Association between Breast Parenchymal Complexity and False-Positive Recall From Digital Mammography Versus Breast Tomosynthesis: Preliminary Investigation in the ACRIN PA 4006 Trial.

RATIONALE AND
OBJECTIVES: We investigate associations between measures of mammographic parenchymal complexity and false-positive (FP) recall from screening with digital mammography (DM) versus digital breast tomosynthesis (DBT).
MATERIALS AND METHODS: We retrospectively analyzed data from 541 women recruited by the American College of Radiology Imaging Network 4006 trial, designed to evaluate callback and detection rates from screening with DM versus combined DM and DBT. Of these, 68 and 56 were FPs based on DM alone versus the combined DM/DBT readings, respectively. Mammographic complexity was quantified with computerized texture analysis and percent density. Logistic regression was performed to evaluate associations between extracted features and FP recall, after adjusting for age and number of previous benign biopsies. Odds ratios and area under the curve (AUC) of the receiver operating characteristic were used to assess association strength.
RESULTS: For DM, age, previous benign biopsies and texture features of correlation, inverse difference moment, sum average, and sum variance were deemed as significant predictors (P <.05) of FP recall, with an AUC = 0.77. For DBT, age was the only significant predictor of FP recall with AUC = 0.64. Using exploratory receiver operating characteristic thresholds for which no true-positives would be missed, a potential FP reduction of 23.5% and 8.9% was demonstrated, respectively, for DM alone versus DM/DBT.
CONCLUSION: Measures of breast complexity measured on 2D digital mammograms are indicative of the likelihood for FP recall from screening with DM, and could help identify women who could benefit from supplemental screening, including DBT.