Artificial Intelligence for Detection and Prevention of Separated Endodontic Instruments: A Scoping Review

Given the growing integration of artificial intelligence (AI) into dental diagnostics and clinical decision-support systems, a comprehensive assessment of existing publications addressing AI applications in the context of separated endodontic instruments is highly relevant. Such an assessment is essential for guiding future research priorities, promoting standardisation of reporting frameworks, and facilitating the transition from experimental AI prototypes towards clinically reliable tools aimed at minimising the incidence of intra-canal instrument separation. To critically evaluate the current landscape of AI models addressing endodontic instrument separation and to determine whether available models support a transition from post-separation detection towards predictive and preventive risk-assessment strategies.

A qualitative evidence synthesis was conducted to summarise published studies evaluating AI and machine learning approaches related to separated endodontic instruments. This scoping review was structured in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) principles to ensure transparent study selection and reporting. Results were mapped and synthesised descriptively in 3 layers: study mapping, model mapping, and performance mapping.

Of the 8 eligible studies, only 2 (25%) were experimental investigations focused on preventive monitoring and prediction of endodontic instrument separation-related changes. Radiology-based detection was the dominant research direction, represented in 6 of 8 studies (75%). For periapical radiograph-based detection of separated instruments, AI-based models reported accuracy values ranging from 0.795 to 0.982, with the highest values observed for Mask Region-based convolutional neural network (accuracy =0.982) and YOLOv8 (accuracy=0.974). Performance was generally lower for panoramic radiographs. In the cone-beam computed tomography (CBCT)-based approach, true-positive detection was reported at 80.8%, although detailed comparative metrics were limited.

Most available AI models related to endodontic instrument separation focus on digital radiological images, including both two-dimensional modalities (panoramic and periapical radiographs) and three-dimensional imaging (CBCT), whereas only a few studies have addressed monitoring the functional condition of instruments during their performance to verify potential critical changes related to the risk of separation. Convolutional neural networks were the dominant methodological approach, whereas traditional ML algorithms were used mainly in preventive experimental studies.