An algorithm that detects prostate cancer could open the door for other applications in medicine
Prostate cancer may have met its match.
The FDA just approved the first artificial intelligence platform to help pathologists detect the cancer, which claims about 34,000 American lives each year, according to the American Cancer Society.
Though this is not the first AI application to enter medicine — the field of radiology has seen clinical adoption of algorithms rise from zero in 2015 to around 30% in 2020, according to a study by the American College of Radiology. The FDA approved some 50 algorithms through Nuance’s AI Marketplace. Breast cancer and detecting cardiovascular disease through retina scans are also growing applications. But prostate cancer diagnosis is new to the field.
This new product, called Paige Prostate, comes out of Paige AI, a company spun out of the renowned cancer-research institution, Memorial Sloan Kettering Cancer Center, and works alongside humans to help increase their chances of catching early-stage tumors. In the study submitted for the FDA clearance, 16 researchers looked at 527 digitally-scanned biopsy slides, 171 of which included cancerous cells. Pathologists who were assisted by the software improved detection of cancer by 7.3%, on average, compared to those who were unassisted.
“I think this is a reason to jump for joy,” says Michael Haffner, Assistant Professor in the Divisions of Human Biology and Clinical Research at the Fred Hutchinson Cancer Research Center and in the Department of Laboratory Medicine & Pathology at the University of Washington. “The work of the pathologist will be so much more efficient because we won’t have to look at every single slide. Instead, we can use Paige AI to pick the slides that have the highest likelihood of showing cancer.”
Pathology as a field has not embraced new technology for decades, bemoans Dr. Haffner. For one, there is a bottleneck of digitization of biopsy slides, making diagnosing extremely time consuming. In essence, doctors are trained to look at slides of cell clusters under microscopes and scan for anomalies. But they’re required to look at many slides, perhaps 20 per patient, of which the vast majority have no tumors. Since algorithms are so great at detecting patterns, Paige AI’s product acts as a diagnostic tool to screen slides (once they are finally digitized) at a rapid pace, then highlight areas that seem abnormal. “So if there is a way to screen out all the slides that have a low chance of cancer, that increases my workflow dramatically,” explains the pathologist.
Though one could argue that there wasn’t an enormous need for such a diagnostic aid — after all, does a 90-year-old mean really need to know that he has Stage 1 cancer? — Dr. Haffner believes that this FDA approval opens the door to other technological advances in the cancer field. “What we really want to know is, just based on how the cancer looks under a microscope, what is the best form of treatment for this patient? There is research supporting the idea you can extract so much information from a digital slide that is not immediately available to even the most skilled pathologist,” he says.
There are, of course, some ethical quandaries and confounders that may stump an AI system, causing it to make the wrong diagnosis. The data sets on which an AI algorithm is trained could be biased in an obscure way, or could be using localized data that is not able to be equitably distributed. Moreover, if an AI program makes a mistake, who is to blame?
All that said, many pathologists are pretty excited, and patients aren’t complaining about having another human-assisted, AI-enhanced diagnostic tool. For anyone getting a cancer screen in the near future, you may find your doctor paging Dr. Paige.