Postoperative computed tomography segmentation with artificial intelligence for a closed-loop workflow in computer-assisted jaw reconstruction.

Postoperative computed tomography (CT) segmentation provides invaluable information for reconstruction verification, follow-up treatment, education, and patient communication, while it remains the most challenging task due to gross anatomical variations and heavy metallic artefacts in the closed-loop workflow of computer-assisted jaw reconstruction (CAJR). This study evaluated the performance of artificial intelligence (AI)-enabled segmentation on postoperative CT. The outcomes of AI-enabled segmentation of postoperative CT were compared to clinical routine segmentation methods as the benchmark using four common evaluation metrics, namely Dice similarity coefficient (DCE), intersection over union, 95th percentile Hausdorff distance, and average Hausdorff distance. Statistical analyses were performed to study the correlation between the segmentation performances and confounding factors. AI outperformed the clinical routine method in all metrics (P < 0.013). The mean and standard deviation values of DCE by AI-enabled segmentation results of the upper skull and mandible were 95.21% ± 2.07% and 94.28% ± 3.03. None of the confounding factors significantly affected the AI-enabled segmentation of the upper skull. Preliminary observation suggested that reconstructed mandible was associated with poorer AI-segmentation outcomes (P < 0.025); deep circumflex iliac artery flap reconstruction of the mandible showed better outcomes compared to fibula free flap cases (P < 0.025) with limited sample size. This study is novel in validating the application of AI segmentation in postoperative CT scans in computer-assisted maxillofacial surgery. The results supported the integration of AI-enabled segmentation in the closed-loop workflow of CAJR.