Robustness Evaluation of Physical versus Virtual Bolus for Breast VMAT Using RegGAN-Based Synthetic CT

Volumetric modulated arc therapy (VMAT) for breast cancer with skin involvement requires a bolus to ensure adequate surface dose. Although both physical and virtual bolus methods are clinically established, their dosimetric robustness against interfraction anatomical variations remains underexplored. This study utilized deep learning-based synthetic CT (sCT) to evaluate the robustness of these two strategies.

Ten patients with skin-involved breast cancer treated with VMAT were retrospectively analyzed. Two plans were generated for each patient: plan_pb (physical bolus during simulation) and plan_vb (virtual bolus optimization). To evaluate robustness, pretreatment cone beam CT images from fractions 1, 6, and 11 were converted to sCT using a registration-based generative adversarial network (RegGAN). Dose distributions were recalculated on these sCT to quantify deviations in the conformity index (CI), homogeneity index (HI), and volume percentage of the prescribed dose (Vpd).

Initial planning showed no statistically significant differences between the two methods in target coverage or organs at risk sparing. In the robustness analysis performed on sCT, the virtual bolus method (plan_vb) demonstrated statistically superior robustness for Vpd (P < 0.05), although the absolute magnitude of improvement was small (mean Vpd deviation: -0.8% vs -1.5%). The physical bolus method (plan_pb) showed better robustness for CI.

Using a RegGAN-based evaluation framework, this study demonstrates that although both bolus strategies are dosimetrically comparable, the virtual bolus method provides marginally improved robustness in target dose coverage against daily anatomical variations.