Radiographic burden of dentin caries in first permanent molars: an international caries classification and management system-based panoramic radiographic study with artificial intelligence-assisted detection

First permanent molars (FPMs) are highly susceptible to dental caries, and radiographic assessment may provide important information about the extent and severity of dentin involvement. This study investigated the radiographic burden of dentin caries in FPMs using an International Caries Classification and Management System (ICCMS™)-based radiographic assessment and evaluated the agreement and detection performance of an artificial intelligence (AI)-assisted system. This retrospective cross-sectional study screened 2,627 digital panoramic radiographs (DPRs) of children aged 6–12 years; 1,000 met the inclusion criteria. A total of 4,000 FPMs were evaluated using ICCMS™ radiographic criteria. The prevalence of dentin caries (RA3+) was analyzed by age and tooth location. Group comparisons were conducted using appropriate statistical methods, and AI-assisted detection was compared with examiner-based consensus classifications using detection performance metrics and Cohen’s kappa. The child-level prevalence of dentin caries (RA3+) was 56.8% and was significantly higher in the late mixed-dentition group than in the early group (66.4% vs. 39.3%). At the tooth level, RA3+ lesions accounted for 30.7% of all molars and were more prevalent in mandibular than in maxillary molars (37.6% vs. 23.9%). AI-assisted detection showed fair agreement with examiner consensus classifications (κ = 0.389), with 78.9% accuracy, 32.1% sensitivity, and 99.6% specificity. AI detection rates increased markedly across examiner-defined lesion severity categories, reaching 80.2% for RC5 and 93.4% for RC6 lesions. A substantial radiographic burden of dentin caries was identified in FPMs in this clinic-based sample of school-aged children. Although AI-assisted detection showed high specificity and improved detection of advanced dentin lesions (RC5–RC6), its low sensitivity for RA3+ lesions indicates that most examiner-defined dentin lesions were missed. AI should therefore not be used as a standalone screening or triage tool and should be regarded only as an adjunct to clinical examination and clinician-led radiographic assessment.