Artificial intelligence (AI) has been increasingly applied to facial-image analysis in orthodontics, with potential applications in radiation-free screening, dentofacial assessment, and treatment-decision support. This systematic review evaluated the diagnostic and predictive performance of AI models applied to two-dimensional (2D) and three-dimensional (3D) facial data.

The protocol was registered in PROSPERO (CRD420261419145). PubMed, Embase, Scopus, Web of Science, Cochrane Library, ClinicalTrials.gov, Google Scholar, and Open Science Framework were searched. Eligible studies evaluated AI models using facial photographs, multi-view photographs, 3D facial scans, reconstructed facial models, or landmark-coordinate inputs for downstream diagnostic, predictive, classification, or treatment-decision tasks. Studies limited exclusively to landmark localization were excluded. Data were synthesized qualitatively. Risk of bias was assessed with QUADAS-2 and certainty of evidence with GRADE.

Fourteen studies were included. Applications comprised cephalometric prediction, skeletal-pattern classification, mandibular-deformity diagnosis, soft-tissue depth assessment, treatment-difficulty classification, and orthognathic-surgery-need prediction. Across classification studies, accuracy ranged from 73.1% to 97.7%, AUC from 0.768 to 0.987, sensitivity or recall from 64.3% to 97.3%, and specificity from 80.5% to 95.6%. Only one study performed true external validation. Methodological heterogeneity, predominantly internal evaluation, and low overall certainty limited clinical translation.

AI-based analysis of 2D and 3D facial data showed promising performance for selected orthodontic and dentofacial tasks. Current evidence supports its use as an adjunct to specialist assessment rather than as an autonomous diagnostic substitute.

CRD420261419145.