To investigate the use of patient-specific (rather than water-based) models for permanent breast seed implant (PBSI) brachytherapy, a retrospective study of 35 PBSI patients is performed. Virtual detailed-tissue patient models are created and overlaid with Pd-103 seed geometries, allowing for simulations with egs_brachy, a new Monte Carlo code.

Considerable discrepancies in dose distributions are demonstrated. Target dose metrics are 4-26\% higher using TG43 assumptions, skin metrics are underestimated by up to 66.5\%, and large disparities are observed in heart, lung, and rib doses. The sensitivity of dose distributions to assumptions in model creation is examined. Individualized adipose-gland segmentation thresholds and realistic seed orientations are shown to be important for accurate modeling. Radioprotective lead shielding has a negligible impact on skin dose.

This thesis demonstrates the importance of detailed patient modeling for PBSI brachytherapy, illustrating the shortcomings of TG43-based simulations and contributing to the future clinical implementation of model-based dose calculation algorithms.