Computer-aided detection inches closer to widespread acceptance

Specificity has long been the weak spot of computer-aided detection. Pattern-matching algorithms have traditionally marked everything suspicious, leading radiologists to complain that CAD systems did little more than redirect them to do a second read.

CAD technology has improved markedly, however, according to Dr. Carl J. Vyborny, an assistant professor of radiology at the University of Chicago. And improvements in specificity have not come at the expense of sensitivity. The best systems can now detect 90% of all malignancies, according to Vyborny, while producing as few as 0.5 marks per image.

“In 25% of the cases, CAD produces no marks at all,” he said during a May 4 presentation at the Society for Computer Applications in Radiology meeting in Salt Lake City. “And the chance that a cancer is present but the computer makes no mark is now only one in 10,000.”

Computer-based expertise may be the best hope for easing the burden being placed on radiologists. Their numbers are dwindling-800 radiologists a year are retiring, compared with a historical rate of 400-Vyborny said. With inherent limitations in human observation and perception, radiologists overlook approximately 33% of all breast cancers.

The radiology community is already beginning to avail itself of this technology. More than 200 ImageChecker units built by R2 Technologies have been installed at nearly 150 mammography sites. The FDA has approved the sale of R2’s technology bundled with GE’s Senographe 2000D full-field digital mammography system. And the agency is reviewing CAD technologies from other vendors, both in stand-alone and bundled configurations. To further encourage the adoption of CAD by mammographers, Congress last year passed a bill, signed by President Clinton in December, authorizing extra reimbursement for mammography that includes the use of a computer.

But before this technology can be part of mainstream practice, radiologists must buy into the idea of CAD. Legal liability could be a motivator, Vyborny said. Radiologists’ concerns about liability, especially pertaining to breast cancer, are high. More than any other

specialists, radiologists are the targets of litigation involving breast cancer. The average judgment is $300,000, and multimillion dollar judgments are not uncommon.

“Radiologists’ acceptance might be directly related to the likelihood that CAD can prevent them from being sued,” Vyborny said.

Missed cancer is a common reason for such litigation, but delayed diagnosis is another. Evidence suggests that radiologists are becoming increasingly vulnerable to the latter. Many mammography centers are swamped, forcing 8% of women seeking screening exams to wait an average of two months and 28% to wait between a week and a month. (SCAN 4/11/01). CAD might be the answer.

“Computers are different from humans and can give reliable, consistent assistance,” Vyborny said. “They never forget or regress, they don’t get tired, and they will only get better.”