UCLA School of Dentistry

Dental education programs face persistent challenges with inter-instructor variability in radiographic caries detection, which undermines the consistency and fairness of student assessment. Large class sizes and subjective interpretation of radiographic findings make it difficult to standardize grading and calibrate instructors effectively.

The school piloted Second Opinion, an FDA-cleared AI-based radiographic evaluation tool, during faculty calibration sessions and examination development for a second-year predoctoral dental student module on radiographic caries detection. Instructor diagnostic performance was evaluated with and without AI-assisted interpretation across varying carious lesion depths.

Instructors demonstrated high baseline diagnostic performance, with group average metrics exceeding 91% across sensitivity, specificity, accuracy, precision, and F1 score. AI-assisted interpretation led to modest, non-significant improvements in overall diagnostic performance and notably increased the rate of unanimous agreement among instructors, particularly for sound surfaces (E0) and early-to-moderate dentinal caries (D1/D2).

• AI-assisted interpretation can serve as a calibration tool to reduce inter-instructor variability even when baseline performance is already high.
• The greatest benefit of AI assistance was in improving consensus on borderline or subtle findings rather than boosting individual accuracy.
• Second Opinion performed comparably to the instructor group, supporting its use in case selection and assessment reliability in dental radiographic education.