To investigate the performance of image registration methods for structure delineation in head and neck (H&N) cancer patients.

This retrospective study randomly recruited 22 patients who had been irradiated in the H&N region between January 2016 and February 2024. The sample group included nasopharyngeal carcinoma (NPC) and oropharyngeal cancer (OPC) patients. The treatment planning structures were delineated as images of computed tomography simulation (CTsim) and were set as the ground-truth. The latest CT diagnostic (CTdiag) image sets of these selected patients were imported into third-party software for delineation. The structures of CTdiag were delineated using an artificial intelligence method except for the target. The performance of rigid and deformable image registration methods (RIR and DIR, respectively) between these two image sets were evaluated using dice similarity coefficient (DSC) and Hausdorff distance (HD). The performance evaluation scores were also compared between NPC and OPC.

The DSC revealed a significant difference in all structures between RIR and DIR, whereas the HD showed no significant difference on the target and the larynx. In terms of a comparison of treatment regions, OPC appeared to sustain the greatest benefit from DIR.

Image registration can provide the benefit of structure delineation, particularly when employing the DIR method. Although the DIR method may not offer a high degree of performance in terms of target delineation, it could effectively serve as a delineation guideline in this process.