Deploying an advanced visualization technology, such as a computer-aided detection (CAD) system for digital mammography, and providing training solely on the operation is no guarantee that the software will be used effectively. A multi-center Japanese research team has found that for the most effective use of CAD for mammographic interpretation, readers must have both appropriate interpretation ability and an understanding of usage.

“To establish CAD’s effective usage, all prospective users must receive training in both the power and limitations of CAD,” the authors wrote. Their findings were presented via a scientific poster presentation at the 2008 International Workshop on Digital Mammography in Tucson, Ariz., last month.

They noted that the impetus for their study was an increase in the breast cancer mortality rate in Japan and the hope that CAD will be effective in preventing overlooks and improving reading results.

The team selected 100 randomized cases, including 40 that had been pathologically diagnosed as cancerous, from 5,788 mammography exams conducted at the National Cancer Center Hospital East over a one-year period. All exams were conducted on a Hologic Lorad M-IV screen-film mammography unit and a Fujifilm Medical System FCR Profect CS digital x-ray system CR plate reader.

Images were transferred directly from the FCR Profect CS to the CAD system, which was developed jointly by the Tokyo University of Agriculture and Technology, Fujifilm, and the National Cancer Center Hospital East. Images were interpreted on a 5-megapixel LCD.

Twelve qualified mammography interpreters participated in the reading test of the 100 exams after a CAD instruction session and an interpretation exercise using cancer cases. They first read all cases without the CAD system, then took a 30-minute break and interpreted the cases with the assistance of CAD.

For the mammography interpretation session without the use of CAD, the readers demonstrated an overall sensitivity of 89 percent and a specificity of 76.1 percent. With CAD utilization, the readers showed a sensitivity of 90.4 percent and a sensitivity of 74.4 percent.

“In the results for all readers, CAD improved sensitivity but specificity decrease,” the authors observed. “This suggests that CAD can contribute to prevent cancer from being overlooked, but perhaps at the expense of an increased recall rate.”

The team also broke out the readers into two groups on the basis of their interpretation performance. Those who scored a true-positive rate of 92.5 percent or higher, which indicated a high skill level in mammographic interpretation, formed one subset while the rest of the readers comprised the other subset.

The more skilled group demonstrated a sensitivity of 93.5 percent both with and without CAD. However, the less-skilled group demonstrated a statistically significant increase in sensitivity with CAD, 88.2 percent compared with 85.7 percent without the technology.

“From this result,” the authors noted, “in order to utilize CAD most effectively for interpretation of mammography, readers must have both an appropriate interpretation ability and an understanding of how to use CAD.”