Deep learning excels at analysis of breast cancer biopsy slides
Researchers have developed a deep-learning computer network that is highly accurate in determining whether invasive forms of breast cancer are present in whole biopsy slides.
A research team led by Case Western Reserve University published results of their study in Scientific Reports, detailing their deep-learning network approach.
The study first involved training the network by downloading 400 biopsy images from multiple hospitals and then presenting the network with 200 images from The Cancer Genome Atlas and University Hospitals Cleveland Medical Center.
According to Anant Madabushi, professor of biomedical engineering at Case Western Reserve and co-author of the study, the network scored 100 percent accuracy in determining the presence or absence of cancer on whole slides.
“This is a study with 600 patients, so it’s fairly robust,” says Madabushi, who also directs Case Western Reserve’s Center of Computational Imaging and Personalized Diagnostics. “And there were a lot of human-machine comparisons done.”
In fact, compared with the analyses of four pathologists, the machine was more consistent and accurate, Madabushi contends.
“Pathologists are extremely busy, and we’re talking about microscopic-level detail in these tissue slides. So, clearly, for them to go in and pick out every cell of cancer was not tenable. There just wasn’t enough time for them to be able to sit down and manually do that,” adds Madabushi. “The network started to get more sophisticated, more granular and more accurate than the pathologists.”
Last month, the Food and Drug Administration approved the marketing of the Philips IntelliSite Pathology Solution, the first whole slide imaging (WSI) system that enables review and interpretation of digital surgical pathology slides prepared from biopsied tissue. The system enables pathologists to read tissue slides digitally to make diagnoses, rather than looking directly at a tissue sample mounted on a glass slide under a conventional light microscope.
According to Madabushi, this is the first time the FDA has permitted the marketing of a WSI system for these purposes, which he says is a major milestone for pathology. “A pathologist can look at an image of a slide on their computer monitor, and that is equivalent to the pathologist looking at a slide under their microscope,” he notes. “That means digital pathology—the digitization of slides—can now be used for primary diagnosis by a pathologist. That’s a game changer.”
He believes that as pathologists increasingly adopt digital pathology there will be “an even greater need for software and analytics like the one we published in this paper.” Ultimately, Madabushi contends that the FDA’s clearance of the Philips system “opens the door to an entire market for the analysis of digital pathology slide images.”