tomoLIBRA: Deep learning software for automated volumetric breast density estimation

This technology is a deep learning-based software tool that automatically estimates volumetric breast density directly from standard 3D digital breast tomosynthesis (DBT) images without requiring raw data.

Unmet Need: Scalable breast density assessment for standard clinical workflows

Breast density is a major independent risk factor for breast cancer, but current methods to estimate volumetric breast density (VBD) rely on raw two-dimensional mammography data. Clinical centers rarely store this raw data due to storage constraints, archiving only the processed "for presentation" images. Consequently, large-scale automated density assessment is limited, and opportunities for personalized risk stratification are missed because current tools cannot operate on the standard archived images.

The Technology: AI-driven segmentation of breast tissue from 3D DBT images

This technology, tomoLIBRA, is a deep learning model designed to calculate volumetric breast density (VBD) directly from 3D reconstructed digital breast tomosynthesis (DBT) images, which are routinely archived in Picture Archiving and Communication Systems (PACS). The software performs 3D segmentation of breast tissue into background, fatty, and dense tissue. By eliminating the dependence on raw projection data, the tool enables retrospective analysis of existing imaging archives and prospective integration into clinical screening workflows. Density measures derived from the software can support personalized breast cancer risk assessment and clinical decision-making.

The model has been validated on 1,080 DBT screening exams and tested in an independent case-control study of 834 patients, demonstrating significant association between density measurements and breast cancer diagnosis.

Applications:

• Breast cancer screening and diagnosis
• Risk stratification for breast cancer
• Longitudinal monitoring during treatment or clinical trials
• Research tool for large-scale epidemiological or population-level studies
• Integration into existing mammography platforms

Advantages:

• Works with standard archived 3D images and does not require raw image files
• Accurate and validated
• Reproducible by providing continuous quantitative density measures
• Enables retrospective analysis on existing images
• Integrates into existing radiology workflows

Lead Inventor:

Despina Kontos, Ph.D.

Related Publications:

• Ahluwalia VS, Doiphode N, Mankowski WC, Cohen EA, Pati S, Pantalone L, Bakas S, Brooks A, Vachon CM, Conant EF, Gastounioti A, Kontos D. “Volumetric Breast Density Estimation From Three-Dimensional Reconstructed Digital Breast Tomosynthesis Images Using Deep Learning.” JCO Clin Cancer Inform. 2024 Dec;8

• Ahluwalia VS, Mankowski W, Pati S, et al. “Deep-learning-enabled volumetric breast density estimation with digital breast tomosynthesis.” Cancer Res. 2022 June; 82(12_Supplement): 1929

Tech Ventures Reference:

• IR CU26015

• Licensing Contact: Joan Martinez