CDS prototype for breast MR does not curb unnecessary biopsies

Diagnostic performance and false-positive biopsy recommendations did not improve among radiologists using a prototype clinical decision support (CDS) system designed for dynamic contrast-enhanced (DCE) breast MRI studies, according to a study published in this month's American Journal of Roentgenology.

As DCE-MRI begins to gain acceptance as a diagnostic and staging tool for breast cancer, the addition of CDS may aid radiologists’ interpretations of suspicious lesions and reduce unnecessary biopsies by providing readers with access to cases beyond their own experience.

Lilla Boroczky, PhD, of Philips Research North America in Briarcliff Manor, N.Y., and colleagues deployed a prototype CDS at the University of Chicago Medicine and sought to evaluate its diagnostic performance and impact on radiologists’ confidence.

The CDS system features two breast databases of DCE-MRI studies. The retrieval database contains 96 malignant and 96 benign lesions, and the test database consists of 46 malignant and 51 benign lesions that were primarily moderately difficult “indeterminate” cases.

Five breast imagers reviewed DCE-MRI exams and indicated their BI-RADS assessment of each lesion as well as their recommendation for biopsy or follow-up. They also provided a confidence rating on a 0 (no confidence) to 10 (certain) scale.

Next, they input a series of morphologic and kinetic features for the lesion into the CDS to retrieve similar cases.

The area under the receiver operating characteristic curve showed that observers’ sensitivity, specificity, accuracy, positive predictive value and negative predictive value did not change significantly with CDS.

Average confidence increased significantly for three of the five radiologists: from 5.9 to 6.3, from 7 to 7.2, and from 4.4 to 5.4, respectively. However, average confidence decreased for the other two radiologists.

“[T]he CDS is most useful for lesions with intermediate BI-RADS categories (3, 4a, 4b and 4c) and not as much for obvious benign and malignant cases,” wrote Boroczky et al.

The researchers acknowledged that the system’s lack of impact on unnecessary biopsies and diagnostic performance was unexpected, and attributed it to the composition of the database, which included benign-looking malignant cases and malignant-looking benign cases, as well the high sensitivity of DCE-MRI.

“We think that, for a radiologist to accept the recommendation of a CDS system, [the area under the receiver operating characteristic curve] would need to be significantly closer to 1. This has implications for further development of the current prototype or design of any CDS systems,” wrote the researchers.

However, despite the system's lack of impact on decision making, three of the five radiologists found the system useful, noted the researchers.