Comparison of artificial intelligence methods with manual measurements for mandibular bucco-lingual bone width assessment using cone-beam computed tomography images

Accurate bucco-lingual bone width measurement is essential for implant planning. Manual cone-beam computed tomography (CBCT) assessment is time-intensive and operator-dependent. This study aimed to evaluate the accuracy of an artificial intelligence (AI)–based system for automated bone width measurement compared with manual methods. A retrospective diagnostic accuracy study was conducted using 300 CBCT scans of posterior mandibular edentulous sites. Manual bucco-lingual bone width was measured at 2-mm intervals from the alveolar crest to 2 mm superior to the mandibular canal. A deep learning framework with U-Net + + was trained to segment the alveolar ridge and mandibular canal, followed by automated bone width measurements. Model performance was assessed using Dice score, Intersection over Union (IoU), precision, and recall for segmentation accuracy, and regression metrics (mean squared error [MSE], mean absolute error [MAE], root mean squared error [RMSE], and coefficient of determination [R²]) for comparison with manual measurements. U-Net + + demonstrated high accuracy for alveolar ridge segmentation (Dice score, 0.9798; IoU, 0.9606; precision, 0.9820; recall, 0.9778) and moderate accuracy for mandibular canal segmentation (Dice score, 0.5640). Automated bucco-lingual width measurements showed reasonable correspondence with manual values (MSE, 1.9700 mm²; MAE, 1.1900 mm; RMSE, 1.4000 mm; R², 0.5300). Qualitative analysis confirmed high visual correspondence for ridge segmentation, though variability persisted in canal delineation. The AI-based U-Net + + system reliably segmented the alveolar ridge and provided bone width measurements with moderate agreement to manual methods. Mandibular canal segmentation remained a limitation. Broader validation across jaw regions and imaging systems is recommended to enhance clinical utility.