From pharmacy robots to space-age telemedicine systems, computerized helpers like those so charmingly portrayed in the pop movie classic "Star Wars" are slowly making their way into hospitals.
One of the most intriguing and futuristic of the recent lot is an automated double-checker for mammograms from the aptly named R2 Technology of Los Gatos, Calif., which was approved for marketing by the Food and Drug Administration late last month.
R2 Technology's ImageChecker is a computerized system that analyzes screening mammograms for suspicious areas that could spell cancer.
Although not designed to replace humans, the computerized system is the first product to so directly support the clinical judgment of doctors interpreting diagnostic images. Down the road, other expert systems are expected to be similarly applied to chest X-rays and CT scans. Meanwhile, expert systems also are working their way as diagnostic supplements into other fields, such as Pap smear screening to identify cervical cancer.
Each year more than 25 million mammograms are taken, according to the American College of Radiology. And although precise figures are hard to come by, the vast majority of those mammograms are used to screen apparently healthy women for breast cancer that can be treated at an early stage.
But mammograms, like Pap smears, are tricky to read and sometimes-especially for very early cancers-a radiologist can miss a shadowy lump or a small cluster of bright white spots that signal the disease. These spots are nearly microscopic bits of calcium that often but not always indicate cancer. But the R2 Technology system, which is intended to look over radiologists' shoulders, is particularly relentless in searching them out.
According to R2 Technology, its system would help doctors find nearly 13,000 cancers that are missed for every 100,000 breast cancers that are currently found during screening mammograms. "We all know we miss things, and this would help us avoid that," says Brian Garra, M.D., a radiologist at Fletcher Allen Health Care in Burlington, Vt., and a member of an FDA panel of experts that reviewed the device.
To do its job, ImageChecker relies on a sophisticated computer system called a neural network, which works like an artificial brain to analyze mammograms that have been scanned and translated into numerical code. The company trained the neural network with more than 400 mammograms of breasts that were later confirmed as cancerous to teach the ImageChecker how to look for malignancies. In clinical tests, the ImageChecker picked up about 83% of the cancers found by radiologists and an estimated 85% of those that humans overlooked on the first go-around. According to clinical studies, it will detect 98% of microcalcifications and 75% of masses.
"It will not pick up all cancers," cautions Garra, emphasizing that the system can't replace a radiologist's diagnosis. But the computerized system could greatly reduce or even eliminate the need for a second radiologist to read a mammogram, a common and rather expensive method for minimizing clinical errors. Institutions have different policies for the proportion of mammograms that get a second look, but some radiologists will ask for another read of every mammogram if given a chance.
During its review of the ImageChecker product, the FDA worried that the technology might inflate the cost of cancer screening by sounding too many false alarms, Garra says. False positives could cause patients to return for extra mammograms or biopsies. But although false alarms were raised fairly frequently with the technology, the number that actually required second mammograms or biopsies was too small to be significant, Garra says.
That was the case at Kaiser Foundation Hospital in San Francisco, one of 13 medical centers where the device was tested. "Our callback rates did not increase," says Sherry Butler, M.D., assistant chief of radiology there. She says the device draws a radiologist's attention to a questionable area. Then it's usually a very quick and easy decision based on human "experience and reasoning ability" to decide whether there is truly cancer or not, she says.
"It's a spell-checker for images," explains Alan Stein, vice president for regulatory affairs at R2 Technology. Like a word processor catching a word that seems odd, the ImageChecker only flags an area for review. A physician decides whether the marked region is cancerous or needs further testing.
But the price of mammography progress is not cheap.
The sticker price for the ImageChecker is $175,000-about the annual salary for a radiologist in some parts of the country.
At that price the ImageChecker probably makes sense only for improving the efficiency of big mammography centers, says Andrew Rosenson, M.D., medical director for UtiliMed, a radiology utilization consulting firm in Northbrook, Ill. Radiologists freed from quality checks could be put to use reading more mammograms at busy centers, he says, but smaller mammography operations probably won't be able to justify the cost. Even large centers, though, will need to crunch the numbers hard before deciding to buy, he says, because a judicious use of second readings by radiologists on staff might yield similar results.
Despite the ImageChecker's high price and current technical limitations, doctors say the time has come to start using automated computer systems to ease some of the burden of radiology image interpretation. "It's a technique that has a lot of potential," says Stephen Feig, M.D., a mammography expert at Thomas Jefferson University Hospital in Philadelphia. "I'm hoping the company will lend us one to evaluate it."