TEANet: Automated tooth extraction and arrangement with tooth-level graph spatial transformation network

Abstract

Background and objective

Orthodontic treatment is a process that involves tooth extraction and arrangement achieved via mesh-based analysis and rigid deformation of teeth. The aim is to achieve balanced relationship in physiologic and aesthetic harmony among cranial structures and occlusion. Existing automated treatment planning methods elaborated point cloud for 3D analysis and utilized Graph Convolutional Network (GCN) for feature propagation. However, these techniques were designed specifically for cases not requiring extraction therefore have limited accuracy for dentition crowding cases that require tooth extraction.

Methods

Motivated by these challenges, we propose Tooth Extraction & Arrangement Network (TEANet), an automated treatment planning method that enables tooth arrangement including cases requiring tooth extraction. The novelty we introduce is a double-coordinate classifier for tooth extraction and an adaptive graph for tooth arrangement. Our method makes the following contributions: (1) We introduce an orthodontic extraction classification for full-mouth point cloud that includes multiple tooth subsets; (2) We design a node-erasable adaptive graph to represent interactions (edges) among teeth (nodes) for feature propagation in GCN; and (3) We introduce a learning-based 3D spatial transformer model with GCN to produce tooth arrangement planning for extraction cases.

Results

We evaluated our method on three tasks in orthodontic treatment (tooth classification, extraction classification and tooth arrangement regression) with a total of 304 clinical cases. Our method achieved consistent and better performance than the state-of-the-art methods. The dataset will be released upon publication.

Conclusion

TEANet offers a novel and robust framework for orthodontic treatment planning, particularly in cases that require tooth extraction. By leveraging a double-coordinate classifier, an adaptive graph, and a learning-based spatial transformer with GCN, TEANet outperforms existing methods and provides an efficient solution for handling complex orthodontic cases.