What AI Can (and Can't Yet) See on a Peri-Implantitis X-Ray

Radiographic diagnosis of peri-implantitis is notoriously operator-dependent, and measurable marginal bone loss is often detected only after the disease has progressed. This comprehensive review set out to synthesize the evidence on artificial intelligence (AI)-based radiographic approaches for detecting peri-implantitis, quantifying marginal bone loss, and stratifying patient risk. The authors searched PubMed and Scopus for original studies published between 2013 and 2025 that applied machine learning or deep learning to peri-implant bone assessment and reported quantitative performance metrics. Ten studies met the eligibility criteria (of eleven initially identified, one full text could not be retrieved), and their imaging modalities, AI architectures, validation strategies, and clinical relevance were extracted for a qualitative synthesis.

Most studies used periapical or intraoral radiographs, with only a handful relying on panoramic or combined imaging. Reported architectures spanned object-detection and segmentation models (YOLO variants, Faster R-CNN, Mask R-CNN, U-Net) as well as classification networks (ResNet, AlexNet). Performance was generally encouraging across tasks including implant localization, bone loss detection, keypoint identification on the implant-bone interface, and severity classification. Only one study went further and attempted to predict clinical outcomes rather than simply describe current bone status.

Despite these promising results, the review is clear about important caveats: every included study was retrospective and validated only internally, reference standards for bone loss varied across studies, and clinical data (smoking status, oral hygiene, prosthetic design) were rarely integrated into the models. This limits how confidently the findings can be generalized to everyday clinical populations.

For clinicians, the take-home message is that AI-assisted radiographic analysis is a maturing adjunctive tool — one that could eventually flag early or subtle marginal bone loss that a human eye might miss on a busy recall day — but it is not yet ready to replace clinical judgment or standardized radiographic protocols. The authors call for prospective, multicenter studies with external validation, longitudinal imaging series, and models that combine imaging with clinical risk factors before AI tools can be recommended for routine peri-implantitis screening.