Digital pathology company Paige announced that it has developed an application leveraging artificial intelligence (AI) that can detect cancer from more than 17 different tissue types including, skin, lung, and the gastrointestinal tract, along with multiple rare tumor types and metastatic deposits.
The new application was developed by leveraging the company’s Foundation Model called Virchow, which allowed developers of the technology to use more than four million digitized slides to train its model across a multitude of cancer types. This is vastly different from the traditional approach of developing AI models that are trained to detect one tissue type at a time, which can take months or years to produce. The company says it is the first of its kind clinical grade application in AI-based cancer diagnosis.
“The early success of our Foundation Model has been possible due to the size, quality, and diversity of the datasets we used to build it,” said Siqi Liu, PhD, director of AI Science at Paige. “Paige has access to one of the largest and most highly regarded pathology datasets globally, which allows us to leverage cutting-edge deep-learning approaches to train systems to detect common, complex, and even very rare cancer entities. Paige’s development provides the pathology community with the most powerful tools for diagnosis, prognosis, biomarker development, and targeted selection of patients for precision therapy,” he continued.
According to Andy Moye, CEO of Paige, the company has a focus of developing clinical grade digital pathology and AI-enabled technologies and will therefore seek FDA regulatory oversight for any of its products that are based on the Foundation Model technology and noted: “We see FDA clearance as being critical to ensure that regulatory and safety standards are being upheld in the application of AI in cancer diagnostics across tumor types.”
Paige was the first company to received FDA marketing approval for an AI-enabled digital pathology diagnostic for Paige Prostate, received in September, 2021. The data from that study showed that the tool improved diagnostic detection of cancer by 7.3% compared with evaluation by a pathologist, while reducing false negatives diagnoses by 70% and false positives by 24%. The company is a 2018 spinoff from Memorial Sloan Kettering and used digitized slides from the cancer center initially, but has now tapped more than 1,000 other institutions in 45 countries to continue to build on its more than four million digitized cancer sample slides.
The new pan-tissue application was built by Paige with Microsoft Research via a collaboration announced in September last year. The goal is to capture much more subtle signatures that reflect the complexity of cancer and provide more accurate, individualized diagnoses.
With the development of this application using Virchow, Paige said its ability to detect cancer leveraging a range of different data has implications broader than diagnosis alone.
“Beyond its multi-tissue capabilities, the Foundation Model and the use of its embeddings can also be leveraged as a critical building block in a variety of upstream and downstream applications across the entire healthcare continuum,” said Razik Yousfi, senior vice president of Technology at Paige. “By combining the outputs of the Foundation Model with data types from other modalities, including genomics, radiology, and other health data, one can derive exponentially greater insights about the nature of cancer, its behavior, and response to specific treatments.”
