Representation of Complex Mammary Parenchyma Texture in Tomosynthesis Using Simplex Noise Simulations

The mammary parenchyma is a complex arrangement of tissues that can greatly vary among individuals, potentially masking cancers in breast screening images. In this work, we propose a Simplex-based method to simulate anatomical patterns and textures seen in digital breast tomosynthesis. Our approach involves selecting appropriate Simplex noise parameters to represent distinct categories of breast parenchyma with variable volumetric breast density (%VBD). We use volumetric coarse masks (70 × 60 × 50 mm³) to outline patches of both dense and adipose tissues. These masks serve as a foundation for volumetric and multi-scale Simplex-based noise distributions. The Simplex-based noise distributions are normalized and thresholded using gradient level sets selected to binarize specific Simplex frequencies. The Simplex frequencies are summed and binarized using post-hoc thresholds, resulting in patches of tissue tailored to represent anatomic-like structures seen in digital breast tomosynthesis (DBT) images. We simulate DBT projections and reconstructions of the patches of breast tissue following the acquisition geometry and exposure settings of a clinical tomosynthesis system. We calculate the power spectra and estimate the power-law exponent (β) using a sample of DBT reconstructions (n=500, equally stratified by four density classes). Our findings reveal an absolute β value of 3.0, indicative of the improvements achieved in both the performance and realism of the breast tissue simulation. In summary, our proposed Simplex-based method enhances realism and texture variations, ensuring the presence of anatomical and quantum noise at levels consistent with the image quality expected in breast screening exams.