Soo-Yeon Kim, Presenter: Nothing to Disclose

Woo Kyung Moon MD, Abstract Co-Author: Nothing to Disclose

Wei-Chih Shen MS, Abstract Co-Author: Nothing to Disclose

Ruey-Feng Chang PHD, Abstract Co-Author: Nothing to Disclose

To quantify the US characteristics of breast masses defined in the ACR BI-RADS and to evaluate the accuracy of the CAD system for the classifications of benign and malignant breast tumors.

Digital US images of 265 pathology-proven cases including 180 benign and 85 malignant masses were used. Eight US features including shape, orientation, margin (angular and unduration characteristics), lesion boundary, echo pattern (the average gray intensity and the variation), and posterior acoustic features were computed and evaluated by the point-biserial correlation coefficient (r). A CAD system was constructed to integrate all proposed features for predicting the likelihood of malignancy by the binary logistic regression model.

On each proposed feature, the mean values of malignant tumors were significantly different from benign tumors (p< 0.05 for all eight features). The correlation between the features and pathological result showed the highest value in angular characteristic of margin (r=0.7) followed by undulation (r=0.56), shape (r=0.56), orientation (r=0.12), posterior acoustic features (r= -0.32), the average gray intensity (r= -0.35), the variation of the echoes (r= -0.46) and lesion boundary (r= -0.57). For the constructed CAD system, the performance indices, i.e., accuracy, sensitivity, specificity, PPV, and NPV were 91.7% (243 of 265), 90.6% (77 of 85), 92.2% (166 of 180), 84.6% (77 of 91), and 95.4% (166 of 174), respectively, with an Az value of 0.97.

Among the quantified BI-RADS US features, margin and shape of breast tumors were most important for predicting the likelihood of malignancy by CAD systems.

The computerized US features and the developed CAD systems confirm the value of BI-RADS lexicon for the classification of breast tumors.

Kim, S, Moon, W, Shen, W, Chang, R, Computer-aided Diagnosis (CAD) of Breast Masses Using American College of Radiology (ACR) Breast Imaging Reporting and Data System (BI-RADS)-Ultrasound (US) Features.  Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL. http://archive.rsna.org/2007/5012108.html