Systematic review on artificial intelligence for alveolar bone density and dimensional assessment in implant dentistry

To systematically evaluate the diagnostic accuracy, quantitative performance, and clinical applicability of artificial intelligence (AI)-based methods for alveolar bone assessment in implant dentistry, and to determine their current level of clinical readiness. A systematic search of PubMed/MEDLINE, Scopus, and Web of Science was conducted from inception to December 2025 in accordance with PRISMA 2020 guidelines. Eligible studies were peer-reviewed human investigations applying machine learning or deep learning techniques to radiographic and/or clinical data for quantitative assessment of alveolar bone density, dimensions, or volume, and reporting diagnostic or measurement performance metrics. Fourteen predominantly retrospective, single-centre studies encompassing approximately 2,900 patients and over 12,000 implant-related imaging sites were included. Cone-beam computed tomography was the most frequently used imaging modality. AI models achieved classification accuracies ranging from 76 to 95% for bone quality and density assessment, with area-under-the-curve values between 0.76 and 0.99. Automated dimensional and volumetric analyses demonstrated high agreement with expert reference standards, with ICC values reaching 0.99 and measurement performance comparable to expert assessment. AI-assisted clinical decision-support applications, including drilling protocol selection and implant site feasibility assessment, achieved reported accuracies of approximately 90–94%. AI-based approaches demonstrate strong quantitative performance and potential to enhance objectivity and efficiency in preoperative implant planning. However, methodological heterogeneity, reliance on surrogate reference standards, and limited external validation restrict current clinical translation. AI-driven quantitative alveolar bone assessment may support clinicians in implant site evaluation and treatment planning; however, at present, these systems should be used as adjunctive decision-support tools pending validation through multicentre, outcome-driven clinical studies.