Shonket Ray PhD, Presenter: Nothing to Disclose

Brad M. Keller PhD, Abstract Co-Author: Nothing to Disclose

Jae Young Choi DPhil, Abstract Co-Author: Nothing to Disclose

Jinbo Chen PhD, Abstract Co-Author: Nothing to Disclose

Emily F. Conant MD, Abstract Co-Author: Scientific Advisory Board, Hologic, Inc

Despina Kontos PhD, Abstract Co-Author: Nothing to Disclose

To investigate the feasibility to predict risk of false-positive recall from breast cancer screening with digital mammography based on computer-extracted parenchymal pattern features of breast tissue complexity.

Digital mammography (DM) images from the ACRIN 4006 trial were retrospectively analyzed. The trial was a reader study to compare screening call-back rates from 2D DM versus a combination of 2D/digital breast tomosynthesis (DBT) in an enriched cohort of women. From a total of 550 women imaged, 76 were recalled on the basis of DM alone, from which 11 were true-positives. Images were acquired using a full-field digital mammography (FFDM) unit. All DM images sets consisted of bilateral CC and MLO views and were vendor post-processed (“For Presentation”, Selenia Hologic Inc.). To characterize breast tissue complexity, breast percent density (PD) was estimated on a per-woman basis using previously validated automated software. In addition, thirteen texture features were extracted using a locally adaptive computerized parenchymal texture analysis algorithm. Logistic regression was performed to identify significant predictors of overall recall and false-positive recall respectively, adjusting for age and number of previous benign biopsies. The area under the curve (AUC) of the receiver operating characteristic (ROC) was used to evaluate model performance.

The logistic regression model has AUC=0.75 (95% CI 0.69-0.81) for predicting overall recall from DM and AUC=0.94 (95% CI 0.87- 0.99) for predicting risk of false-positive recall; outperforming prediction based on age and number of previous benign biopsies alone that have AUC=0.64 (95% CI 0.57- 0.70) and AUC=0.73 (95% CI 0.51- 0.94) respectively. Significant predictors (p<0.05) are energy, inertia, inverse difference moment, sum average, sum variance, difference average, difference variance and difference entropy. Sensitivity for predicting false-positive recalls is 80% at a 100% cancer detection ROC operating point.

Prediction of false-positive recall from DM screening mammography could be improved with the inclusion of computer-extracted features of breast tissue complexity.

Prediction models could identify women at high-risk for false-positive DM screening due to their breast tissue complexity, who may be offered supplemental modalities for breast cancer screening.

Ray, S, Keller, B, Choi, J, Chen, J, Conant, E, Kontos, D, Prediction of False-positive Recall from Screening Mammography Using Computer-extracted Breast Tissue Complexity Features: Data from the ACRIN 4006 trial.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14011487.html