Evaluation of artificial intelligence in 3D dental imaging: A comprehensive review of segmentation, registration, and soft-tissue prediction

Objectives: To critically evaluate the current capabilities of artificial intelligence (AI) in three-dimensional (3D) dental imaging, with emphasis on automated segmentation, multimodal registration, and soft-tissue outcome prediction using cone-beam computed tomography (CBCT), intraoral scans (IOS), and 3D facial scans. Data: Key performance metrics were extracted from the literature, including Dice similarity coefficients (DSC) for segmentation accuracy, point-to-point deviation errors for registration precision, and mean absolute errors (MAE) for soft-tissue prediction. Sources: A comprehensive literature search was performed using PubMed, Scopus, IEEE Xplore, and ScienceDirect. The search was limited to English-language studies published between 2015 and 2025. Study selection: More than 60 recent peer-reviewed studies were included. Eligible studies applied deep learning or hybrid AI techniques to 3D dental imaging and reported validated quantitative outcomes for segmentation, registration, or predictive modeling. Conclusions: AI demonstrates near-human performance in segmentation tasks, particularly for hard tissues on CBCT and IOS, with Dice scores frequently exceeding 0.90. Multimodal registration across CBCT, IOS, and facial scans is technically feasible and consistently achieves sub-millimeter precision. In contrast, soft-tissue outcome prediction remains variable and represents the least mature application. While segmentation and multimodal fusion tools are approaching clinical readiness, predictive soft-tissue modeling requires further methodological refinement. Clinical significance: Automated 3D segmentation of dental and craniofacial structures is largely clinic-ready and offers substantial time savings for diagnosis and treatment planning. Although multimodal fusion is technically robust for virtual patient construction, clinicians should exercise caution when using current AI tools for soft-tissue prediction, as these simulations continue to exhibit clinically relevant error margins.