Abstract Digital dental implant planning using panoramic radiographs and cone beam computed tomography (CBCT) imaging is labor-intensive and prone to error due to clinician fatigue, limited digital expertise, and time constraints. Artificial intelligence (AI) offers a promising solution by automating image analysis. This study aimed to develop a deep learning system for segmenting edentulous maxillary ridges to support automated implant planning. A total of 209 CBCT scans were retrieved from the University Dental Hospital Sharjah image archive (Romexis, Planmeca), of which 77 met the inclusion criteria. Manual segmentation was performed using 3D Slicer software and reviewed by a third examiner. A convolutional neural network (CNN) based on the U-Net architecture was developed using the Medical Open Network for AI (MONAI) framework. The dataset was split into training (90%) and testing (10%) sets. Model performance was evaluated using the Dice Similarity Coefficient (DSC). Low-scoring cases (DSC <0.70) were inspected in detail to identify sources of discrepancy. The 77 cases comprised 30 unilateral and 47 bilateral edentulous spaces. Most involved posterior edentulism (n = 57), with fewer anterior (n = 5) or combined (n = 15). The model achieved a mean DSC of 76.57%. Discrepancies between manual and model segmentation mainly arose from annotators excluding narrow bone regions (<4 mm) or irregular sinus floors, and from smoothing during manual labelling. In several instances, the model provided greater anatomical precision than manual segmentation. The developed AI model segmented maxillary edentulous spaces with moderate-to-high accuracy. With larger, more balanced datasets and refined manual labelling protocols, this approach shows strong potential to streamline digital implant planning and enhance clinical outcomes.
Moufti et al. (Fri,) studied this question.