Agreement between digitally predicted and clinically verified occlusal contacts in computer-aided design and...

The growing adoption of digital workflows and artificial intelligence (AI)-assisted design has revolutionized occlusal device fabrication, offering improved precision, efficiency, and reproducibility. However, the clinical reliability of digitally predicted occlusal contacts, particularly across different intensity levels and anatomic regions, remains insufficiently validated.

This study aimed to evaluate the accuracy and reliability of AI-predicted occlusal contacts in computer-aided design and computer-aided manufacture (CAD-CAM) occlusal devices fabricated through a fully digital workflow, with comparisons stratified by contact intensity and anatomic region.

A cross-sectional observational study was conducted with 53 occlusal devices (45 maxillary, 8 mandibular) fabricated using intraoral scanning, AI-based CAD software program (Medit Link v3.2.2; Medit), and 3-dimensional (3D) printing. Predicted occlusal contacts were automatically generated by the software program and compared with clinical markings obtained using articulating paper during the initial adjustment session. Agreement rates were calculated by contact intensity (low, medium, high) and anatomic region (anterior, posterior). Statistical analyses included descriptive statistics, the Friedman test for repeated-measures comparisons across intensity levels, Bonferroni-adjusted post hoc tests, and intraclass correlation coefficients [ICC(2,1)] to assess internal consistency among continuous measurements within each device (α=.05).

Agreement between digitally predicted and clinically verified contacts increased systematically with intensity, with mean coincidence rates of 59.4%, 69.7%, and 79% for low-, medium-, and high-intensity contacts, respectively. Low-intensity contacts showed the greatest dispersion, while high-intensity contacts displayed the most stable agreement values. Internal consistency among the 3 continuous agreement measurements was supported by ICC(2,1) values ranging from 0.797 to 0.867. Region-specific analyses demonstrated clearer separation among intensity levels in the anterior region, whereas medium- and high-intensity contacts overlapped more extensively in the posterior region.

AI-assisted CAD-CAM workflows demonstrated progressively higher predictive accuracy as occlusal contact intensity increased, with high-intensity contacts showing the most consistent correspondence with clinical markings. In contrast, the reduced accuracy and greater variability observed for low-intensity contacts-particularly in posterior regions-underscore the continued necessity of clinical verification during device adjustment.