Published in Nature Medicine, a paper on the study describes an AI system for computational pathology that achieves clinical-grade accuracy levels. Researchers developed deep-learning algorithms to build a system that can detect prostate cancer, skin cancer and breast cancer with near-perfect accuracy, according to the paper. These algorithms are based on a dataset of nearly 45,000 de-identified, digitized slide images from more than 15,000 cancer patients from 44 countries.
“After years of in-depth, comprehensive modeling, training, and testing, we are thrilled that Nature Medicine has published our paper, which demonstrates our ability to train accurate classification models at unprecedented scale, and validates our mission to create the world’s first clinical-grade, artificial intelligence in pathology,” said Paige co-founder and chief scientific officer Dr. Thomas Fuchs in a news release. Fuchs led the work at his lab at Memorial Sloan Kettering Cancer Center in New York.
The paper outlines how a series of novel algorithms created using datasets 10 times larger than those that have been manually curated performed better and also are more generalizable. Curating datasets can be prohibitively expensive and time intensive, according to New York-based Paige. By eliminating the need to curate datasets, the company said it can now develop many more algorithms that can be built into clinical decision support products.
“The publication in Nature Medicine of the algorithm developed by Dr. Fuchs’ lab is an important milestone for Paige,” said Christopher Kanan, lead AI scientist at Paige. “It demonstrates that AI has the potential to support pathologists in delivering quantitative and more accurate diagnoses, improving treatment for patients worldwide. Leveraging even larger training sets, over the past year, Paige has created novel vendor-agnostic systems that demonstrate even better accuracy.”
Paige has already built on the academic work described in Nature Medicine to develop a clinical product, based on technology currently under review by the FDA as a designated Breakthrough Device, for an intended indication different than the one described in the article.