Systematic Review of AI Applications in Orthodontic Surgery and Treatment Planning: Diagnostic Performance and Explainability Tools

Artificial Intelligence (AI) has emerged as a valuable tool in orthodontic surgery planning, offering solutions to the limitations of manual diagnosis, including subjectivity, inter-observer variability, and errors in image interpretation. This systematic literature review, conducted in accordance with PRISMA 2020 guidelines and examined 14 peer-reviewed studies published between 2020 and 2025. The included studies show that AI models frequently achieved high diagnostic performance, with several studies reporting classification accuracies above 90% for identifying the need for orthognathic surgery. For postoperative outcome prediction, mean facial morphology errors ranged from approximately 0.69 to 0.94 mm. In three-dimensional maxillofacial segmentation tasks, deep learning models reported Dice similarity coefficients between 93% and 99%. AI assisted cephalometric analysis reduced assessment time in datasets of up to 1500 cephalograms while maintaining clinically acceptable agreement with manual tracings, although variability between platforms and limited external validation were observed. Explainability methods such as Grad CAM heatmaps, attention mechanisms, three-dimensional shape-based explanations, and SHAP style feature attributions improved interpretability by highlighting relevant anatomical regions and influential features. However, these techniques were inconsistently integrated and rarely evaluated for their effect on clinician trust or automation bias. Overall, the evidence indicates that AI improves efficiency and local diagnostic performance but does not consistently outperform conventional approaches for complex global outcome prediction. Future research should focus on large multicenter datasets, standardized evaluation metrics, and hybrid workflows that combine AI support with clinician oversight to enable safe clinical adoption.