Using AI to read mammograms
Hester Hill Schnipper, LICSW, OSW-C Program Manager Emeritus, Oncology, Social Work
JANUARY 27, 2020
We all know that mammography is a good, but imperfect, tool. There continues to be some controversy about when women should begin to have mammograms and how often they should schedule them. The American Cancer Society recommends that women, at normal risk, begin to have mammograms at age 45 and continue every other year until 55. At that point, the recommendation is an annual exam. Women who are at higher risk are told to begin sooner and always have a yearly exam. As a general rule of thumb, a daughter of a breast cancer survivor should have a first mammogram ten years earlier than her mother's age of diagnosis or at age 45, whichever comes sooner.
A second controversial area of mammograms is their accuracy. They can miss things, and they can't always discern what they do see. A biopsy is always recommended if a radiologist is concerned about a finding. There are recommendations that all mammograms should be double read, meaning two doctors should review them. As we all know, humans make mistakes.
Finally, dense breasts limit the view and accuracy of mammograms, and younger, pre-menopausal women, and some older women, have dense breast tissue. In some states, including Massachusetts, laws have been passed that mandate breast density be included in the mammogram report that is sent to patients. This can be very upsetting for a worried woman as the required legal language is serious. It likely reads something like this:
“Your mammogram shows that your breast tissue is dense. Dense breast tissue is common and is not abnormal. However, dense breast tissue can make it harder to evaluate the results of your mammogram and may also be associated with an increased risk of breast cancer. This information about the results of your mammogram is given to you so you will be informed when you talk with your doctor. Together, you can decide which screening options are right for you. A report of your results was sent to your primary physician.” (Excerpted from the American Cancer Society.)
A new and encouraging area of research is the use of AI or artificial intelligence to read mammograms. It turns out that computers are better at this than humans. (And, yes, I agree that is a little scary.) A recent article in the journal Nature reported that researchers from Google, the United States, and the United Kingdom are finding that computers can be trained to recognize patterns and interpret images. Something similar is already being used to detect lung cancers on CT scans, diagnose eye disease in people with diabetes, and view cancer cells on slides.
The study tested the computers' readings of images where the diagnosis was already known, and the new system did better than did radiologists. In the United States, there was a 9.4% reduction in false negatives and a 5.7% reduction in false positives. This seems quite significant and will eventually be a step forward in correctly finding early breast cancers.