Computers read mammograms to detect breast cancer

By using computers programmed to recognize suspicious mammograms, doctors can find breast cancers that would otherwise escape diagnosis, say radiologists who are among the minority in their profession currently using the technique.

Stamatia Destounis of the University of Rochester in New York and her colleagues employed an X-ray–scanning computer to reanalyze old mammogram results of 318 women. Although doctors hadn’t originally read the mammograms as being abnormal, all the women in the study had been diagnosed with breast cancer at least 1 year after their mammograms were performed. In 52 cases, a pair of physicians who had initially read the results hadn’t noticed developing cancers that, in retrospect, were visible on the breast images.

The computer-assisted analyses correctly pegged 37 of these missed cancers, Destounis and her colleagues report in the August Radiology. However, the computer flagged as suspicious about twice that number of masses that turned out to be harmless. If doctors had acted on all the computer’s suggestions by performing follow-up biopsies, they would have caught more breast cancers, but they would also have performed more unneeded procedures.

Machines can’t replace doctors who read mammograms, but computer-aided detection, or CAD, can guide them, Destounis says.

In a second study, pairs of radiologists on Destounis’ team first performed 18,586 new mammograms and identified more than 400 that were suspicious enough to require biopsies. Of those, 85 proved to be cancer. That biopsy-to-cancer ratio is typical. The radiologists also consulted the computer and, on the basis of its recommendations, performed six additional biopsies. All six identified a cancer that would have been missed.

The expense of installing the CAD systems limits how widely they are used, Destounis notes. Nevertheless, she says, “I would recommend CAD for all mammograms.”