Wu, Fangfang

Abstract Accurate estimation of mammographic breast density could aid in augmenting breast cancer risk assessment for women undergoing breast screening with full-field digital mammography (FFDM). Breast density can be estimated from FFDM and is most commonly assessed in the clinic by visual grading into one of the four categories defined by the American College of Radiology BI-RADS. However, BI-RADS density assessment is highly subjective and does not provide a quantitative, continuous measure of percent density (PD), which would allow for more refined risk stratification and assessment of density changes. Here, we introduce Deep-LIBRA, an artificial intelligence (AI) tool for fully-automated assessment of breast PD from FFDM images. Two key modules form the core of Deep-LIBRA: 1) an implementation of a modified U-Net architecture for breast segmentation and 2) a radiomic machine learning module that performs PD estimation within the segmented breast region. To develop and validate Deep-LIBRA, raw (i.e., "For Processing") FFDM images (Selenia Dimensions, Hologic Inc.) acquired at two breast cancer screening practices were retrospectively analyzed. For the breast segmentation module, we used a total of 12,100 FFDM studies from 2,200 individual women and a 90%-10% split-sample training-validation approach, using the Dice coefficient to evaluate the accuracy of Deep-LIBRA versus ground-truth manual breast segmentation. For the PD estimation module we used a total of 3,304 FFDM images from 1,652 individual women; manual PD scores obtained with the widely used Cumulus software were used as the "gold standard" in a three-fold cross-validation setting to assess the accuracy of Deep-LIBRA in PD estimation. PD estimates from Deep-LIBRA were also compared with breast density estimates from the commercially available Volpara software. Breast segmentation had a Dice coefficient of 95.31% when compared to ground-truth manual breast segmentation in the validation set. Deep-LIBRA average differences from ground-truth PD scores in the three cross-validation folds were 4.91%, 4.65%, and 4.22%, while Volpara had corresponding average differences of 6.20%, 6.01%, and 5.94%. Deep-LIBRA PD scores were also significantly different from Volpara PD (t-test p-value < 0.001 in all three folds). Preliminary evaluation results show that Deep-LIBRA is a promising AI approach for accurately assessing PD from FFDM images. Citation Format: Omid Haji Maghsoudi, Scott Christopher, Aimilia Gastounioti, Lauren Pantalone, Fang-Fang Wu, Eric A. Cohen, Winham Stacey, Emily F. Conant, Celine Vachon, Despina Kontos. Deep-LIBRA: An artificial intelligence approach for fully-automated assessment of breast density in digital mammography [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2600.

Abstract Gene expression profiling has identified molecular subtypes that classify invasive breast cancers into distinct categories that vary in their clinical behavior and response to treatment. These subtypes highlight the many possible biologically and clinically distinct types of breast cancer. Immunohistochemical (IHC) staining of tumor sections using antibody panels has been used as a surrogate for the molecular subtypes identified through gene expression profiling, resulting in a set of intrinsic molecular subtypes. Given the heterogeneity within breast cancer, one might expect that different risk factors influence specific subtypes of breast cancer through various etiologic pathways. We evaluated whether percent mammographic density (MD), the proportion of fibroglandular or white tissue on a mammogram and one of the strongest and most consistent risk factors for breast cancer, is associated equally with all intrinsic molecular subtypes. Data were pooled from six cohort or case-control studies including 3389 women with invasive breast cancer and 9860 without, who underwent screening mammography. Percent MD was assessed from digitized film-screen mammograms (for cases, at a median 4 years prior to diagnosis) using a computer-assisted threshold technique, and categorized as 0-10%, 11-25%, 26-50% and 51%+. Receptor status was abstracted from clinical pathology records and supplemented by IHC staining of tumor sections or microarrays. With estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor (HER2) information, we classified tumors into the intrinsic molecular subtypes: Luminal A ( ER+ and/or PR+ and HER2- and grade 1 or 2), Luminal B (ER+ and/or PR+ and HER2+ or Luminal A and grade 3), HER2 expressing (HER2+/ER-/PR-) and triple negative (ER-/PR-/HER2-). For cancers found to be triple negative (TN), we also stained for epidermal growth factor receptor (EGFR) and cytokeratins (CK) 5/6 to differentiate basal-like tumors (positive for EGFR and/or CK 5/6) from the unclassifieds (negative on both EGFR and CK 5/6). We used polytomous logistic regression to calculate the odds (OR) of each of the intrinsic subtypes of breast cancer by categories of percent MD, adjusting for age, BMI and study. Contrasts were constructed and tested within the polytomous model framework to investigate differences in the strength of association of percent MD across subtypes. On average, cases were slightly older (56.4 years [SD=11.3]) than controls (55.8 [10.9]). Of the 3389 invasive breast cancer cases, 2146 (63%) were classified as Luminal A, 709 (21%) as Luminal B, 169 (5%) as HER-2 expressing, and 365(11%) as triple negative (TN). Of the TN tumors, 203 had tumor tissue available and were evaluated for CK 5/6 and EGFR, with 167 (82%) classified as basal-like and 36 (18%) as unclassified. Percent (%) MD was associated with breast cancer risk across all intrinsic subtypes. For Luminal A, compared to women with 11-25% MD (reference), women with 0-10% MD had a reduced risk of breast cancer (OR=0.64 [95% confidence interval (CI): 0.55, 0.74]) while women with 26-50% MD had an OR=1.5 (1.3,1.7) and women with 51%+ MD had the highest risk, OR=2.4 (2.0, 2.8). Similar associations were seen for Luminal B, HER2 expressing, and TN across percent MD categories. The small differences among the percent MD-breast cancer associations observed for the different subtypes were not statistically significant, either when comparing the four (Luminal A, Luminal B, HER-2 expressing and TN; p=0.61) or five categories of subtypes (Luminal A, Luminal B, HER-2 expressing, basal-like, unclassified; p=0.66). Our results suggest percent MD is a risk factor for all intrinsic molecular subtypes. Additional study is needed to examine whether these results hold for younger and older aged women. Understanding the importance of mammographic density measures for subtypes of breast cancer has significance for development of subtype-specific risk models. Citation Format: Celine Vachon, Rulla Tamimi, Yunn-Yi Chen, Christopher Scott, Kimberly Bertrand, Matthew Jensen, Daniel Visscher, Aaron Norman, Fergus Couch, John Shepherd, Bo Fan, Fang-Fang Wu, Lin Ma, Andrew Beck, Steve Cummings, V. Shane Pankratz, Karla Kerlikowske. Mammographic density: A risk factor for all breast cancers or only specific subtypes? [abstract]. In: Proceedings of the Eighth AACR Conference on The Science of Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; Nov 13-16, 2015; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2016;25(3 Suppl):Abstract nr IA22.