UCLA

Dental caries remain highly prevalent worldwide, underscoring the need for more accurate and efficient diagnostic methods in dental education. Traditional radiographic interpretation, though essential, often suffers from variability and limited sensitivity, prompting exploration of artificial intelligence (AI) as a supportive tool. This study evaluated whether an AI platform (Second Opinion®) could enhance radiographic caries detection in a dental school setting by comparing its diagnostic performance to that of second-year dental students and by assessing its impact on faculty accuracy and consensus. AI performance was compared with caries detection exam results from second-year dental students in the 2023 cohort (Cohort 1). The same exam was later repurposed as a self-assessment quiz for the 2024 cohort (Cohort 2), and their performance was compared with that of the AI. Subsequently, a new AI-assisted caries detection exam was developed for Cohort 2, incorporating ≥75% faculty agreement as the gold standard for lesion classification. Diagnostic metrics (sensitivity, specificity, accuracy, precision, and F1 score) were calculated for students, AI, and faculty members—both without and with AI annotations. The AI platform outperformed both student cohorts, achieving higher sensitivity (89.5%) and accuracy (93.6%). Cohort 2 demonstrated significant improvement after structured self-assessment, with accuracy increasing from 40.4% in the self-assessment to 61.7% in the caries detection exam. Notably, Cohort 2 surpassed Cohort 1’s first attempt pass rate (96.25% vs. 55.8%). Among faculty, three of four members showed increased sensitivity and accuracy with AI annotations, and unanimous (4/4) consensus improved from 73.33% to 86.67%. The AI platform consistently exhibited higher diagnostic performance than second-year dental students, reinforcing its potential as a reliable adjunct in caries detection training. Moreover, AI-assisted workflows streamlined exam development and improved faculty consensus. While AI provides diagnostic support, it should complement—rather than replace—clinician-led education and judgment. With careful curriculum integration, AI holds substantial potential for elevating diagnostic standards and refining dental training.