Clinical use of CAD in mammography
Michael J. Ulissey, MD, presented the results of a study of 12,860 women with a median age of 49 who underwent mammography during a refresher course Monday at the Radiological Society of North America meeting in Chicago. The researchers tracked the mode of detection, whether it was traditional radiology, computer-aided diagnosis, or both. The total recall rate was 7.7 percent. Without the computer, the recall rate was 6.5 percent. The increased detection rate was 19.5 percent and almost all of the additional detections were stage 0 or stage 1 cancers.
When computer-aided diagnosis will become the new standard of care is anyone's guess, Ulissey said. That will require more and more radiologists to use the technology and more research to replicate the data. "Computer algorithms are only going to get better and better," he said, so the industry is certainly on that path. Plus, Ulissey said computer-aided diagnosis is more likely to defend a radiologist than prosecute a radiologist, which could also drive the technology toward routine care.
Breast CAD in detail
Joseph Yo, Ph.D. presented the results of his study on computer-aided diagnosis in breast imaging. He found variability-the area of interest picked up some of the time but not all of the time-in 14 out of 50 cases, a rate of 28 percent. He found reproducibility, after digitizing and analyzing the 50 cases some 10 times, of 40 to 43 out of 50 cases each time. That represents a rate of 80 to 86 percent.
As for the accuracy of breast segmentation, Yo found the results perfect or acceptable in 97 percent of cases. Yo found that computer-aided diagnosis in breast imaging limited false positives, allowed for rapid reading, and resulted in 10 to 21 percent false negatives. All should help with the legal burden radiologists face since missed breast cancer is the number one source of medical malpractice suits.
"CAD is cutting-edge technology," Yo said. "Not long ago it was only a research tool."
In the future, Yo sees the industry embracing digital. That includes a transition to full field digital mammography. CAD analysis of CAD images reduces CAD variability, he said. There is also less time delay and no digitization error.
Radiologists also will incorporate known information, he said. That includes using ultrasound for screening and diagnosis. The use of MRI for breast imaging is increasing, he said, but four dimensional data sets are laborious to handle. Researchers are studying tomosynthesis but it requires 10,000 slices to diagnosis one cancer, Yo pointed out. Despite the research going on in these other methodologies, "CAD will become a crucial tool," Yo said.
CAD methods for lung cancer
CAD should help reduce the number of unnecessary biopsies, according to Kunio Doi, Ph.D. Although 15 to 40 percent of biopsies are malignant, 60 to 86 percent of biopsies are unnecessary.
At this point, however, there are several limitations to CAD's ability to reduce unnecessary biopsies. Those include segmentation, reproducibility, sensitivity, and specificity.
Doi pointed out that in 2000, there were 55 CAD papers presented at RSNA but 163 presented this year. "CAD is applicable to all modalities and all parts of the body," he said.
For lung cancer, CAD's purpose is to improve diagnosis, such as the accuracy and consistency and reduction in reading time.