Can computers spot breast cancers that radiologists miss? That's the question doctors with Intermountain Health Care are hoping to answer once and for all in a yearlong study of 40,000 women.
Mammogram technology has improved in the past decade, providing sharper images of breast tissue, but mammography still misses 10 to 30 percent of breast cancers, says Dr. Brent Parkinson, director of breast imaging services for IHC. That's why doctors have turned to computer aided detection equipment for a second opinion.
But previous studies of CAD and mammography have proved inconclusive. One large study, published in the Journal of Radiology in 2001, found that the CAD technology detected one additional breast cancer case per 1,000 screening mammograms, or 12 additional cases in the 12,000 women studied. But another, larger study sponsored by the National Cancer Institute found that the computers were no more skillful than radiologists in finding irregularities that turned out to be cancerous.
The IHC study, funded by a $21,500 grant from the LDS Hospital Deseret Foundation, will avoid the weaknesses of the previous studies, says Parkinson, the study's principal investigator. It will provide an additional 40,000 women for a control group, and will also include not just breast radiology specialists but a mix of radiologists.
"What happens in our practice may be more representative of the whole country," he said.
During the study, which began on Sept. 27, each mammogram will be read by both a human radiologist and a computer, programmed to have the same kind of pattern recognition the radiologist gets after years of practice.
If the CAD finds something amiss — a suspicious area that might reveal a cancerous mass, or a white dot that might signify calcification — it marks the spot on a digitized version of the mammogram. The radiologist then rereads the original mammogram to see if the computer has picked up something missed by the human eye.
Radiologists view traditional mammogram films by looking through a contraption that both magnifies the image and blocks out distracting light. Like an astronomer gazing at the heavens through a telescope, the radiologist tries to sort through the shades of dark and light to find something that might be significant.
For a radiologist like Parkinson, who has scanned hundreds of thousands of mammogram images in his career, abnormalities stand out. Still, it's always possible to miss something, which is why the computer assist could be a boon.
"If it picks up more cancers earlier, we know it will save more lives," Parkinson said. It could also save money, because treating a breast cancer early rather than at a later stage is less expensive.
The CAD technology is sometimes more adroit than radiologists at spotting subtle calcifications, Parkinson says, but most of these turn out, on a skilled radiologist's second reading, to be nothing to worry about. CAD is less successful than radiologists at spotting masses.
"What we know is, you don't rely on this alone to read a mammogram," he said.
The IHC study will track every person who has an abnormal mammogram, and will tabulate how many of these irregularities were caught by radiologists, by the computer or both. Then it will follow each patient to see how many people with abnormal mammograms turned out to have breast cancer.
The CAD study will be conducted on mammograms taken at LDS, Cottonwood, Alta View and McKay-Dee hospitals, plus the Bryner Clinic.