Computer-Aided Detection May Not Improve Breast Cancer Screening
Computer-aided detection doesn't seem to improve screening mammogram accuracy; doesn't mean mammograms aren't valuable.
A large study suggests that using a computer to help read screening mammograms -- called computer-aided detection (CAD) -- doesn’t improve the accuracy of screening mammograms.
It’s important to know that this doesn’t mean screening mammograms are less valuable. It simply means that using a computer to read mammograms doesn’t add anything more to the interpretation than what a skilled radiologist can see.
The study was published online on Sept. 28, 2015 by JAMA Internal Medicine. Read the abstract of “Diagnostic Accuracy of Digital Screening Mammography With and Without Computer-Aided Detection.”
With CAD, a computer program highlights areas on the mammogram image that MAY be abnormal. A radiologist has to decide if the areas really are abnormal. CAD isn’t the same as having a second radiologist look at the mammogram. With CAD, only one radiologist looks at the mammogram.
After the U.S. Food and Drug Administration approved CAD for mammography in 1998, many mammogram centers began using the tool. Research shows that by 2012, 83% of all screening mammograms were digital and read with CAD. Reading a mammogram with CAD costs more than having only a radiologist read the mammogram.
In this study, the researchers wanted to know if using CAD improved the accuracy of screening digital mammogram readings. So they compared 495,818 digital mammograms read with CAD to 129,807 digital mammograms read without CAD.
The mammograms were done in 323,973 women from 2003 to 2009 and read by 271 radiologists at 66 facilities across the United States.
The researchers found that CAD didn’t improve any measure of accuracy of screening mammograms:
- how often cancers were detected
- how often cancers were missed
- how often an abnormal area that wasn’t cancer was incorrectly labeled as cancer (a false positive)
The radiologists detected cancer in about four of every 1,000 women whether or not they used CAD. About three of those four breast cancers were invasive.
Radiologists using CAD did find more DCIS, which is non-invasive, stage 0 cancer, than radiologists not using CAD.
"Even more troubling, when we studied the 107 radiologists who interpreted [mammograms] both with and without CAD, we found that a given radiologist tended to miss more cancers when using CAD than when he or she didn't use the software," said the study’s lead author, Constance Lehman, M.D., Ph.D. "It may be that radiologists reading with CAD are overly dependent on the technology and ignore suspicious lesions if they are not marked by CAD." Dr. Lehman is the director of breast imaging and co-director of the Avon Comprehensive Breast Evaluation Center at Massachusetts General Hospital in Boston.
The researchers said they were concerned that reading screening digital mammograms with CAD offers no additional benefits to women and may possibly increase the risk that a radiologist misses a breast cancer.
The researchers are also concerned about the cost of CAD. CAD reading of mammograms is estimated to cost $400 million per year in the United States, though some doctors think this estimate is low.
“Despite lack of evidence of benefit, use of CAD mammography has taken off," said Diana Buist, Ph.D., a senior investigator at the Group Health Research Institute and another study author. “Evidence has accumulated demonstrating no added benefit and potentially more harm from this technology. Yet insurance companies, employers, and women are still paying for CAD."
Still, some radiologists have concerns about the study.
“CAD was first developed for older film screen mammography systems and then adopted for digital systems,” said Emily Conant, M.D., chief of breast imaging at the University of Pennsylvania Medical Center and member of the Breastcancer.org Professional Advisory Board. “It may certainly be true that the previously documented clinical impact of CAD is not as great with present day digital mammography. However, there are issues with this paper. While a large number of studies across many practices were investigated, there is always great variability in the performance of the actual readers or radiologists that interpreted each mammogram and the authors were unable to account for this in their study design. Nevertheless, the study suggests that for 2D digital mammography, CAD may not be as useful as previously thought.
“I would be cautious in extending this analysis to digital breast tomosynthesis [3D mammography] and the new synthetic 2D mammograms that are now being used clinically,” Dr. Conant continued. “CAD in these scenarios may actually help with the accuracy and efficiency of reads by localizing lesions in 3D space and by summing information from the 3D images that might not be detected by the radiologist. Certainly, CAD applications for the rapidly evolving 3D tomosynthesis platform needs to be thoroughly evaluated.”
We know that mammograms aren’t perfect, but they’re the best way to find breast cancer early, when it’s most treatable. Just because CAD doesn’t add any extra value to mammograms doesn’t mean mammograms aren’t valuable for women. As you and your doctor develop your breast cancer screening plan, you may want to ask these questions:
- Are any of the mammography centers near me better than the others?
- Is digital mammography available? (Research has shown that digital mammography can be more accurate than film-based mammography.)
- Will the radiologist read my mammogram using CAD?
- Is my most recent mammogram being compared to my older mammograms when it’s being read? (Comparing the new to the old has been shown to improve reading accuracy.)
- Does a second radiologist routinely review any suspicious mammograms before a final interpretation is made? (Second readings improve mammogram accuracy.)
In the Breastcancer.org Screening and Testing section, you can learn more about mammograms and other ways to screen for breast cancer.
— Last updated on February 22, 2022, 9:55 PM
Share your feedback
Help us learn how we can improve our research news coverage.
Was this article helpful?