Real-Time Human–AI Collaborative Radiotherapy Planning for Nasopharyngeal Carcinoma: Multi-Center Validation and Prospective Clinical Deployment
Abstract
Purpose
To evaluate a deep-learning based real-time automated radiotherapy planning pipeline for nasopharyngeal carcinoma (NPC) in both retrospective multi-center study and prospective clinical deployment.
Methods
We developed a deep learning-based model specifically designed for real-time automated planning in NPC. The model was trained on a dataset of 890 patients and iteratively refined through four versions, incorporating innovations such as quantile loss, priority-based constraint encoding, and hybrid CPU-GPU acceleration to improve plan quality and efficiency. Its performance was evaluated in a five-center retrospective study, including 125 patients from the model development center and 120 patients from four external centers. Subsequently, we established an online All-in-One (AIO) radiotherapy workflow that integrates simulation, planning and delivery into a single session to enable same-day treatment and streamline clinical operations. Within this workflow, the planning stage was powered by the deep learning-based planning model, deployed in a human-AI collaborative manner to generate clinically deliverable plans. The entire AIO workflow was then validated in a prospective cohort of 242 consecutively treated NPC patients using a CT-linac system.
Results
In retrospective multi-center evaluation, AI-generated plans consistently achieved superior or comparable dosimetric quality relative to expert manual plans, delivering reliable target coverage despite variations in imaging, contouring, and prescription practices across institutions. In prospective deployment, 95% of plans were clinically accepted after a single optimization cycle, with a mean generation time of 3.5 minutes, indicating high clinical readiness. All plans met target coverage criteria and passed both secondary dose verification and in vivo EPID analysis.
Conclusion
This study represents the largest prospective validation to date of AI-based planning for NPC, demonstrating real-time feasibility, robust generalizability, and consistent clinical quality. Our development-to-deployment framework supports the scalable adoption of AI-driven planning and provides a transferable model for intelligent radiotherapy across disease sites.