For Mammogram's Meaning, Radiologists Reign Over Computer

SEATTLE, Dec. 5 -- When it comes to mammogram interpretation, the eyes and intuition of radiologists still have it over computer chips, according to researchers here.

Computers can help interpret mammograms, but radiologists should trust their own judgment first, concluded Stephen Taplin, M.D., of the Group Health Cooperative Center for Health Studies here, and colleagues.

That's the bottom line of their retrospective study, reported in the December issue of the American Journal of Roentgenology, that evaluated how radiologists interpreted mammograms with and without the benefit of computer-assisted detection software.

Overall, the computer-assisted detection program slightly increased physicians' ability to determine that a woman without cancer was, in fact, cancer-free. But it had no effect on their ability to detect cancer, or potential cancer, that was present.

But when the computer-assisted detection program did not mark a visible lesion as suspicious, doctors were less likely to recommend further evaluation than they were when they didn't use the program, the researchers found.

"This means that the radiologists may have been deferring to computer-assisted detection and believing its interpretation rather than their own interpretation," Dr. Taplin, now at the National Cancer Institute, said.

"This study shows that it is hard to ignore the technology," he said, "and it raises the question of whether there is a potential for computer-assisted detection to do harm."

The researchers began with more than 56,000 screening mammograms taken from 1996 through 1998 and identified cases of breast cancer diagnosed within two years after the mammograms were taken.

From those, they selected a set of 341 mammograms from three different groups - 114 women who were cancer-free for two years after their mammograms, 114 who developed breast cancer within a year, and 113 who developed breast cancer within 13 months through two years.

The set was used to test the performance of 19 radiologists, each of whom read the mammograms with and without computer-assisted detection during test sessions six months apart. Dr. Taplin and colleagues compared the results of the two approaches.

The study found that computer-assisted detection assistance did not affect overall sensitivity. For instance, without the use of computer-assisted detection the radiologists identified 63.2% of the cancers that appeared within a year and with computer-assisted detection they identified 62%.

But the effect differed for visible masses that were marked by computer-assisted detection, compared with those that were unmarked, the researchers found. If the masses were marked, sensitivity was increased by the use of computer-assisted detection, but was decreased if they were unmarked.

For instance, the sensitivity for marked cancer by one year was 82.7% without computer-assisted detection and 83.1% with computer assistance. But if the lesion wasn't marked, the doctors identified 37.4% of the cancers without computer-assisted detection, but only 30.1% when they were aided by computer.

A similar pattern was seen for cancer that developed from 13 months through two years and both interactions between assistance and computer-assisted detection marking were statistically significant at P