2193 - Simulation and Planning Free Ultra Hypofractionated Whole-Breast Radiation Therapy (Fast Forward) Using AI Empowered Adaptive Radiation Therapy (ART)

Purpose/Objective(s): Direct-to-treatment radiotherapy bypasses simulation and planning, shortens the consultation-to-treatment time from weeks to hours, and expands access to care. We hypothesize that direct-to-treatment ultra hypofractionated whole breast radiotherapy (Fast Forward¹) is feasible with CBCT-guided ART that integrates accurate Hounsfield numbers for direct dose calculation on CBCT and artificial intelligence for segmentation.

Materials/Methods: We retrospectively analyzed 11 left sided whole breast patients who previously underwent CBCT-guided ART. The CT images, contours, and plan of a representative case was used to create a generalized template patient. The template patient was then applied as placeholders in the adaptive system for the 10 remaining patients. During online adaption, AI automatically segmented organs-at-risk (OARs)—lungs, heart, right breast, stomach, and spinal canal—as well as the target (left breast) on the patient-specific CBCT. The placeholder plan was then adapted to each patient’s anatomy to generate treatment plans. Dose constraints included PTV normalization to V26Gy = 90%, with V105% < 7%, V107% < 2%, and Dmax < 110%. Hard constraints were applied for the heart (V1.5Gy < 50%, V3Gy < 10%, V7Gy < 5%) and left lung (V8Gy < 15%), while all other OARs followed the ALARA principle. The AI-generated PTV was compared to the physician-delineated contours using the Dice similarity coefficient (DSC).

Results: The templated placeholder plan, derived from a single representative patient, was successfully adapted to the anatomy of all 10 patients without requiring major modifications. The entire adaptive workflow—including AI segmentation, plan adaptation, and dose calculation—was completed in 18.7 ± 2.2 minutes per patient. All adaptive plans met the predefined planning objectives and constraints. The heart dose metrics were V1.5Gy = 27.5 ± 6.9%, V3Gy = 5.8 ± 1.9%, and V7Gy = 1.0 ± 0.9%, while the left lung dose remained within safe limits (V8Gy = 12.1 ± 0.5%). The AI-generated PTV demonstrated a high degree of agreement with the clinically treated PTV, achieving a DSC of 0.90 ± 0.02. AI underestimated the superior–inferior direction of the PTV but would only require minimal physician intervention online. This demonstrates that a single pre-defined template can be effectively applied across multiple patients, ensuring both efficiency and plan quality in a direct-to-treatment workflow.

Conclusion: Direct-to-treatment Fast-Forward radiation therapy is feasible with current clinical technology, allowing seamless adaptation of a templated plan to multiple patients without compromising plan quality. While physician edits to AI-generated volumes are still required, integrating adaptive planning into clinical workflows remains efficient. Future improvements in AI models—customizable to individual practice contouring policies—could further enhance automation and streamline the process.