Nearly a year after it announced a collaboration with Microsoft, digital pathology player Paige unveiled the second generation of an image-based artificial intelligence model to help detect cancer.
In September, Paige tapped Microsoft to construct what it called the “world’s largest” AI model built upon countless images of cancer biopsies.
The updated versions of those models, called Virchow 2 and Virchow 2G, feature more data diversity and depth, the company claims. The models were built with a diverse data set of more than 3 million pathology slides from over 800 labs and 45 countries, according to Paige executives.
The AI models were then trained using de-identified data from more than 225,000 patients, encompassing a “broad spectrum of gender, race, ethnicity, and geographical regions” and providing a more “holistic understanding” of cancer, according to Paige,
The data set also includes more than 40 different tissue types stained with H&E and diverse immune stains, making it suited to a wider variety of applications.
Paige touts the size of its Virchow 2G model in particular, boasting 1.8 billion parameters. The company, a former Fierce 15 winner, claims the AI models break new ground in AI training scale, surpassing previous performance standards.
The company collaborated with Microsoft researchers to train the AI models, using the tech giant’s supercomputing infrastructure.
A study published in Nature Medicine in July found that the Virchow AI model detected nine common cancer types and seven rare cancers as well as or better than tissue-specific clinical-grade models in production.
“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 at Paige. in a statement. “We are merely scratching the surface of what these foundation models can achieve in transforming our understanding of cancer through computational pathology.
Razik Yousfi, Paige’s senior vice president of technology, boasted that the second generation of the Virchow model “outperforms anything in the industry.”
“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,” Yousfi said in a statement.
Paige, which was originally launched out of Memorial Sloan Kettering Cancer Center, previously developed an AI model for spotting the telltale signs of tumors among tissue samples by using more than 1 billion images taken from about 500,000 pathology slides spanning multiple types of cancer.