In collaboration with Microsoft, Paige has unveiled the second generation of Virchow, its million-slide foundation model for cancer diagnosis, which the team says are “the world’s largest and most advanced artificial intelligence (AI) models in clinical pathology, Virchow2 and Virchow2G. They say, offer a deeper understanding of cells and tissue, aiming to redefine cancer diagnosis and treatment.”
“Pathology is the cornerstone of medicine, but pathologists have been reliant on: microscopes for over a 100 years,” said Razik Yousfi, senior vice president of technology at Paige. “Now we have a chance to improve accuracy while speeding things up.”
Accelerating slide digitization, he points out, will be one of the key goals for the field. Also, achieving reimbursement, “It’s slow and poor now,” he said.
Foundation models are machine learning models trained on a broad spectrum of generalized and unlabeled data and capable of performing a wide variety of general tasks. Paige has used its foundation model technology to develop a universal clinical AI application for pathologists to aid in the identification and diagnosis of cancer across over 40 tissue types.
“This second generation of Paige’s Virchow model outperforms anything in the industry and continues to grow in knowledge and capability, bringing us closer to making precision medicine a reality,” said Yousfi.
The models were built with a diverse dataset of over three million pathology slides from over 800 labs and 45 countries and trained using de-identified data from over 225,000 patients. The models aim to offer a broad spectrum of gender, race, ethnicity, and geographical regions, providing a more holistic understanding of cancer.
The dataset also includes over 40 different tissue types stained with H&E and diverse immune-stains (IHC). With 1.8 billion parameters, tripling the size of the team’s previous models, they say Virchow2G is the largest pathology model ever created.
Trained in collaboration with Microsoft researchers and using Microsoft’s advanced supercomputing infrastructure, these models set a new record in AI training scale, surpassing previous performance standards have been detailed, as showcased in a recently published technical report.
“Our collaboration with Microsoft has been pivotal in the development of Virchow, which has already earned recognition in Nature Medicine,” said Thomas Fuchs, founder and chief scientist of Paige.
He added, “We are merely scratching the surface of what these foundation models can achieve in transforming our understanding of cancer through computational pathology. Virchow’s immense scale unlocks key information that can be used to drive groundbreaking innovations, enabling precise diagnostics, targeted treatments, and personalized patient care…This is the beginning of a new era in oncology, where technology and science converge to combat cancer more effectively than ever before.”
“We are not only expanding capabilities, increasing accuracy, and reducing time in the cancer diagnosis process, but also pushing the boundaries of what’s possible. Our goal is to continue to bring the most advanced AI to pathology, leading to better patient outcomes and significant advancements in disease understanding and treatment,” said Yousfi.
