AI Is Learning to Read Mammograms
Fri, December 3, 2021

AI Is Learning to Read Mammograms

The researchers' AI system performed better than radiologists / Photo Credit: Chompoo Suriyo

 

Per the research of Scott Mayer McKinney and colleagues of journal portal Nature, AI can help doctors find breast cancer on mammograms, as reported by Denise Grady of New York City-based newspaper The New York Times. X-rays of the breast are the new system for reading mammograms. However, they are not yet available for widespread use, as it is one of Google’s ventures into medicine. This paper will help move things along quite a bit. This paper will help move things along quite a bit, according to Dr. Constance Lehman, director of breast imaging at the Massachusetts General Hospital in Boston, who was not involved in the study.

The new system was tested on images where the diagnosis was already known. “We took mammograms that already happened, showed them to radiologists and asked, ‘Cancer or no?’ and then showed them to A.I., and asked, ‘Cancer, or no?’” Dr. Mozziyar Etemadi recounted, an author of the study from Northwestern University. The researchers found that it performed better than radiologists. On scans from the US, the new system generated a 9.4% reduction in false negatives, in which a mammogram is misread as normal, thereby missing the cancer. It also produced a lowering of 5.7% in false positives, where the scan is misidentified as abnormal but there is no cancer. On mammograms performed in Britain, the system also performed better than radiologists with a reduction in false negatives by 2.7% and false positives by 1.2%. 

In the US, about 33 million screening mammograms are performed annually, but it misses about 20% of breast cancers, with false positives being common, as found by the American Cancer Society, a voluntary organization. This results in women being called back for tests and biopsies. Hence, doctors wanted to make mammograms more accurate. Dr. Lehman argued, “There are many radiologists who are reading mammograms who make mistakes, some well outside the acceptable margins of normal human error.” 

However, Dr. Lehman noted that the patients studied might not reflect the general population. Racial makeup was not specified and only six radiologists were included in the study, which were not always reliable. She added, “We have to be very careful. We want to make sure this is helping patients.”