Reliability of AI-Automated and Semiautomated Upper Airway Volume Segmentation

Background/Objective: To evaluate the reliability, diagnostic accuracy and time efficiency of an artificial intelligence (AI)-automated method (CephX) and a semiautomated method (INVIVO) for upper airway segmentation, the manual digital method (ITK-SNAP) was used as the reference standard. Methods: This retrospective study analyzed cone-beam computed tomography (CBCT) scans from 133 patients. The upper airway volume and narrowest cross-sectional area were measured via the three methods. Reliability and repeatability were assessed via the intraclass correlation coefficient (ICC). The time required for each segmentation method was also recorded and compared. Results: Both the AI-automated (ICC = 0.945) and semiautomated (ICC = 0.992) methods demonstrated excellent reliability for total volume measurements compared with the manual reference. For the narrowest area, the automated method showed excellent agreement (ICC = 0.943), whereas the semiautomated method showed good agreement (ICC = 0.868). All methods demonstrated excellent intrareader repeatability (ICC > 0.95) and high test-retest reliability. The AI-automated method was significantly more time-efficient, requiring less than 30 s per analysis, compared with 161.4 s for the semiautomated method and 336.6 s for the manual method. Conclusions: AI-automated and semiautomated segmentation methods are reliable and accurate alternatives to manual upper airway analysis. The AI-based approach offers a substantial advantage in time efficiency, making it a valuable tool for clinical practice.