Evaluation of software developed for automated segmentation of digital dental models.

University of Minnesota M.S. thesis. March 2012. Major: Dentistry. Advisor: Brent Larson, DDS. 1 computer file (PDF); vi, 50 pages, appendices p. 33.Introduction: The use of digital models for simulation of orthodontic treatment is currently hampered by time-consuming manual segmentation processes which rely on the user to manually define tooth boundaries for separation into individual tooth objects. A new software tool has been developed that almost completely automates the segmentation process requiring little input from the user. The software combines novel methods to analyze the 3D mesh curvatures that make up a digital model, identify the gingival margin for separation of the teeth from the gingival tissues, repair the gingival margins, and then identify the high curvature vertices on the gingival margins to determine segmentation boundaries between teeth. Methods: Thirty pretreatment models (15 upper, 15 lower) with varying amounts of crowding and spacing were digitally scanned using an R700 digital model scanner (3Shape, Copenhagen, Denmark). Each digital model was then segmented using the newly developed software tool and then visually evaluated for segmentation accuracy. To determine inter- and intra- operator repeatability, two different examiners participated in this study. To use as a comparison, the same 30 digital models were also segmented using 3Shape’s OrthoAnalyzer® and evaluated for segmentation accuracy in the same manner. Results: Of the 387 possible separators on the 30 models tested, 380 were correctly placed by the new software tool (98.2% accuracy, 5 omitted separators, 2 misplaced separators). On a tooth by tooth basis, a total of 77 errors were observed on 417 teeth; 67 being minor errors that did not affect the segmentation of the models and 10 being major errors that did affect the segmentation of the models (83.0% success in accurate tooth anatomy reproduction after segmentation). Both inter- and intra- operator repeatability was high. Using OrthoAnalyzer® to segment the same 30 digital models, all 387 separators were correctly placed and there were 14 tooth anatomy reproduction errors. Conclusions: Initial results indicate that mesh segmentation algorithms can be developed to accurately segment digital dental models in most situations while requiring little user time. Further development of these algorithms could provide the orthodontic practitioner fast, easy treatment simulation for help in treatment planning