370 - Pioneering an AI-Empowered All-in-One Radiotherapy Workflow for Breast Cancer: Advancing Dosimetric Consistency and Patient Care

09:45am - 09:55am ET

Room 107

Presenter(s)

Xiaofang Wang, MD, PhD - Fudan University Shanghai Cancer Center, Shanghai, Shanghai

L. Yu, X. Wang, J. Zhao, J. Wang, W. Hu, Z. Zhang, and X. Yu; Department of Radiation Oncology, Fudan University Shanghai Cancer Center; Department of Oncology, Shanghai Medical College, Fudan University; Shanghai Clinical Research Center for Radiation Oncology; Shanghai Key Laboratory of Radiation Oncology, Shanghai, China

Purpose/Objective(s):

Conventional radiotherapy workflows involve multi-day delays between CT simulation and the first treatment, often causing patient anxiety and potential anatomical variations. An emerging "All-in-One" workflow seeks to eliminate these delays by integrating simulation, contouring, planning, and beam delivery into a single on-couch session. This study evaluates the first clinical implementation of this AI-driven workflow for breast cancer, assessing its procedural performance and overall clinical impact.

Materials/Methods:

A prospective cohort of 25 patients (14 left-sided, 11 right-sided) underwent the All-in-One workflow between 2021 and 2025. All received whole breast irradiation with a simultaneous integrated boost (40.05-48Gy/15 fractions) following breast-conserving surgery. Deep learning models, trained on historical institutional data, were utilized for target auto-segmentation and dose prediction-based auto-planning. Contouring accuracy was assessed using the Dice Similarity Coefficient (DSC) before and after manual revision. Dosimetric quality was compared against a conventional workflow cohort of 177 cases (71 left, 106 right) using Inverse Probability of Treatment Weighting (IPTW) to balance anatomical differences. Session duration, waiting intervals, and body volume changes were analyzed.

Results:

The one-stop workflow was successfully completed for all 25 patients, averaging 24.7±4.8 minutes from simulation to delivery. Auto-segmentation required minimal manual intervention, achieving excellent DSCs of 0.978±0.022 and 0.821±0.158 for the CTV-breast and tumor bed, respectively. The AI-generated plans exhibited superior dosimetric consistency over the IPTW-matched manual plans. Specifically, ipsilateral lung V4Gy and V16Gy were significantly reduced for left-sided patients (from 40.2% to 33.3%, p<0.01; and 16.3% to 13.9%, p=0.04, respectively), while mean heart dose was lowered for right-sided patients (from 1.2 Gy to 0.8 Gy, p<0.01). Notably, the workflow eliminated the sim-to-treat body volume changes in target slices (0.14%-9.13%) observed during the median 21-day waiting period of the conventional cohort, ensuring treatment initiation within the optimal clinical window. In the All-in-One cohort, no Grade =2 acute skin toxicity or pneumonitis was reported. Patient satisfaction was notably high due to the simplified clinical pathway, daily life integration, and reduced non-medical expenses.

Conclusion:

The first implementation of the All-in-One workflow for breast cancer delivers substantial clinical value through dosimetric superiority and enhanced patient care. By condensing the preparation timeline into a single, seamless session, this approach significantly improves patient experience, mitigates risks associated with interval anatomical changes, and establishes a highly efficient new paradigm for breast cancer radiotherapy.