Random and systematic errors associated with 3D-printed models.

Objectives: To evaluate the reliability and validity of 3D models printed at 3 different resolutions.
Methods: 30 sets of orthodontically treated plaster models were selected; 15 from patients with and 15 from patients without extracted premolars. The models were scanned using Trios by 3Shape, saved as STL files, and prepared for printing with Meshmixer and 3D Builder. Each model was printed using a grey-colored resin with a MoonRay printer at 20, 50, and 100 um resolutions. Two blinded examiners measured the stone and printed models using the ABO Cast-Radiograph Evaluation (CREval) system; points were scored for 8 components, including alignment, marginal ridges, buccolingual inclination, occlusal contacts, occlusal relationships, overjet, and interproximal contacts.
Results: Replicate analyses showed intra-observer random method errors ranging from 0.4 (occlusal relations) to 1.4 (alignment) points. There were no statistically significant systematic inter-observer errors. With the exception of marginal ridges, which were rated significantly lower on the 20 mm (p=.008) and 60 mm (p=.023) printed models than on the stone models, there were no statistically significant difference in component scores between the printed and stone models. Median total scores ranged from 22 to 24 points, with no statistically significant differences between the printed and stone models. Based on the ranges between the 10^th and 90^th percentiles, the stone models were slightly less variable than the 20 mm models, which were less variable than the 50 mm and 100 mm models
Conclusions: CREval scores measured on models printed at 20, 50 or 100 mm resolution are similar to scores measured on the stone model.