Training matters most with digital mammography

Digital mammography's replacement of screen-film has become inevitable now that head-to-head comparisons of the systems are showing little difference in performance. The sticking point now lies with the radiologists reading soft-copy mammograms, not with the technology itself, according to speakers at the ECR.

Participants in a special session on digital mammography presented data on the efficacy and accuracy of screen-film versus full-field digital mammography (FFDM) for lesion detection and characterization. They dismissed differences in sensitivity and specificity as statistically insignificant and targeted inter-reader variability as the prime cause of diagnostic vulnerability.

"We are the problem, and we have to learn," said session chair Prof. Rolf Schulz-Wendtland, a professor of radiology at Erlangen University Hospital in Germany.

The European Screen-Trial project, which is evaluating the feasibility of digital mammography in breast screening programs, is assessing the effect of training on radiologists' soft-copy reading performance in an ongoing three-part study.

A group of 19 radiologists experienced in reading hard-copy screening mammograms attended a one-day workshop in workstation operation and soft-copy reading skills before viewing 150 full-field digital mammograms (20% cancers) in both hard- and soft-copy form. A two-hour limit was imposed for reading 75 cases. Study participants were also asked to read 500 to 1000 FFDM cases (5% cancers) during a six-month self-training period. They then assessed the initial 150 FFDM cases again in soft-copy.

Results from the nine radiologists who have completed the study reveal no significant change in either sensitivity of specificity before and after training, said Nora Wedekind of the Center for Medical Diagnostic Systems and Visualization at University of Bremen, Germany, who presented the data at the ECR. Radiologists' reading speed generally increased, however, to become comparable with their soft-copy reading rate.

"The reading performance of this group of experienced radiologists was already very high, so the room for improvement by training was limited," Wedekind said. "Also, this study did not involve computer-aided detection software. We would expect an improvement in reading speed and in performance with a combination of training and CAD, and we will be performing another study to test this."