Artificial intelligence for detection and classification of furcation defects using radiographic imaging: A systematic review

Abstract

Purpose: This systematic review aimed to identify, appraise, and synthesize evidence on the diagnostic accuracy of artificial intelligence (AI) algorithms for detecting and classifying furcation defects on radiographic images, addressing limitations of traditional methods. Materials and Methods: A comprehensive search of databases and registers was conducted through April 2025. Inclusion criteria comprised diagnostic accuracy studies evaluating AI algorithms against a reference standard (clinical examination, expert consensus, or surgical findings) for furcation defect detection/classification on dental radiographs. Risk of bias was assessed using QUADAS-2. Because of study heterogeneity, a meta-analysis was not performed. Results: Eight retrospective studies were included, utilizing various AI algorithms (e.g., ResNet, UNet, YOLO-v4, Vision Transformers) and radiographic modalities (periapical, panoramic, CBCT). Studies employing advanced deep learning models on 2D radiographs generally reported high diagnostic accuracy for detecting furcation involvement, with several reporting high sensitivity, specificity, and AUC values. However, performance varied by AI model and imaging modality. Proprietary AI tools showed suboptimal results in some studies. Classification of furcation severity was less consistently reported. Conclusion: AI algorithms, particularly advanced deep learning models applied to well-annotated 2D radiographs, show promise for accurate furcation defect detection. Nonetheless, the field exhibits methodological and reporting heterogeneity. Future research should prioritize standardized protocols, direct comparisons with clinicians, and development of clinically translatable AI tools to improve early and accurate diagnosis of furcation involvement.

Ähnliche Arbeiten