Radiologists' detection of mammographic abnormalities with and without a computer-aided detection system.

Abstract

The aim of this study was to to evaluate the role of a computer-aided program (CAD) in assisting detection of mammographic lesions by radiologists not specifically trained in mammography and its potential utility in breast screening. Mammograms were evaluated by radiologists not specifically trained in mammography first without, and then with, the CAD. Finally, the mammograms were evaluated by experienced mammographers who formed the reference standard. Two hundred and ninety four breasts were studied. In 257 breasts (87.4%), the CAD system did not help with the detection of abnormalities. It assisted radiologists in detecting abnormalities in 21 breasts (7.1%) with potential for detecting abnormalities in a further 13 breasts (4.4%). Only in three cases (1%) did it cause confusion in interpretation. There is overall increased sensitivity in detecting mammographic abnormalities with the aid of the CAD system from 74.4 to 87.2%, which is statistically significant. However, it failed to detect suspicious abnormalities in 71 breasts (24.1%). The CAD system improved detection of suspicious mammographic abnormalities by radiologists who are not specifically trained in mammography. However, there is also a substantial failure to detect suspicious mammographic features that cautions against over-reliance on the system, emphasizing its role as a second reader at best.