Using artificial intelligence in the evaluation of periapical, caries, and restoration status: a new methodological and technological study.

Authors

Abstract

This study aimed to assess and compare the accuracy of the radiographic diagnosis of caries and periapical status conducted by undergraduate students, endodontics residency students (ERS), and artificial intelligence (AI) applications. The gold standard for diagnosis was based on the conclusions of an endodontist. Among the 1,540 patients who visited the Department of Endodontics of the Faculty of Dentistry for endodontic treatment from 2020 to 2023, 68 patients met the inclusion criteria and were selected for this retrospective study. A total of 191 endodontically treated teeth were examined based on 68 panoramic radiographic images at two years post-treatment. The examiners were two undergraduate students, two endodontics residency students, two endodontists, and an AI application. Each examiner made independent diagnoses by evaluating the same panoramic radiographs and evaluating periapical conditions, caries, and restoration assets of the included teeth. Restoration and caries were scored as present or absent. The radiographic periapical status of teeth was evaluated using Periapical Index Scoring (PAI). Data were statistically analysed using the chi-squared test and Pearson’s correlation coefficient. Intra-observer and inter-observer agreement were calculated using the kappa test (α = 0.05). The AI system demonstrated moderate agreement with the gold standard in detecting caries (κ = 0.653) and dislodged coronal restorations (κ = 0.879), but only slight agreement for periapical status (κ = 0.155). In the detection of caries, AI showed 65.3% sensitivity and 96.8% specificity, while expert raters achieved over 91% in both metrics. For dislodged coronal restorations, AI yielded high performance, with 82.1% sensitivity, 100% specificity, and 100% positive predictive value (PPV). In periapical status assessment, although AI achieved high sensitivity (92.9%) and negative predictive value (99%), its low specificity (58.6%) and PPV (15.3%) indicated a high false-positive rate. Overall, expert raters consistently outperformed students and AI across all diagnostic categories. These findings indicate that AI has the potential to support clinicians in detecting dislodged coronal restorations with high accuracy. Meanwhile, its performance in caries detection remains moderate, and its periapical status assessment is insufficient due to high false-positive rates. Although expert raters consistently demonstrated superior diagnostic performance, AI may serve as a supplementary tool in dental diagnostics, particularly for the identification of restoration failures. Further refinement and training of AI algorithms are necessary to enhance their reliability in comprehensive dental radiographic evaluation. Not applicable. The online version contains supplementary material available at 10.1186/s12880-025-02019-y.

Tags

Topics