1063 - From Feasibility to Clinical Practice: Direct-to-Treatment Ultra-Hypofractionated Whole-Breast ART with AI-Based Segmentation

03:20pm - 03:25pm ET

Room 156

Presenter(s)

Shanshan Tang, PhD - UT Southwestern Medical Center, Dallas, TX

S. Tang¹, J. Visak¹, T. Zhuang¹, C. S. Lin², C. Y. Liao¹, M. Arbab¹, S. J. Domal¹, N. Wandrey¹, C. Tye¹, P. G. Alluri¹, D. J. Sher¹, S. N. Badiyan³, A. S. Rahimi¹, M. H. Lin¹, and D. D. M. Parsons¹; ¹Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, TX, ²University of Texas Southwestern Medical Center, Dallas, TX, ³University of Texas Southwestern Medical Center, Department of Radiation Oncology, Dallas, TX

Purpose/Objective(s): Direct-to-unit (DTU) radiotherapy, which omits CT simulation and prospective planning, can shorten consultation-to-treatment intervals from weeks to hours and expand access to timely breast cancer care. Building on prior feasibility work, this study integrates ART emulation testing with early clinical implementation of an ultra-hypofractionated whole-breast workflow enabled by CBCT-guided adaptive radiotherapy (ART), AI-generated segmentation, and direct CBCT dose calculation. A secondary aim was to evaluate workflow robustness to treatment isocenter uncertainty, a key consideration for DTU models.

Materials/Methods:

Feasibility was first assessed through ART emulation in 22 patients (11 left-, 11 right-sided) retrospectively analyzed. For each laterality, a template patient generated a generalized placeholder plan, which was then adapted on previously acquired CBCTs (Ethos; Varian). AI-based segmentation, adaptive optimization, and normalization preserved PTV coverage while meeting hard constraints for the heart and ipsilateral lung. Segmentation quality was quantified using Dice similarity coefficients (DSC). Isocenter robustness was evaluated by applying systematic shifts up to 2 cm.

Clinical implementation was subsequently evaluated in 10 prospective patients treated with both the standard and DTU adaptive pathways. One fraction per patient was delivered with the DTU intent, with online physician review and editing of AI-generated target and OAR contours. Workflow duration, DSC values between online-adapted and preplan volumes, and dosimetric metrics were recorded.

Results:

Emulation: All patients were successfully adapted with no manual edits; workflows averaged 13 minutes. Adapted plans met all PTV and OAR constraints, and AI-generated PTVs showed high concordance with physician contours (DSC 0.91 ± 0.02). Isocenter deviations up to 2 cm did not compromise PTV coverage or OAR dose.

Clinical implementation: Breast volumes varied widely (mean 1383 ± 464 cc). Treatment times were comparable between conventional and DTU workflows (24.2 ± 3.7 vs. 24.5 ± 4.0 minutes). Physician-reviewed PTV generated during both the DTU and conventional adaptive workflows showed similarly high agreement with patient’s reference plan PTV (DSC 0.92 ± 0.02 for conventional fractions; 0.90 ± 0.03 for DTU fractions), and all DTU fractions met target coverage and OAR limits.

Conclusion: ART emulation confirmed strong feasibility and isocenter robustness for a simulation- and planning-omitted direct-to-treatment breast workflow. Early clinical deployment demonstrated seamless integration into routine practice without increasing treatment time and with preserved plan quality. These results support broader adoption of DTU breast radiotherapy to accelerate care delivery and reserve simulation and planning capacity for more complex cases.