3765 - AI-Enhanced Automated Planning System for Online Adaptive Radiotherapy in Nasopharyngeal Carcinoma: Development and Prospective Clinical Validation
Presenter(s)
Y. Sun1, Y. X. Yang2, J. Tang3, L. Jia4, H. Li4, Y. Liu4, W. Zhang3, J. Zhou3, R. Guo2, X. Yu5, X. Yang2, G. Y. Wang2, and G. Q. Zhou2; 1State Key Laboratory of Oncology in South China, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangdong Provincial Clinical Research Center for Cancer, Department of Radiation Oncology, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China, 2Department of Radiation Oncology, Sun Yat-sen University Cancer Center, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangzhou, China, 3Shanghai United Imaging Healthcare Co., Ltd., Shanghai, China, 4Shenzhen United Imaging Research Institute of Innovative Medical Equipment, Shenzhen, China, 5Sun Yat-sen Memorial Hospital, Guangzhou, China
Purpose/Objective(s): To develop and validate an AI-enhanced automated planning system (AI-APS) that simultaneously optimizes plan quality and optimization efficiency for online adaptive radiotherapy (OART) in nasopharyngeal carcinoma (NPC).
Materials/Methods: The automated planning system (OmniPlan, United Imaging, China) was enhanced by integrating two novel AI-driven modules: 1) A constraint-prioritized dose prediction (CPDP) model to resolve dosimetric conflicts between target coverage and organ-at-risk (OAR) constraints, with specialized optimization for advanced-stage (T3-T4) cases; 2) A rapid planning tool calculation model supplanted conventional MC calculation during plan optimization, reducing computation time from minutes to seconds while maintaining dosimetric accuracy. System validation was conducted in two phases: Initial technical validation utilizing 20 historical T1-T4 NPC cases to assess plan quality metrics and optimization efficiency, and subsequent prospective clinical validation involving 120 treatment-naive NPC patients undergoing online adaptive radiotherapy (OART) on the CT-Linac platform (uRT506c, United Imaging, China), with the clinical phase evaluating plan first-pass success rates and optimization time to confirm clinical feasibility.
Results: Compared to the baseline automated planning system, the AI-APS demonstrated significant improvements in technical validation, increasing the first-pass success rate from 60.0% to 100.0% and reducing the mean optimization time from 15.0 minutes to 4.0 minutes. In a clinically challenging advanced NPC case involving severe PTV-brainstem anatomical conflict, the CPDP model achieved an 11.3% reduction in brainstem maximum dose (6,768?6,000 cGy) while maintaining clinically acceptable target coverage (98.1% vs. 96.5%). The RMCC model further enhanced efficiency, enabling a 4.5-fold reduction in optimization time (18.0?4.0 minutes) without compromising plan quality. Prospective clinical validation yielded a 98.3% first- pass success rate (118/120 cases), with a mean optimization duration of 3.5 minutes.
Conclusion: This AI-APS enables rapid generation of clinically acceptable OART plans for NPC, effectively addressing the dual challenges of plan quality and optimization efficiency in online adaptive radiotherapy workflows.