AI-Based Clinical Decision Support Systems for Secondary Caries Detection and Staging on Bitewings: A Multi-Algorithm Comparison.

INTRODUCTION

Radiographic detection of caries lesions adjacent to restorations is challenging because of the limitations of two-dimensional imaging and difficulties distinguishing true lesions from restorative or anatomical radiolucencies. Artificial intelligence (AI)-based clinical decision support systems (CDSSs) have been introduced to assist radiographic interpretation; however, different AI tools may yield variable diagnostic outputs, and their comparative performance remains unclear.

OBJECTIVE

To compare the diagnostic performance of commercial and experimental AI algorithms for detecting and staging secondary caries lesions on bitewings.

METHODS

This cross-sectional diagnostic accuracy study included 200 anonymized bitewings comprising 885 restored tooth surfaces. A consensus-based reference standard classified each restored surface as non-carious, secondary caries, or indeterminate. Secondary caries lesions were further staged as enamel-stage or dentin-stage. Five commercial AI systems (Second Opinion®, CranioCatch, Diagnocat, DIO Inteligência, and Align™ X-ray Insights) and three experimental systems based on Mask R-CNN and Mask DINO architectures were tested. Diagnostic performance was assessed using sensitivity, specificity, accuracy, positive predictive value, and negative predictive value (95% confidence intervals). Comparisons were performed using generalized estimating equations adjusted for clustered data.

RESULTS

Specificity was high across all systems (0.957-0.986), whereas sensitivity was moderate (0.327-0.487), reflecting missed detections of secondary caries lesions, particularly dentin-stage lesions. Accuracy ranged from 0.882 to 0.917, with no significant differences among the evaluated systems (p ≥ 0.05). Similar patterns were observed in the additional analysis restricted to dentin-stage lesions. Misclassifications between non-carious and secondary caries surfaces were frequently associated with radiographic overlap, restoration-related radiolucencies, and cervical artifacts.

CONCLUSIONS

AI algorithms, regardless of architecture or commercial status, showed similar diagnostic capabilities and a conservative diagnostic profile, favoring specificity over sensitivity. Improvements in dataset diversity, reference-standard quality, and model explainability may further enhance the reliability of AI-assisted secondary caries detection.