Artificial Intelligence in the Radiological Diagnosis of Impacted Maxillary Canines: A Systematic Review

Objectives: The aim of this systematic review was to evaluate whether artificial intelligence systems improve the diagnosis and localization assessment of impacted canines in radiological imaging. Methods: A systematic literature search was conducted across four electronic databases (MEDLINE/PubMed, Scopus, Embase, and Web of Science) for studies published after 2020, with no language restrictions. Eligible studies were comparative studies involving human subjects that evaluated AI-based systems against experienced clinicians or accepted radiological reference standards for the detection and localization of impacted canines. The risk of bias and applicability were assessed using the adapted QUADAS-3 tool. The review protocol was prospectively registered in PROSPERO (CRD42023487320). Results: The search strategy identified 110 records. After the removal of 41 duplicates, 69 articles were screened by title and abstract. Seventeen studies underwent full-text evaluation, and eight studies met the inclusion criteria and were included in the qualitative synthesis. Across the included studies, the overall risk of bias was considered high, primarily due to retrospective study design and limitations in reporting of methodological procedures. Conclusions: The available evidence does not provide high-quality studies addressing the studied issue. AI appears to yield more favorable results in CBCT analysis when compared to panoramic radiographs. However, this observation should be interpreted with caution, because the compared studies did not address the same clinical task, since these radiographs were taken in different clinical situations. Further well-designed studies with standardized datasets and external validation are required to better define the potential of artificial intelligence in orthodontic radiological diagnostics.