Individual Graph Representation Learning for Pediatric Tooth Segmentation From Dental CBCT.

Area of Science:

• Medical Imaging
• Computer Vision
• Dental Research

Background:

• Pediatric teeth segmentation from cone-beam computed tomography (CBCT) is challenging due to age-related variations in tooth type and spatial distribution.
• Existing segmentation methods, primarily developed for adult dentition, lack the adaptability for the unique spatial distribution of pediatric teeth with individual changes (SDPTIC).
• This limitation results in suboptimal accuracy for segmenting pediatric teeth in CBCT images.

Purpose of the Study:

• To develop a novel deep learning approach for accurate pediatric teeth segmentation from CBCT data.
• To address the challenge of SDPTIC by introducing a topology structure-guided graph convolutional network (TSG-GCN).
• To improve the adaptability and accuracy of automated pediatric teeth segmentation.

Main Methods:

• A novel Topology Structure-guided Graph Convolutional Network (TSG-GCN) was developed, integrating a 3D GCN-based decoder for segmentation and a 2D decoder for dynamic adjacency matrix learning (DAML).
• SDPTIC information was captured by learning a dynamic adjacency matrix from specially-designed 2D projection labels derived from 3D CBCT data.
• A novel loss function was implemented to ensure inter-task consistency between the 3D segmentation decoder and the 2D DAML decoder, addressing potential convergence issues.

Main Results:

• The TSG-GCN approach demonstrated significant improvements in pediatric teeth segmentation accuracy compared to seven state-of-the-art methods.
• Validation on both public and multi-center datasets confirmed the effectiveness of the proposed method.
• The dynamic graph representation effectively captured SDPTIC, leading to enhanced segmentation performance.

Conclusions:

• The TSG-GCN is a highly effective method for pediatric teeth segmentation from CBCT, outperforming existing approaches.
• The novel approach successfully addresses the challenge of individual spatial variations in pediatric dentition.
• This work offers a promising advancement for automated analysis of pediatric dental CBCT scans.