The newly open-sourced foundation models accelerate cancer innovation with unprecedented access to the most trusted and advanced AI technology available today
NEW YORK–(BUSINESS WIRE)–Paige, a pioneer in AI-driven pathology solutions designed to improve the fight against cancer, today announced the open-source release of Paige’s Foundation Models, Open Paige Foundation Model (“OpenPFM”), including Virchow and PRISM, two pioneering foundation models for computational pathology, as well as an ML SDK to support the use of these models in the research community. Virchow is the first million-slide foundation model for cancer, while PRISM built on Virchow, is a vision and language model for slide-level cancer insights. These models are now available on Hugging Face, enabling AI developers, academic researchers, and innovators to create customized AI models and applications that accelerate cancer research and clinical development.
Developed in collaboration with Microsoft Research, Virchow and PRISM are designed to unlock deeper insights within pathology slides for a variety of end uses, surpassing historical capabilities. Harnessing an unmatched and globally diverse dataset of 1.5 million digitized pathology slides from Memorial Sloan Kettering Cancer Center – which is internationally recognized as a world-leading cancer center and ranked among the top two cancer hospitals in the United States – Virchow’s training data scale is the largest to-date. Recently published in Nature Medicine, Virchow aims to set a new standard of what is possible in computational pathology to change the way cancer is diagnosed and ultimately treated.
PRISM builds on the image representations generated by Virchow, combining them into a whole slide image signature. Since many pathology and cancer tasks are defined at the slide-level, PRISM adds a critical component to the computational pathology ecosystem. Trained with an additional 587,000 whole slide images and 195,000 clinical reports, PRISM generates written diagnostic summaries, offering a new way to gain insights into cancer and the pathological basis of disease*. PRISM can be used for standard diagnostic tasks such as cancer detection and sub-typing without needing additional training. It has also been shown to improve data efficiency, reducing the number of training samples traditionally required for AI development, as demonstrated in biomarker prediction across a wide range of tissue types.
“We are thrilled to extend access to our AI models by making them accessible to all,” said Siqi Liu, Director of AI Science at Paige. “This initiative highlights our best-in-class technology and underscores our commitment to fostering innovation, collaboration, and accessibility in AI research. The launch of OpenPFM, including Virchow and PRISM is just the beginning of the groundbreaking Foundation Model technology we are developing at Paige. It serves as a crucial building block in our revolutionary journey to transform cancer diagnosis and treatment.”
“The introduction of our open-source models unlocks new possibilities to further advance research for better patient care and accelerate clinical discovery,” said Razik Yousfi, SVP of Technology at Paige. “Our goal in offering access to our technology, that underpins our clinical-grade applications, is to drive innovation and help push the boundaries of what is possible in cancer diagnostics and drug development, ultimately transforming patient care.”
Virchow is available under the Apache 2.0 License. Both PRISM and the Paige ML SDK are available under the CC-BY-NC-ND 4.0 License. More advanced versions of Virchow, PRISM and other latest AI models from Paige are accessible through various commercial licensing options. These are designed to support AI requirements for clinical trials, R&D, and other commercial endeavors. Paige also offers collaboration opportunities with its team of experienced AI scientists and medical experts to enhance scientific discoveries.
For more information about licensing Virchow for commercial use contact [email protected].
*Diagnostic summaries provided by PRISM are for research use only.
