AI in Mammography: A Second Reader, Not a Final Answer

What NCI actually says a mammogram does

Recent studies have tested AI as an extra reader for screening mammograms. Some report finding more cancers. Some report lighter reading workloads. The practical question is how the tool behaves once it is added to a real screening program.

Start with the test itself. NCI describes mammography as an imaging test that uses low-dose x-rays to make pictures of the breast. Mammograms are used for breast cancer screening because they can find tumors at an earlier stage, before symptoms appear.

NCI also notes limits in the same breath. Mammography is more likely to miss cancer in women with dense breasts. And an abnormal result still needs more imaging, or a biopsy, before anyone can say the word cancer. AI does not change that sequence. Our page on how mammograms work covers the basics.

This page explains federal sources. It is not medical advice and does not suggest a test or treatment.

"AI in mammography" is not one thing

The software can be used in several different ways, and they are not interchangeable:

• ranking images so the most suspicious are read first
• marking an area for the radiologist to look at
• deciding whether a second human reader is needed
• reading independently as one of two readers

A useful headline names which of these was studied, and says whether a radiologist still reviewed every examination. Those are different interventions with different risks.

Callbacks are part of screening, not a mistake

Any story about finding more cancers needs the other half of the ledger.

NCI reports that in the United States about 10% of mammograms lead to a woman being called back for more testing. Of those, only about 7% end in a cancer diagnosis. More than half of women screened every year for ten years will have a false-positive result at some point, and many will have a biopsy as part of the follow-up.

Diana Miglioretti, who led an NCI-funded study of 3.5 million mammograms, pushed back on the idea that a false positive means someone erred. The purpose of a screening mammogram is to sort people into those unlikely to have cancer and those who need a closer look. The follow-up, she said, is a normal part of screening.

That does not make it costless. The same study found that women who had a false-positive result were less likely to return for routine screening later. False positives are also more common among younger women, women with dense breasts, women with previous biopsies, and women with a family history.

Overdiagnosis sits alongside this. NCI-covered research suggests the risk of overdiagnosis with routine screening mammography is substantial for women in their 70s and older, and rises with age and other health problems. Overdiagnosis is not a false positive. It is a real cancer that would never have caused harm.

A cautionary parallel already on the record

Three-dimensional mammography, also called digital breast tomosynthesis, offers a useful comparison. NCI states that combining it with standard mammography is better at finding tumors than standard mammography alone. NCI also states that it is still unknown whether it is more effective at reducing deaths from breast cancer. The NCI-sponsored TMIST trial was launched to answer exactly that.

Better at finding, not yet shown to save more lives. That gap is the one an AI study has to close too.

What a program should measure before switching AI on

A center should test the actual software, scanner mix, and reading process it plans to use. Then it should track cancer detection, callback rates, interval cancers, and workload. Checks need to continue after software updates, and results should be broken out by breast density and by demographic group, because an average can hide unequal errors.

What does not change

• AI support is not a diagnosis and does not replace a radiologist or a biopsy.
• Results from one product do not transfer to another product.
• Finding more abnormalities also creates more follow-up, more biopsies, and more worry.
• Screening decisions still rest on age and risk. NCI notes that the US Preventive Services Task Force currently recommends screening mammograms every two years for women at average risk between ages 40 and 74.

If you have been told you have dense tissue, what to ask after a density notice is a separate conversation from AI. Background on the disease itself is in our breast cancer overview.

Questions for your imaging center

• Is AI used here, and does a radiologist still read every exam?
• Has this system been tested in people like me?
• How are disagreements between the software and the radiologist handled?
• Has your callback rate changed since the software was introduced?

How this article was prepared

An AI-assisted editorial system helped prepare this page. No named medical reviewer has reviewed it unless one is listed.

The National Cancer Information Foundation publishes Cancer Explained. This page is for learning. It is not medical advice and does not suggest a test or treatment.

See an error, old source, or unclear wording? Tell us.