Effectiveness of human-artificial intelligence collaboration for interproximal caries detection using bitewing radiographs: a systematic review.

OBJECTIVE: To evaluate the effectiveness of human-artificial intelligence (AI) collaboration in detecting interproximal caries on bitewing (BW) radiographs compared with unaided human interpretation. STUDY DESIGN: A systematic search of six databases was performed. Studies comparing the diagnostic accuracy of AI-assisted and unaided human interpretation of interproximal caries on BW radiographs were included. Due to clinical and methodological heterogeneity, results were synthesized qualitatively. Risk of bias (RoB) was assessed using the QUADAS-2 tool. RESULTS: Seven studies met the inclusion criteria, including 6,400 proximal tooth surfaces (total of 1802 BWs). Observers were dental students, general dentists and postgraduate clinicians. AI assistance increased sensitivity for proximal caries detection, particularly for early lesions, while specificity was preserved or minimally affected (≤3.4%). Three studies reported improved interobserver agreement with AI support. Changes in reading time were modest and inconsistent. RoB was low to moderate, with most concerns related to patient selection and index test reporting. CONCLUSIONS: Human-AI collaboration improves proximal caries detection on BW radiographs by increasing diagnostic sensitivity while maintaining specificity, supporting the use of AI as a clinician-centered decision support tool.