SEGMENTATION AND MOTION ANALYSIS OF TEXTURED THREE-DIMENSIONAL SCANS OF TEETH

Abstract

Teeth movement is an important process for a dentist that helps in gauging the progress of the treatment. However, the lack of a stable reference with respect to which one could measure the teeth movement makes this a challenging problem. In this work, the Rugea are used as stable reference on which a segmentation and motion measurement of all individual teeth in the upper jaw is performed. The approach in this work is to utilize deep learning Convolutional Neural Networks (CNNs) to segment the rugae and the individual teeth. Building upon the robustness of two-dimensional image semantic segmentation, this work develops a method to convert three-dimensional textured scans of the upper palate to two-dimensional data on which the semantic segmentation is applied. Moreover, the achieved two-dimensional segmentation is pulled-back to segment the original three-dimensional textured mesh. After the segmentation of two three-dimensional scans of the same patient before and after an orthodontic treatment, an algorithm is developed to match the scans at the stable rugae region from which the three-dimensional, translation and rotation, motion of the individual teeth is computed.