To develop and validate an artificial intelligence (AI) model for automated gingival segmentation on cone-beam computed tomography (CBCT) using the intraoral scan as a reference. A total of 101 CBCT scans (180 arches) were divided into training (50 scans/80 arches), validation (15/40), and testing (36/60) sets. Manual segmentation (MS) served as the ground truth based on intraoral scan. AI-driven segmentations (AS) were refined by an expert (R-AS) to correct oversegmentation and undersegmentation and then compared to AS to assess model performance. A subset of the testing dataset (30 arches, 15 maxilla/15 mandible) was used to compare AS and MS. Accuracy was evaluated using Dice similarity coefficient (DSC), median surface deviation (MSD), and root mean square deviation (RMS). MS, AS, and R-AS times were recorded and compared. The AI model achieved high accuracy with DSC of 93% (maxilla) and 91% (mandible). MSD values were 0.07±0.05 mm (maxilla) and 0.04±0.04 mm (mandible). RMS values were 0.40±0.29 and 0.24±0.11 mm, respectively. Maxillary segmentation was significantly more accurate across all metrics (p<0.01). AS performed similarly to MS for maxillary and mandibular gingival segmentation, except for maxillary MSD, where MS outperformed AS (0.09±0.04 vs. 0.02±0.03 mm; p<0.05). AS (5.9±1 s) and R-AS (1096±339 s) were significantly faster than MS by 725 and 4 times, respectively. The AI model demonstrated high accuracy and time efficiency in overall gingival segmentation, with expert refinement supporting clinical optimization of the digital workflow.
Papasratorn et al. (Wed,) studied this question.