CAD improves radiologists' ABUS performance

By Kate Madden Yee, staff writer

November 11, 2015 --

Monday, November 30 | 10:10 a.m.-10:20 a.m. | RC215-09 | Arie Crown Theater
In this study, a Dutch team compared unaided radiologist readings of automated breast ultrasound (ABUS) exams with those interpreted using computer-aided detection (CAD) to determine if adding CAD as a "second reader" could help reduce unnecessary recalls.

Screening for breast cancer with ABUS can be a useful adjunct to mammography, but it can also boost recalls for lesions that are ultimately proved to be benign.

The team led by Jan van Zelst of Radboud University Medical Center in Nijmegen included ABUS scans from 89 women in the study. Of these, 19 exams were malignant, 30 were benign, and 40 had no lesions at all. Three radiologists and one fourth-year resident read the exams without help from CAD, marking any findings based on the likelihood of malignancy using a 0 to 100 scale.

The 89 cases were then analyzed by CAD, and any suspicious findings were compared with the findings of the four readers. The research team considered the findings suspicious only when the same lesions were marked by the CAD program; any reader marks that were not also flagged by the CAD program were considered benign. Finally, the researchers used analysis based on the area under the receiver operator characteristics (ROC) curve to compare unaided readings with CAD readings.

The area under the ROC curve improved after the CAD program was used, from 0.77 to 0.88 (p = 0.01), the group found. Among the reader marks uncorroborated by CAD, none were malignant.

Using CAD may help clarify ambiguous findings on automated breast ultrasound, reducing unnecessary referrals for further imaging, van Zelst and colleagues concluded.