New AI System Beats Radiologists for Locating Breast Cancer
Artificial intelligence tools are on the way to improving the practice of medicine, including mammography.
Artificial intelligence (AI) and automation are changing the world within which we live. For decades humans have worked intensely on improving machines to help with tasks—from robotic assembly lines to robotic surgery. Even as we try to imagine self-driving cars, braking sensors have been assisting us on slippery roads for years, and monitoring devices are common in every healthcare setting.
A recent study published in the journal Nature reports on advances in mammography tools created by Google in association with research institutions in the UK and the US. Although the new diagnostic AI tools are only experimental at this point, continued fine-tuning of the application holds promise for detecting breast cancers missed by human counterparts.
Radiology reads by human radiologists of mammograms x-rays are sometimes wrong. According to the American Cancer Society, screening mammograms do not locate approximately one in five cancers. False-positive mammogram results require women to go through additional testing or biopsy to rule out cancer. Approximately 50 percent of women who get an annual mammogram for ten years will be given a false-positive result.
Using images already read by radiologists, the experimental tool reduced false negative results (outcomes that do not identify a cancer that is present) by 9.4 percent. The number of false positive results was reduced by 5.7 percent. Interestingly, in Britain, the same tool also reduced false reads across the board—but by lower percentages. This suggests radiologists in the UK are more accurate, or there is some other variability in testing metrics.
During the study (funded by Google), study authors asked radiologists to read image results and ran the same images through the Google AI tool. Overall, the AI tool did a better job.
The experimental tool also makes mistakes, even as it continued to learn how to read mammograms. While in several trial phases the tool outperformed a team of six radiologists, in another instance, the tool missed a cancer than all six radiologists had identified.
Importantly, the AI tool learns iteratively, which means the more training the AI tool receives the more accurately it can identify what it is “seeing.” In time, the tool may consistently exceed the ability of experienced human radiologists to read mammography results. As well, the AI tool does not become distracted, take coffee breaks, or become fatigued after eight hours of straight image review.
Mammograms are an important screening tool. AI could become an important assist to human radiologists to improve accuracy—and reduce the possibility of misdiagnosis.
Skilled medical malpractice lawyers serving Maryland and Washington, DC
With 35 years of successful experience recovering compensation for those injured by medical negligence, Schochor, Federico and Staton, P.A. pursues your right to compensation after medical mistake. Call 410-234-1000 or contact us to set up a free appointment to discuss your case.