3486 - AI-Enabled Multimodal Image-Guided Radiotherapy for Nasopharyngeal Carcinoma: An Integrated Diagnostic-Therapeutic Workflow

Purpose/Objective(s):

Radiotherapy for nasopharyngeal carcinoma (NPC) increasingly depends on advanced imaging to achieve accurate staging and biologically informed target delineation. Simultaneous PET/MR imaging provides complementary metabolic and high-resolution anatomical information but is rarely incorporated into a continuous diagnostic-to-treatment radiotherapy workflow. We developed and evaluated an artificial intelligence (AI)–enabled clinical workflow centered on PET/MR–based assessment, integrating AI-assisted staging with CT-linac–based one stop radiotherapy to establish a unified diagnostic–therapeutic treatment pathway.

Materials/Methods:

All patients underwent thermoplastic mask immobilization followed by same-day head-and-neck PET/MR imaging acquired in the treatment position. AI–based system generated structured imaging reports to support staging confirmation. On the treatment day, physicians delineated primary and nodal gross tumor volumes (GTVp and GTVn) using PET/MR data. On a CT-linac system, fan-beam CT (FBCT) imaging was acquired. AI algorithms automatically segmented target volumes and organs at risk, followed by automated treatment planning. Workflow efficiency was evaluated by measuring the time from FBCT acquisition to completion of the first treatment fraction. Safety and feasibility were assessed using predefined dosimetric metrics.

Results:

Fifteen patients completed the workflow successfully. The median time from FBCT acquisition to completion of the first treatment fraction was 28 minutes (26–34 minutes). AI-generated contours showed strong agreement with physician-defined references, with Dice similarity coefficients greater than 0.85. All treatment plans met institutional dose constraints for target coverage and organ-at-risk protection. Workflow completion was achieved in all patients.

Conclusion:

This study demonstrates the clinical feasibility of an AI-enabled PET/MR–guided radiotherapy workflow for NPC. Integration of metabolic PET information with high-resolution MR imaging supported accurate target delineation while enabling efficient adaptive treatment delivery. Automated contouring and planning simplified the treatment process without compromising dosimetric quality. These results suggest that integrating advanced imaging and AI into radiotherapy workflows may improve the consistency and efficiency of precision radiotherapy for NPC.

Table 1. Some details of the workflow.