using mammography to detect cancer in normal breast tissue is fundamentally a signal-to-noise exercise, it is particularly suited to CAD technology.30 Because factors such as radiologist fatigue and distraction, the complexity of breast structure, and the subtle characteristics of early stage disease make interpreting mammograms challenging and contribute to both false-positive and false-negative results, the use of CAD with mammography becomes particularly attractive, offering experienced radiologists the option of a “double read.”
Basic CAD systems consist of a workstation with display and signal processing software. The CAD unit reads either manually digitized mammography films or directly digitized images and highlights areas of concern such as masses, calcifications, or architectural distortions, for the radiologist’s review. Images can be printed or displayed in soft copy on a monitor. CAD for mammography was formally introduced in 1998 when the FDA approved the first CAD device, ImageChecker M1000®, made by R2 Technology of Sunnyvale, California. In addition to ImageChecker, two other CAD devices cleared for use in the United States: (1) Second Look® by Nashua, New Jersey-based iCAD® and (2) MammoReader® by Intelligent Systems Software of Clearwater, Florida. ImageChecker and Second Look are also approved by the FDA for use with full-field digital mammography devices.
In a 2001 study, radiologists who interpreted mammograms, using both conventional mammography reading techniques as well as CAD technology, found nearly 20 percent more cancers with CAD than they did without, and the proportion of early stage malignancies detected increased from 73 to 77 percent. But they also found that the recall rate increased, from 6.5 percent when the radiologist interpreted the mammogram without CAD to 7.7 percent when CAD was used.46 This study analyzed only the ImageChecker M1000® system produced by R2 Technology and the results cannot be assumed to apply to every CAD system.
The reproducibility of CAD results has improved as the technology has been advanced. Bin Zheng and colleagues used 100 mammographic cases with four views each from a database of more than 1,000 digitized images and diagnostic results. The cases included 25 with microcalcification clusters and 75 with masses. Two-thirds of the cases had been confirmed malignant. Using ImageChecker®, Zheng scanned the images three times over a period of 3 weeks, checking for sensitivity, false-positive rates, and reproducibility of the results. The researchers found identical results in 213 of 400 images, for a reproducibility rate of 53 percent, an improvement from 38 percent found in a 2000 study based on an earlier version of the CAD system.125
The greatest clinical value in CAD probably does not lie in its ability to raise the performance level of all breast imagers, but rather in its potential