Tooth Alignment Network Based on Landmark Constraints and Hierarchical Graph Structure.

Area of Science:

• Dentistry
• Biomedical Engineering
• Artificial Intelligence

Background:

• Automatic tooth alignment prediction is crucial for efficient orthodontic treatment planning and aligner design.
• Current methods often lack clinical interpretability as they directly regress tooth motion.
• Subjectivity and reliance on dentist experience impact the quality of alignment targets.

Purpose of the Study:

• To develop a novel neural network for predicting tooth alignment targets with enhanced clinical interpretability.
• To leverage anatomical landmark constraints to improve the accuracy of automated tooth alignment.
• To provide a knowledge-driven solution for less experienced dentists in orthodontic planning.

Main Methods:

• A novel tooth alignment neural network integrating tooth landmark constraints and a hierarchical graph structure (jaw-tooth-landmark) was developed.
• Tooth landmarks were detected and used to construct a graph characterizing relationships between teeth and landmarks.
• Landmark constraints were defined to guide the network in learning normal occlusion and predicting tooth transformations.

Main Results:

• The proposed method demonstrates improved results in tooth alignment target prediction compared to existing approaches.
• The architecture, incorporating tooth data and landmark constraints, enhances explainability in clinical tooth alignments.
• The network successfully predicts rigid transformations for each tooth during the alignment process.

Conclusions:

• The novel neural network effectively predicts tooth alignment targets using anatomical landmark constraints and a hierarchical graph structure.
• This approach offers improved accuracy and clinical interpretability, benefiting both experienced and inexperienced dental practitioners.
• The method represents a significant advancement in automated orthodontic treatment planning and aligner design.