Integration of AI-Based Synthetic CT Generation and Auto-Segmentation for CBCT-Guided Adaptive Radiotherapy in...

Background/Objectives: Adaptive radiotherapy (ART) is essential yet expensive in prostate cancer treatment. This study aims to preliminarily validate the feasibility of an integrated AI workflow for CBCT-guided adaptive radiotherapy in prostate cancer, and to provide a technical foundation for subsequent clinical translation. Methods: Planning CT and CBCT images from 120 patients were used for training and validation, while 21 patients were reserved for testing. A CycleGAN-ResNet generated synthetic CT (sCT) from CBCT, and an nnU-Net model performed auto-segmentation on the sCT. Image quality and segmentation accuracy were quantitatively assessed. The original plan was recalculated on sCT to evaluate actual dose delivery; if clinical constraints were unmet, adaptive re-optimization was performed, and plans were compared. Results: The total time per fraction in this study was approximately 19 ± 6 min, falling within the reported feasibility range for online ART. sCT image quality was significantly improved, making them suitable for subsequent auto-segmentation and treatment planning. The auto-segmentation technique substantially enhanced contouring efficiency, with the automatically generated contours requiring only minor modifications to meet clinical standards. In dosimetric analysis, the adaptive plans provided superior target coverage, CI, and HI. Compared with the actual dose delivered by the original plan, the adaptive plans yielded lower bladder V40 and lower rectal mean dose/V30/V40/V50. Conclusions: This study preliminarily validated the feasibility of an integrated AI workflow that concatenates CycleGAN-based sCT generation, nnU-Net-based auto-segmentation, and sCT-based adaptive plan re-optimization. Compared with conventional segmented studies, this integrated exploration facilitates a more comprehensive assessment of the potential value of AI technologies in CBCT-guided prostate cancer ART, offering a preliminary solution for promoting a cost-effective adaptive radiotherapy approach.