The generative adversarial networks (GANs) for case- or scenario-based teaching in orthodontic education, a scoping review

Abstract Background Generative Adversarial Networks (GANs) can produce high-fidelity synthetic images without exposing real patient data, offering a potential solution to the patient-consent and privacy constraints that limit case- and scenario-based teaching in postgraduate orthodontic education. Objective The present scoping review maps the evidence that has explored or developed GAN-based applications relevant to orthodontic education, with particular focus on the synthesis of facial photographs (including de-identification), panoramic radiographs (OPGs), lateral cephalograms, cone-beam computed tomography (CBCT) images, and intraoral photographs. Methods PubMed, Scopus, ERIC, and IEEE Xplore were searched from 2014 (the year GANs were first described) to 28 June 2026, combining terms for orthodontics or craniofacial conditions, GAN architectures (vanilla, conditional, deep convolutional, style-based, cycle-consistent), and imaging or synthesis-related terms, supplemented by manual reference-list screening of relevant reviews. Records were de-duplicated, then screened by title and abstract; studies were included if they primarily used or demonstrated the potential of GANs for generating orthodontic-relevant images. Full texts of remaining studies were reviewed, and included studies were grouped by image type. Results After abstract review, 79 articles were identified for full text review and from them, 7 studies met inclusion criteria, trained on 302 to 50,000 datasets, and were grouped into five categories: facial de-identification of frontal or profile facial images ( n = 2), OPG synthesis ( n = 2), lateral cephalogram synthesis ( n = 1), CBCT synthesis ( n = 1), and intraoral photograph synthesis ( n = 1). Across all groups, GAN-generated images were frequently indistinguishable from real images even to trained clinicians, particularly at moderate resolutions, while very high-resolution outputs and fine anatomical detail (individual tooth morphology, cortical bone texture, soft tissue) remained the most reliable cues for expert detection. Synthetic images improved downstream diagnostic AI accuracy when used to balance imbalanced training datasets, and AI-enhanced low-dose CBCT preserved clinical decision-making while reducing radiation exposure. Conclusions GANs hold promise for advancing orthodontic education; However, the lack of standardization across GAN studies remains an issue. Limited evidence suggests that GAN-generated dental and facial images are sufficiently realistic, but current evidence derives from single-centre, single-population pilot studies with limited external validation. Larger, more diverse multi-centre datasets and conditional models tailored to specific malocclusion types or facial forms are needed before wider adoption. Furthermore, GAN-generated images contain clinically relevant information that can be used to expand existing datasets and train deep-learning models.