Sep 30

QP 10 - DHI 2: Quick Pitch: The Digital Revolution in Radiation Oncology: AI Models for Enhanced Patient Care

1054 - Development and Prospective Validation of a Fully AI-Driven Online Adaptive Radiotherapy Workflow

12:50pm - 12:55pm PT

Room 20/21

Presenter(s)

Guan-Qun Zhou, MD, PhD - Sun Yat-Sen University Cancer Center, Guang Dong Province, Guangdong

Y. X. Yang¹, G. Y. Wang¹, X. Yang¹, L. Lin¹, Y. Li¹, K. Zhang², B. H. Li², H. Li³, L. Jia³, Y. Liu³, B. H. Li², W. Zhang², J. Zhou², F. Chi¹, Y. P. Mao¹, R. Guo¹, X. Huang⁴, G. Q. Zhou¹, and Y. Sun⁴; ¹Department 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, ²Shanghai United Imaging Healthcare Co., Ltd., Shanghai, China, ³Shenzhen United Imaging Research Institute of Innovative Medical Equipment, Shenzhen, China, ⁴State 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

Purpose/Objective(s):

Radiotherapy efficacy critically depends on millimeter-precise dose delivery, yet anatomic changes during treatment compromise precision, adversely affecting survival and quality of life. Online adaptive radiotherapy (ART) addresses these variations through rapid plan adjustments, but an efficient online ART workflow remains lacking. We developed and validated a fully AI-driven online ART workflow, using nasopharyngeal carcinoma (NPC) as a model.

Materials/Methods: AI algorithms were integrated across all stages of the online ART workflow. For image construction, a deep learning-based algorithm was used to metal artifact correction. For contouring, a multi-class model was designed to simultaneously segment anatomically adjacent organs at risk (OARs) to improve efficiency, while knowledge-guided AI models were developed to segment target volumes with anatomical and positional priors to enhance accuracy. For planning, a label-guided AI model was utilized for prioritizing clinical objectives to enhance plan quality, while a fast Monte Carlo model was used to improve efficiency. For quality assurance and treatment, a machine-learning method was used to predict gamma passing rate. These models were integrated into a CT-linac platform to establish a fully AI-driven online ART workflow. We prospectively enrolled 122 treatment-naïve NPC patients to validate its clinical feasibility.

Results: Iterative optimization reduced AI contouring time to 1.5 minutes while improving segmentation accuracy (Dice Similarity Coefficients of 0.79 for involved cervical lymph nodes, 0.91 for high-risk clinical target volume, and 0.92 for low-risk clinical target volume). Planning optimization time decreased to 4 minutes with 100% first-pass acceptance. Prospective validation showed 98.4% success rate, a median time of 20.8 minutes, and superior dosimetry. Early clinical outcomes showed 99.2% complete response rates and manageable acute toxicity.

Conclusion: The fully AI-powered online ART workflow is clinically feasible, improving radiotherapy precision and potentially enhancing patients'quality of life.