When Artificial Intelligence Meets Pathology

In a recent Nature Medicine article, MSK’s Dr. Thomas Fuchs and his colleagues describe how they trained artificial intelligence (AI) to process pathology slides at large scale. The research was completed with the start-up Paige.AI and is reported by The Cancer Letter, Tech Crunch, Becker’s Healthcare, and 360 DX.

Dr. Thomas Fuchs. Credit: Richard DeWitt

To train deep learning systems, researchers typically need to manually curate data, a time-consuming process that has previously prevented large-scale AI applications to pathology datasets. In this study, researchers instead trained the system using only the slide-level diagnostic information already in the electronic health record. The study included 44,732 slides of prostate, skin, or axillary lymph node tissue from 15,187 patients at 800 institutions. Their calculations indicate that the model could reduce a pathologist’s workload by up to 75%, classifying cancer without sacrificing sensitivity.

To learn more about artificial intelligence, including a list of the latest AI-related publications from MSK authors, check out the Library’s new LibGuide.

Of Mammograms and Men

A study by MSK researchers published this month in Breast Cancer Research and Treatment and reported by AuntMinnie.com found that men at increased risk for breast cancer could benefit from screening mammograms.

The authors performed a retrospective review of 163 asymptomatic men at increased risk for breast cancer due to family history, personal history, or BRCA1/BRCA2 genetic mutations. After reviewing 806 screening mammograms done over nearly 7 years, the authors calculated a cancer detection rate of 4.9 per 1,000 mammograms, comparable to average-risk women. This indicates that screening mammography could be a useful tool in the high-risk male population, though the authors state that larger studies are needed to strengthen this conclusion. There are currently no guidelines on the use of screening mammography in men.