Artificial Intelligence-Supported Performance-Based Assessment in Oral Radiology Education: A Constructive Alignment Perspective

Background/Objectives: This study aimed to examine the educational implications of an AI-supported performance-based assessment platform in oral radiology education within a constructively aligned framework. Methods: A total of 266 fourth-year dental students evaluated five panoramic radiographs using an AI-supported lesion detection platform with a predefined detection confidence threshold of 40%. Student performance was assessed using a confusion matrix framework, and precision and sensitivity values were calculated. Receiver operating characteristic (ROC) analysis was conducted to explore the relationship between region selection frequency and AI-supported diagnostic performance. Group comparisons were performed using the Mann–Whitney U test. Results: ROC analysis demonstrated a statistically significant but weak inverse association between region selection frequency and AI-supported diagnostic performance (AUC = 0.414, 95% CI: 0.342–0.486, p = 0.018). Students exceeding the cutoff of 8.5 marked regions demonstrated significantly lower performance scores than those at or below the cutoff (p < 0.001). No significant correlation was observed between AI-supported performance scores and traditional summative examination outcomes. Conclusions: AI-supported performance-based assessment demonstrated value in measuring applied diagnostic reasoning distinct from traditional written examinations. Within a constructively aligned framework, such tools may serve as complementary strategies for competency-oriented assessment in dental education.