Systematic Review of Artificial Intelligence in Pre-Implant Radiographic Imaging: A Structured Qualitative Synthesis of Performance Across Imaging Modalities and AI Architectures.

PURPOSE: To evaluate contemporary evidence on artificial intelligence (AI) applied to radiographic imaging for pre-implant dental assessment and treatment planning, with emphasis on the distinction between technical algorithmic performance and clinically oriented planning outcomes. MATERIALS AND METHODS: Peer-reviewed original studies published from January 2021 through December 2025 were identified through PubMed/MEDLINE, the Cochrane Library, ScienceDirect, JSTOR, and supplementary searches. Eligible studies used radiographic imaging as an integral input to AI for pre-implant assessment, planning, decision support, virtual positioning, or outcome prediction. Reporting completeness was assessed with CLAIM 2024. Risk of bias was assessed with QUADAS-3 for eligible diagnostic/test-accuracy studies, PROBAST+AI for the prognostic prediction-model study, and RoB 2 for the randomized clinical study. Owing to heterogeneity, findings were synthesized qualitatively. RESULTS: Thirty-six studies were included. CBCT was the predominant modality. Applications included anatomical detection and segmentation, bone assessment, decision support, implant-position prediction, automated virtual placement, workflow integration, and preoperative outcome prediction. Segmentation and anatomical-analysis studies generally reported high task-specific performance, while clinically oriented positional accuracy and planning outcomes were more variable. AI frequently reduced workflow time. QUADAS-3 was applicable to five studies; four had high overall risk of bias and one unclear risk. The prognostic prediction-model and randomized clinical studies also had high risk of bias under their respective tools. CONCLUSIONS: AI shows greatest current support for anatomical analysis and selected workflow tasks. Evidence for autonomous implant positioning and outcome prediction remains limited and heterogeneous. Current findings support clinician-supervised integration, with further multicenter external testing, standardized clinically meaningful outcomes, reproducibility, and prospective evaluation required before broader routine implementation.