Development and validation of artificial intelligence-based full dentition labiolingual reconstruction using CBCT images

Objective To development and validate an artificial intelligence (AI) -based system for automatic full dentition labiolingual reconstruction (DenRec-LL) using Cone beam computed tomography (CBCT) images. Methods DenRec-LL was developed by generating 2D panoramic images, segmenting and identifying teeth with a transformer framework, mapping tooth centroids into 3D coordinates for 2D–3D alignment, performing 3D tooth segmentation using deep learning, and reconstructing the full dentition labiolingually based on estimated tooth axes, centers, and missing-tooth prediction. 75 external CBCT datasets (25 normal dentition, 25 mixed dentition and 25 crowded dentition) were collected and used for testing. The performance of DenRec-LL system was evaluated on a 4-point scale, the geometric deviation metrics were analyzed and the clinical acceptability rate (CAR) was calculated, and the reconstruction time was recorded. Results 75 CBCT datasets with 1500 teeth demonstrates high precision and stability across normal, mixed, and crowded dentition. A clear majority of labiolingual reconstruction cases (93.1%) achieved the highest score, and CAR was approximately 94% across groups, with no significant differences. The automated process required 63 ± 4 s per case, representing an approximately 4.85-fold reduction compared to the manual process. Conclusions The DenRec-LL system could be used for full dentition reconstruction with high accuracy, robustness and efficiency. Clinical significance DenRec-LL bridges the gap between raw CBCT data and clinical decision‑making in routine orthodontic practice by automatically generating full‑dentition labiolingual images, which facilitate the assessment of root‑bone relationships and torque control.