Abstract PS3-04-07: Anisotropic active dense tissue: a strong image-based cancer risk predictor

Abstract Introduction: Mammography-based risk prediction models show promise, with top AI models achieving AUCs near 0.801, potentially outperforming traditional clinical models (TC8, BCSC, BRCAPRO, Gail) whose AUCs are ∼0.60-0.652. However, AI models face limitations in widespread use due to issues with generalizability, calibration, and interpretability. Building on the understanding that tumor microenvironment changes promote breast tumorigenesis3,4 and tissue organization influences tumor invasion5, our patented interpretable computational biophysics approach6,7 separates mammographic dense tissue into "active" (structurally reorganizing) and "passive" subtypes8,9. We found that active dense tissue associates with early tumor onset10, that our method is agnostic to processing vs. presentation mammograms11, and our measurements synergize with volumetric breast density (VBD) and fibro-glandular volume (FGV) for a 3-yr risk AUC of 0.82 [under review]. We present a retrospective case-control analysis demonstrating how a novel morphological metric, called anisotropic mass, augments the predictive power of our active dense tissue measurement. Methods: Using the OPTIMAM database12,13, bilateral mammographic screenings were acquired on Hologic’s Selenia units ∼3 years prior to diagnosis for n=197 cases (mean age = 59.4), or the last screening on record for n=3712 controls (mean age = 59.8). We calculated areas and anisotropic mass of fatty, active dense, and passive dense tissues. Anisotropy measures directional preference by quantifying departure from uniform directional gradients14 and anisotropic mass is the sum of anisotropy across an area. VBD and FGV were obtained from Volpara® (Lunit Inc.) software. An interaction generalized linear model (GLM) predicted cancer vs. control status using VBD, FGV, patient age, amounts of fatty, passive dense, and active dense tissue, our mammographic percent density, and active dense tissue's anisotropic mass. AUC and 95% confidence intervals were calculated from bootstrapped ROC curves.The images and data used in this publication are derived from the OPTIMAM imaging database ( https://medphys.royalsurrey.nhs.uk/omidb/about-omi-db/ ) 12,13 . We acknowledge the OPTIMAM project team and staff at the Royal Surrey NHS Foundation Trust who developed the OPTIMAM database, and Cancer Research UK who funded the creation and maintenance of the database. Results: Our model outputs an AUC of 0.85, with (0.84-0.86 95% CI), a very significant improvement (p-value < 10-7) over our prior model (AUC = 0.82), validating anisotropy as a complementary measure to active dense tissue. Conclusions: Our biophysics-based approach offers a significant advantage in breast cancer risk prediction. It generates interpretable measurements that match or exceed deep learning models' performance while overcoming their drawbacks in generalizability, calibration, and interpretability. References: 1.IBDW 2025, https://breastdensityworkshop.org/wp-content/uploads/2025/06/IBDW2025_program_online_6-2-2025.pdf.2. McCarthy et al. J Natl Cancer Inst, 2020. 112(5): p. 489-497.3. Bissel et al. Nat Med, 2011. 17(3): p. 320-9 4. Tanner et al. Proc Natl Acad Sci USA, 2012. 109(6): p. 1973-8 5. Nazari et al. Breast Cancer, 2018. 25(3): p. 259-267 6. Khalil & Batchelder U.S. Patent 10,467,755 B2, 2019 7. Khalil & Batchelder EU Patent 2,988,659 B1, 2022 8. Marin et al. Med Phys, 2017. 44(4): p. 1324-36.9. Gerasimova-Chechkina et al. Front. in Physiol., 2021. 12(595).10. Batchelder et al. medRxiv 2024: p. 2024.02.17.24302978.11. Khalil & Hamilton IBDW2025 poster #70 2025.12. Halling-Brown et al. Radiol Artif Intel, 2021. 3(1): p. e200103.13. OPTIMAM (OMI-DB) project: https://medphys.royalsurrey.nhs.uk/omidb/about-omi-db/.14. Hamilton et al. Front. in Oncol., 2022. 12: p. 991850. Citation Format: A. Khalil, J. Juybari, J. Hamilton. Anisotropic active dense tissue: a strong image-based cancer risk predictor [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-04-07.