The Feasibility of Direct-to-Treatment Ultra-Hypofractionated Whole-Breast Irradiation Using AI-Enabled Adaptive Radiation Therapy
Abstract
Purpose
Direct-to-treatment radiotherapy eliminates CT simulation and pre-planning, reduces consultation-to-treatment intervals from weeks to hours. This study evaluated the feasibility of an ultra-hypofractionated whole-breast workflow that omits simulation and advanced planning, instead using CBCT-guided adaptive radiation therapy (ART) with AI-generated segmentation and direct CBCT dose calculation. A secondary objective was to assess the robustness of treatment isocenter placement, a key uncertainty in direct-to-treatment workflows.
Methods
Eleven left-sided and eleven right-sided breast cancer patients previously treated using CBCT-guided ART were retrospectively analyzed. For each laterality, one patient served as the template for generating a generalized placeholder plan based on the planning CT, structure set, and reference plan. The remaining 20 patients were used for ART emulation on previously acquired CBCTs. Online workflow included AI-based segmentation, adaptive optimization, and plan normalization. Hard dose limits were applied to the heart and ipsilateral lung, with all other organs-at-risk (OARs) managed according to ALARA principles. Agreement between AI-generated and physician contours was calculated using the Dice similarity coefficient (DSC). Isocenter robustness was evaluated by introducing systematic shifts of up to 2cm in all directions and analyzing the dosimetric impact on target and OAR metrics.
Results
All placeholder plans were successfully adapted on CBCT without manual edits, and the full adaptive workflow was completed in 12.6±1.9 minutes per fraction. All adapted plans met PTV coverage and OAR constraints. AI-generated PTVs showed high concordance with physician contours (DSC=0.91±0.02), with minor under-segmentation primarily in the superior–inferior direction. Simulated isocenter deviations of up to 2 cm did not degrade PTV coverage or ipsilateral lung dose, demonstrating strong robustness to isocenter placement uncertainty.
Conclusion
Ultra-hypofractionated whole-breast radiotherapy is feasible in a simulation- and planning-omitted direct-to-treatment workflow when delivered with CBCT-guided ART and AI-assisted segmentation. The workflow demonstrates robustness to isocenter variation and supports safe, rapid initiation of breast radiotherapy.