April 16, 2007

Computers Don't Help Mammogram Analysis

Photo of female patient getting a mammogram Woman undergoing a mammogram. National Cancer Institute.

An increasingly used computer software system designed to help doctors spot evidence of breast cancer instead appears to reduce the accuracy of mammogram readings and may lead to additional and unnecessary medical testing, according to a recent study.

Mammograms, or X-rays of the breasts, are often the first line of defense in detecting tumors that are too small to feel. By looking at mammograms, trained radiologists can identify suspicious regions that should undergo further testing. But visual inspection is subject to human error. So in the late 1980s, researchers began to develop computer software to help radiologists identify mammogram abnormalities they may have missed.

Computer-aided detection systems received approval from the U.S. Food and Drug Administration in 1998 and have since become more widespread and popular. By some estimates, about one-third of mammograms are now interpreted with assistance from these devices. But no large-scale studies had yet been done to assess their accuracy.

A multi-site team of researchers conducted the first large-scale community-based review of computer-aided detection in mammography with funding from the NIH National Cancer Institute (NCI), the Agency for Healthcare Research and Quality and the American Cancer Society.

Led by Dr. Joshua J. Fenton of the University of California Davis Health System, the scientists looked at the use of screening mammography in 222,135 women who had 429,345 mammograms. The mammograms were obtained at 43 facilities in three states from 1998 through 2002. During that time, 15% of the study sites began to use computer-aided detection as a complement to traditional mammograms.

As reported in the April 5, 2007, issue of the New England Journal of Medicine, the researchers found that the computer software had no clear impact on early identification of breast cancer. Instead, the devices resulted in significantly lower overall accuracy of mammogram readings. The authors estimated that for every additional woman diagnosed with breast cancer through the use of computer analysis, 156 women would be falsely recalled for more tests and 14 have unnecessary biopsies to exclude the possibility of cancer. The study suggests that the software may help to spot the least dangerous breast cancers, which may never develop into more serious cancers.

Despite the disappointing performance of the computer software, the researchers note that their study size was not large enough to judge with certainty whether the benefits of routine use of computer-aided detection could outweigh its harms. In addition, several research teams are working to improve the cancer-detecting software, which may ultimately enhance the accuracy of the devices.

Related Links