There are few fields as ripe for AI improvement as medicine. We covered it extensively back in May, but the advances keep coming at an ever-quickening pace (it’s also worth noting that that post is more about why medicine is a great candidate for AI improvement and how doctors can use AI to do their jobs better, whereas this one is about a specific development showing that concept in action). Some of the intervening advances have proven so fascinating/promising, we couldn’t help but amend our coverage to include them. November is just such a month because Google’s deep-learning AI proved, in a limited-scale test, to be more adept at identifying breast cancer that has metastasized to a patient’s lymph nodes than human pathologists.
Breast cancer is a silent and pernicious murderer. As with any type of cancer, metastasis into the lymph nodes typically makes for a more complex and arduous treatment and recovery plan. So, detecting nodal metastases is a huge part of pathologists’ prerogatives when diagnosing any type of cancer. In this specific case, researchers deployed an ‘off-the-shelf’ Google deep-learning approach — which they titled LYNA for LYmph Node Assistant — to aide pathologists in identifying nodal metastases in breast cancer patients.
According to the BBC’s ScienceFocus, what makes LYNA’s approach novel (for machines, anyway) is how it reviews samples at differing magnifications, much as a pathologist would do the same under a microscope.
Apparently, pathologists’ accuracy wanes the more time pressure they’re under (and when are doctors not under time pressure?); and, the smaller the offending metastasis, the harder for the pathologist to see, especially when in said time crunch. According to the same article, in some cases, “only 38 per cent of small metastases are picked up by pathologists when samples are reviewed under time constraints, and right now, that pathologist’s examination is the gold standard in diagnosis of nodal metastases.
Use LYNA, of course! “The algorithm’s first test showed that LYNA was able to correctly distinguish a slide with cancer from a slide without 99 per cent of the time, even when the regions were too small to be detected by pathologists,” from the BBC again.
In the second test the researchers arranged, six pathologists conducted a diagnostic test, both with and without LYNA’s help. With LYNA’s help, the doctors “found it ‘easier’ to detect small metastases, and on average the task took half as long” the BBC continued. “Pathologists working with LYNA’s assistance were more accurate than both unassisted pathologists and the LYNA algorithm working alone.”
No, LYNA is not better than a pathologist right this second. This was an incredibly limited sample size looking at one specific type of cell sample — we can’t simply extrapolate that to all pathology. But what it does show is that when used strategically, AI really can help doctors do a better, more accurate job faster. And, it shows that in the course of a few short months, advances involving this type of technology and innovation are being tested in real-world scenarios more and more.
Medicine really is a phenomenal frontier for improvement via AI. We’re excited to working with this transformational technology to hopefully bring about better care and improved outcomes for patients and their families.