Multi-Structure Segmentation in CBCT Volumes

Abstract

This paper is devoted to the processing of medical 3D images for the task of teeth and jaw structures segmentation. Two modern neural networks are used as the main architecture: 3D U-Net and Swin UNETR, which are adapted to work with multiclass medical data. The processing steps include image preprocessing, model training, and evaluation of their accuracy. Special attention is paid to the data preprocessing process, including intensity normalization, spatial resolution alignment, and label recoding. The study uses the Tooth Fairy 2 dataset, which contains annotated 3D images of dental and jaw structures. Segmentation accuracy is estimated using Dice and HD 95 metrics, which allows you to judge the quality of recognition of complex anatomical structures and compare the performance of both models.

More by Aleksandr Sinitca

• Seborrheic dermatitis: Identification of key disease factors and treatment with a topical solution of betamethasone 0.05% + salicylic acid 2%Meditsinskiy sovet = Medical Council · 2026
• Reconstruction of long-term complex system dynamics from short-term fragments of observational records using high-order Kalman filteringChaos An Interdisciplinary Journal of Nonlinear Science · 2026
• Microscopy Image Dataset of pulmonary vessels for Quantitative assessment of fibrosisScientific Data · 2026
• Microscopy Image Dataset of pulmonary vessels for Deep Learning-Based Quantitative assessment of fibrosisFigshare · 2026