. It was also found to outperform existing pathology models.
In a study published in Nature on May 22, Providence, Microsoft, and the University of Washington researchers created a benchmark using data from Providence and the Cancer Genome Atlas program, covering nine cancer subtyping tasks and 17 pathomics tasks. With pretraining and advanced modeling, Prov-GigaPath achieved top performance on 25 of 26 tasks and outperformed the second-best method on 18 tasks.
With Prov-GigaPath now globally available, Providence's Chief Medical Officer, Carlo Bifulco, MD, told Becker's that its comprehensive whole-slide modeling capabilities could unlock new approaches to studying the tumor microenvironment.
This, according to Dr. Bifulco, could have significant downstream applications in cancer diagnostics and prognostics, aiding clinicians in treatment selection and offering broader biomedical impacts.