How accurate is the computer-aided mammogram?
Computer-aided detection software was supposed to further help radiologists interpret mammograms. Instead, according to new research led in part by a UC Davis research physician, that software may be making readings less accurate.
The study, which included work done by Joshua Fenton, a family and community medicine physician at UC Davis, shows that women who got screening mammograms at centers using the detection software known as CAD were more likely to be told their mammograms were abnormal and thus to undergo biopsies to rule out breast cancer.
CAD software analyzes the mammogram image and marks suspicious areas for radiologists to review, to assist them in determining which images may point to invasive tumors. CAD was approved by the U.S. Food and Drug Administration in 1998 and has been incorporated into many mammography imaging practices. Its effect on the accuracy of interpretation, however, has been unclear.
“Within three years of federal approval, 10 percent of the mammography facilities in the country were using CAD,” says Fenton, who is an assistant professor and was lead researcher on the investigation. “There had been no large-scale, community-based review of CAD efficacy despite the rapid adoption of this technology, so we did this study to see if CAD was proving to be beneficial.”
Fenton and his colleague say the study points out the need for the use of other techniques to find cancer at its earliest stages. They note that the National Cancer Institute is incorporating techniques for imaging at the molecular level into many of its studies and is also conducting studies on how to improve the use of CAD and conventional mammography.
Using Data from 429,000 Mammograms
— Joshua Fenton, lead mammography study researcher
Investigators in the CAD study looked at the use of screening mammography in 222,135 women who had 429,345 mammograms. The period of observation was from 1998 through 2002 and took place at 43 facilities in Colorado, New Hampshire, and Washington states. The study included 2,351 women who received a diagnosis of breast cancer within one year after screening and also received a mammogram that did or did not use CAD.Seven facilities, representing 16 percent of the study sites, implemented computer-aided detection during the study period. With the use of CAD, 32 percent more women were recalled for more tests and 20 percent more women had a breast biopsy. Use of the software had no clear impact on the early detection of breast cancer. The study suggests that, if anything, the software may promote the detection of the least dangerous breast cancers, such as localized, in situ breast cancers. The effect of in situ cancers on breast cancer mortality remains unknown and some evidence suggests that not all develop into invasive cancers.
A UC Davis radiologist looks over a computer-aided mammogram.
Every time the CAD software marks a real cancer, a radiologist has to consider about 2,000 additional false-positive marks, making it very difficult to distinguish between real cancers and those that are not cancer. The authors estimate that for every additional woman diagnosed with breast cancer on the basis of CAD, 156 women are falsely recalled for more tests and 14 had unnecessary biopsies to exclude cancer.
Proliferation of Mammography Software
“It's unfortunate that the use of the software has proliferated so widely before we are certain of its benefits,” said Fenton. “We need studies to determine if the benefits of the software outweigh its harms and costs. There is also the potential for new studies to improve the performance of CAD software.”
The authors estimate that if all mammography facilities adopt CAD, the annual cost of mammograms in the United States could increase 18 percent, or an additional $550 million nationwide
Fentons's investigation, “Influence of Computer-Aided Detection on Performance of Screening Mammography,” was published in the April 5, 2007 issue of the New England Journal of Medicine.