Computer-aided detection in mammography is achieving good grades. Researchers continue to affirm CAD's sensitivity for detecting breast masses and microcalcifications, and they have recently determined that it yields reproducible results. But its
Computer-aided detection in mammography is achieving good grades. Researchers continue to affirm CAD's sensitivity for detecting breast masses and microcalcifications, and they have recently determined that it yields reproducible results. But its Achilles heel might be its struggle to detect architectural distortion.
"Architectural distortion is a less common appearance for breast cancer, but it is more frequently overlooked, and CAD systems may be of less help for this particular finding," said Dr. Jay A. Baker, chief of breast imaging at Duke University Medical Center.
Baker and colleagues evaluated CAD's sensitivity to detect architectural distortion, the third most common mammographic appearance of nonpalpable breast cancer (AJR 2003;181:1083-1088). They used two commercially available CAD systems to examine screening mammograms from 43 patients who had 45 mammographically detected areas of architectural distortion. Five breast radiologists identified present and actionable architectural distortion in 80 views of the 45 cases.
One CAD system detected distortion in 22 of 45 cases and in 30 of 80 mammograms, yielding a 49% case sensitivity and a 38% image sensitivity, respectively. The other system featured a case sensitivity of 33% and an image sensitivity of 21%. The study, according to Baker, does not indicate that one system is better than the other.
In another study, Dr. Rachel F. Brem, director of breast imaging and intervention at George Washington University Medical Center, and her colleagues found that the use of CAD could raise radiologists' detection sensitivity by as much as 21% (AJR 2003;181:687-693). The researchers identified 377 potentially missed cancers. Three radiologists evaluated the mammograms, which had been interpreted as normal or benign nine to 24 months before cancer diagnosis. To enhance blinding, the radiologists were told that normal cases would be mixed in. In 313 cases, at least one of the three radiologists recommended additional workup. In these cases, the regions identified for workup corresponded to the location of the pathologically proven cancer in 177 cases, which were then evaluated by CAD.
The investigators determined that 123 additional cancer cases could have been found. The sensitivity of radiologists using CAD was 91.4%, an increase of 21.2% over radiologists not using CAD. While this study, among others, has validated the parameters of sensitivity and accuracy, additional research testing CAD's reproducibility has found that it is attainable using commercially available technology.
Dr. Bin Zheng, a research associate professor of radiology at the University of Pittsburgh, and colleagues scanned and analyzed 400 positive mammographic images three times, using CAD (Radiology 2003;228:58-62). Abnormality-based sensitivity for mass detection ranged from 66.7% to 70.8% for the three different scans. These values reflect the percentage of correct markings of at least one true-positive region in either the craniocaudal or mediolateral oblique views (or both). Region-based sensitivity for masses ranged from 51% to 52.6%, and the range for microcalcifications was 85% to 87%.
Reproducibility for the true-positive regions identified on all three CAD scans was substantially higher than for false-positive regions. CAD generated 188 cues in three scans for the true-positive regions, 82 of which were marked at the same locations. True-positive cluster regions elicited 88.9% same-location cues from CAD.