Abstract
BACKGROUND: Breast positioning and compression are integral to digital mammography (DM) and digital breast tomosynthesis (DBT) imaging. Therefore, their accurate representation in computer simulations using anthropomorphic digital breast phantoms can be crucial. However, existing compression models often neglect external conditions, such as breast support elevation and radiographer handling, and lack quantitative validation against reference compressed breast shapes. PURPOSE: To optimize and validate a finite element compression model for digital breast phantoms in the cranio-caudal (CC) view using a statistical shape model derived from patient structured light (SL) scans. METHODS: Uncompressed breast phantoms were generated from segmented dedicated breast CT scans of 88 patients. A subset of 30 phantoms was selected to balance representativeness and computational efficiency. Simulated compression was performed using a finite element solver while varying combinations of adipose tissue, fibroglandular tissue, and skin Young's moduli, breast support height, and positioning force magnitudes. After compression, principal component analysis (PCA) was performed on the phantom surfaces, and shape feature distributions were compared to those from 236 compressed patient breasts acquired via structured light (SL) scanning. A weighted Kolmogorov-Smirnov (K-S) distance was used to quantify shape agreement through the distance between the phantom and patient shape feature distributions. A multi-stage optimization strategy combining random sampling and grid searches was used to minimize this distance and identify physiologically plausible parameter ranges. RESULTS: The optimized parameter set improved agreement between phantom and patient shape feature distributions, reducing the average K-S distance from 0.50 to 0.32. External conditions, namely the breast support elevation and positioning forces, had a greater impact on agreement than elastic parameters. The analysis identified physiologically plausible parameter ranges that consistently produced realistic shapes. Tissue elastic parameters had a smaller but measurable effect and the results highlighted the importance of modeling skin as a distinct tissue type. CONCLUSIONS: This study highlights the importance of external forces and boundary conditions in achieving realistic compressed breast shapes and provides validated parameter ranges for compression simulations. The proposed intervals can serve as a basis for introducing variability in virtual clinical trials and other digital phantom studies. Further research should extend validation to internal tissue distribution and MLO compression.