Artificial intelligence for radiographic caries detection in dental education: a systematic review with functional...

Introduction Dental caries remains a major global oral health burden, and accurate radiographic detection of carious lesions is a core competency in dental education. Artificial intelligence (AI), particularly machine learning and deep learning systems, has shown promising diagnostic performance in caries detection on dental radiographs. However, evidence on how AI influences dental students' diagnostic performance and learning remains fragmented and methodologically heterogeneous. This systematic review aimed to synthesize student-based evidence on the role of AI in radiographic caries detection within dental education. Methods A systematic review with functional meta-synthesis was conducted following PRISMA 2020 guidance and a preregistered PROSPERO protocol. PubMed, Embase, Scopus, and SciELO were searched up to April 30 2026. Eligible studies involved undergraduate or postgraduate dental students performing radiographic caries detection or classification tasks using AI-assisted decision support, AI-based training, or AI-system versus student benchmarking. Risk of bias was assessed using an adapted QUADAS-2 framework, and certainty of evidence was evaluated using GRADE. Quantitative findings were interpreted descriptively because methodological heterogeneity precluded pooled effect estimation. Results Nine studies involving 537 dental students were included. Functional synthesis identified three roles of AI: AI-assisted decision support, AI-based training scaffold, and AI-system performance benchmark. AI-assisted interpretation was associated with improved lesion detection in some studies, particularly sensitivity, although specificity and overall diagnostic balance varied. Pre-post AI-based training studies suggested short-term improvements, but effects were inconsistent and not clearly superior to alternative educational approaches. Benchmarking studies showed that AI systems performed comparably to or better than students in some settings, but several included very small student comparator groups. Substantial heterogeneity was observed in study design, imaging modality, AI systems, reference standards, diagnostic thresholds, and outcome reporting. Risk of bias ranged from low to high, although most studies were judged as low or moderate risk. Certainty of evidence was very low across exploratory comparative outcome domains. Conclusion AI may support radiographic caries detection education, particularly as decision support or structured training under controlled conditions. However, current evidence mainly reflects short-term performance effects and does not establish durable learning gains, educational superiority, or transferability to clinical practice.