Machine learning algorithms enhance the accuracy of radiographic diagnosis of dental caries: a comparative study

Abstract

Objectives This study evaluated the influence of cognitive aids, including machine learning (ML) algorithms and checklists, on the diagnostic accuracy and confidence of dental students in detecting dental caries on bitewing radiographs. Methods Fifty-two third-year dental students were randomly assigned to control, ML, or checklist groups. The participants recorded their caries diagnoses (charting) on 10 bitewing radiographs and rated their confidence. Diagnostic accuracy and reliability were compared between groups for caries detection (present/absent). The inter-rater reliability for International Caries Detection and Assessment System II (ICDAS II) caries grading was assessed using weighted kappa. Participants also completed questionnaires on their perceptions of cognitive aids. Results ML group showed the highest diagnostic accuracy and confidence levels. For caries detection, the ML group achieved the highest sensitivity (79%) and diagnostic odds ratio (20.3), while the checklist group had the highest specificity (90.9%) (P