Using AI to Detect Breast Cancer: What We Know

Whether you realize it or not, artificial intelligence (AI) is a part of everyday life—guiding everything from new playlists on Spotify to the responses of customer service chatbots. Doctors have long used computers to flag when something appears off in medical images. But recent studies suggest that AI may be able to spot cancer in a mammogram or other breast imaging that even a well-trained radiologist might miss. It may also predict people most likely to develop breast cancer between mammograms.

Artificial intelligence is the ability of a computer to imitate human behavior (for example, to learn and act). AI developers train computers to recognize patterns in large amounts of data. Once a program has been trained, it can begin evaluating new data on its own and start making predictions. 

To train AI to read mammograms, technicians input information from hundreds of thousands to millions of mammograms. The AI software creates a mathematical representation of what a normal mammogram looks like and what a mammogram with cancer looks like. The AI system checks each image against the standards to distinguish what’s normal from what’s not. As the program is exposed to more images of mammograms, it can learn over time (called machine learning) — and become more accurate, explained Amy K. Patel, MD, a breast radiologist and medical director of The Breast Care Center at Liberty Hospital in Liberty, MO. AI is also being used to spot cancer on ultrasounds and MRIs. 

AI-assisted breast cancer screening is available in some European countries, but is not yet standard in the U.S. According to the American College of Radiology Data Science Institute, about 9% of U.S. radiologists are using AI mammography or breast imaging. (Radiologists are doctors that use imaging to diagnose and treat diseases.)

There are many different ways radiologists can use AI when reading mammograms, including checking their reading against the computer’s reading, or turning to the computer for mammograms to prioritize first based on “likely suspicious” results. Research is ongoing in countries where there is more use of AI to check for signs of cancer, but the studies done so far suggest that AI could improve breast cancer detection in several ways:

According to the National Cancer Institute, screening mammograms miss about 20% of breast cancers. AI systems appear to have the ability to pick up very subtle signs of an early cancer that the human eye might miss. 

A study published in The Lancet Oncology describes how researchers used AI to help screen mammograms of more than 80,000 women in Sweden. Half of these women had their mammogram read by AI before it was looked at by a radiologist, while the other half had theirs read by two radiologists. The study revealed that the AI group had 20% more cancers detected than the radiologist-only group. 

Another study in Germany and the U.S. that used AI to look at nearly 1.2 million mammograms found that having a radiologist and AI system working together was 2.6% better at detecting breast cancer than a radiologist alone. The results were published in The Lancet Digital Health in July 2022. 

A false positive result occurs when the radiologist detects an abnormal finding on a mammogram that doesn't ultimately prove to be a cancer. Before cancer can be ruled out, however, doctors may need to order multiple follow-up tests, such as additional mammogram images, ultrasound and/or a biopsy, which can be emotionally and financially draining.

A study of more than 91,000 mammograms from women in the U.S. and the U.K. found that the use of an AI system lowered the rate of false positives by almost 6% in the U.S. and by 1.2% in the U.K. The findings appeared in Nature in 2020.

A breast biopsy allows your doctor to determine whether an area flagged during a mammogram or other breast imaging contains cancer or not. “About 80% of biopsies performed on areas of concern turn out to be benign [non-cancerous],” says Dr. Patel. AI systems may be able to reduce the number of unnecessary biopsies.  

For instance, an AI tool called iBRISK (intelligent-augmented breast cancer risk calculator) could accurately predict whether abnormal tissue flagged by doctors was more likely to be benign or cancerous, according to a study in Radiology: Artificial Intelligence. 

“[I]t is exciting to imagine a future where AI is used successfully to avoid unnecessary biopsies in thousands of women and men each year,” Elizabeth S. McDonald, MD, PhD, and Emily F. Conant, MD, wrote in a separate article about the iBRISK findings. “Although the lofty goal of successful AI integration might seem elusive, the potential benefit to our patients is worth our collective efforts to achieve optimal screening and diagnostic performance.” Dr. McDonald is an associate professor of radiology and co-director of the Penn Breast Cancer Translational Research Group and Dr. Conant is a professor of radiology and the vice chair of faculty development in the department of radiology at the University of Pennsylvania.

AI may also lead to improvements in doctors’ ability to predict those people at greatest risk of developing breast cancer. 

A study published in June 2023 found that AI was more accurate in predicting breast cancer risk than the Breast Cancer Surveillance Consortium (BCSC) risk model. The Breast Cancer Surveillance Consortium Risk Calculator estimates a woman’s 5-year risk of developing invasive breast cancer based on such factors as a woman’s age, her family history of the disease, race/ethnicity, breast density, and any history of benign breast biopsies.

Using screening images collected from 13,600 women who had normal mammograms, five AI systems generated risk scores for developing cancer over that five-year period. AI was more accurate than the BCSC model in predicting breast cancer; the best results were achieved when AI and the BCSC model were used together. The findings appeared in Radiology.

More clinical trials are needed to make sure AI systems are safe and reliable to use as a second or even third reader of mammograms, in addition to a radiologist. AI can only be as good as the information that is used to train the technology. If there is a lack of diversity in the data used to train AI, for instance, it is impossible to know if the AI systems will be accurate for all people. 

“Adoption has been slow because we want to get this right for patients, and we want to make sure we can implement these tools to best serve them and not just for secondary gain [by companies that make AI systems],” Dr. Patel says. (Dr. Patel serves as a medical advisor to Kheiron Medical, a maker of AI systems for cancer diagnosis.)

With regard to AI and breast cancer screening, future studies will help to determine whether AI systems can:

• Sponsor Message

  produce accurate results for women of all ages, body types, and ethnic backgrounds

• Sponsor Message

  recognize all forms of breast cancer

• Sponsor Message

  reliably reduce rates of false negatives (when cancer is missed) or false positives versus standard readings by radiologists only

For example, the Leeds Investigation of Breast-screening AI study, or LIBRA, is a British study enrolling nearly 7,000 women who will have their mammograms read by AI software as well as two radiologists. If all three readings agree, women will be given the all-clear. However, if any of the three disagree, the radiologists will decide whether a patient should be called back for more imaging. The goal is to figure out if AI can increase cancer detection rates, reduce unnecessary patient recalls, and ease workforce pressures. 

Dr. Patel notes that there are many other potential uses for AI-assisted imaging in breast cancer diagnosis and treatment, too, such as: 

• Sponsor Message

  informing next steps after the detection of certain benign breast conditions that might increase the risk of breast cancer 

• Sponsor Message

  predicting the risk of cancer being present in a patient who appears to have ductal carcinoma in situ (DCIS), Stage 0 cancer

• Sponsor Message

  assessing whether a breast cancer is responding to chemotherapy given before surgery

• Sponsor Message

  determining the risk of metastatic breast cancer, particularly to the lymph nodes

You may wish to discuss the following questions with your doctor or other members of your care team.

In the U.S. at least, AI-assisted breast imaging is not yet widely available, and there’s not an online directory you can search to find it. If you're interested in AI-assisted imaging, whether for screening or monitoring, ask your care team what is available in your area.

Yes, there are clinical trials of AI-assisted imaging in breast cancer, including mammography, ultrasound, and MRI. Right now, most of the trials tend to be concentrated outside the U.S. You can ask your doctor or care team if there are any studies available in your area. You also can search online directories such as ClinicalTrials.gov and BreastCancerTrials.org. 

No. The consensus is that AI systems will not replace radiologists; rather, it will help them make more accurate readings and manage workflow. 

“You have to have the radiologist in the driver’s seat, as there is a human element that AI will never have,” says Dr. Patel. “I do think AI is going to help radiologists be much more effective in the years to come.”