Automated virtual treatment planning in orthodontics: modeling and algorithms.

Abstract

Computer-based virtual treatment planning and simulation has become increasingly important in orthodontics due to its potential to lower costs and provide better treatment outcomes. However, its clinical use is currently limited by the need for significant human intervention in segmenting the dental arch (a 3-dimensional surface mesh) into individual tooth objects and then (re)aligning these tooth objects on the arch to satisfy prescribed treatment criteria. The goals of this thesis are to develop modeling techniques, computational algorithms, and associated software to automate key components of the treatment process, including dental arch segmentation, feature identification, and tooth alignment, thereby making the advantages of virtual treatment planning and simulation available to the vast majority of orthodontic patients. Towards this end, this thesis makes the following contributions: First, an algorithm is designed to segment a dental arch into individual tooth objects (sub-meshes) that can be manipulated downstream in the treatment planning and simulation pipeline. The algorithm is largely automated and requires human intervention in very difficult or unusual cases only. The algorithm avoids the limitations of known approaches to dental segmentation by dividing the segmentation task into two key subtasks—separation of the gums from the teeth and separation of the teeth from one another—and also incorporates a new technique to repair defects in the gumline caused by noise in the data. Second, algorithms are developed to automatically identify suitable features on the surfaces of individual tooth objects for use in the alignment stage. These include intrinsic features such as cusps, incisal edges, grooves, marginal ridges, and the occlusal surface boundary, as well as derived features such as the archform and occlusal plane. A key technique underlying some of these algorithms is the identification and clustering of vertices of high curvature on certain planar cross-sections of the tooth objects and stitching these clusters together to extract the features of interest. Third, an algorithm is developed to automatically align the tooth objects on two opposing arches so that the best possible intra-arch and inter-arch occlusion (i.e., contact relationship) of teeth is achieved, while respecting certain natural dental constraints. The alignment process is modeled as a simulation-based optimization of certain configurations of constraints defined with respect to the tooth features identified previously. The simulation is based on a spring-mass model where the teeth gradually move to their final positions under the influence of forces exerted by (hypothetical) springs attached to the features. Finally, all of the algorithms developed in this thesis have been implemented and incorporated into a comprehensive software tool. This tool has been used by dental practitioners to evaluate and validate the algorithms on clinical data. These experimental studies have demonstrated the efficacy of the algorithms for orthodontic treatment planning.