Abstract:This paper presents our approach to scaling the nnU-Net framework for multi-structure segmentation on Cone Beam Computed Tomography (CBCT) images, specifically in the scope of the ToothFairy2 Challenge. We leveraged the nnU-Net ResEnc L model, introducing key modifications to patch size, network topology, and data augmentation strategies to address the unique challenges of dental CBCT imaging. Our method achieved a mean Dice coefficient of 0.9253 and HD95 of 18.472 on the test set, securing a mean rank of 4.6 and with it the first place in the ToothFairy2 challenge. The source code is publicly available, encouraging further research and development in the field.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to achieve accurate segmentation of multiple structures in cone - beam computed tomography (CBCT) images, especially for applications on complex anatomical structures such as teeth, jaws and nerves**. Specifically, the paper focuses on how to improve the nnU - Net framework to meet the unique challenges of dental CBCT images and achieve excellent performance in the ToothFairy2 competition. ### Problem Background 1. **Importance of Accurate Segmentation** - Accurate segmentation of dental structures such as teeth, jaws and nerves is crucial for dental diagnosis, treatment planning and surgical navigation. - Accurate segmentation can automate the analysis of dental images, helping to identify dental diseases, plan implant surgeries and navigate complex anatomical areas. 2. **Existing Challenges** - The high variability of dental anatomical structures, the proximity of key structures and the need for precise positioning make it very important to develop robust segmentation algorithms. - The lack of large - scale publicly - labeled datasets was once a bottleneck in the progress of this field, but this situation has improved with the emergence of the ToothFairy2 competition. ### Research Objectives The goal of this paper is to optimize the multi - structure segmentation task in CBCT images by improving the nnU - Net framework, specifically including: - **Adjusting Network Configuration**: Modify the patch size, network topology and data augmentation strategy to adapt to the characteristics of dental CBCT images. - **Improving Segmentation Accuracy**: Verify different improvement measures through experiments, and finally achieve a higher Dice coefficient and a lower Hausdorff distance (HD95). - **Addressing Specific Challenges**: Solve problems such as left - right distinction, training time, data augmentation, etc., to ensure the robustness and generalization ability of the model in practical applications. ### Main Contributions - **Performance Improvement**: Through a series of improvements, the model achieved an average Dice coefficient of 0.9253 and an HD95 of 18.472 in the ToothFairy2 competition, winning first place. - **Method Innovation**: Introduced a larger patch size, a deeper network topology, and optimized the data augmentation strategy. - **Code Publication**: All source codes have been made public to encourage further research and development. In conclusion, this paper aims to solve the key problems in dental CBCT image segmentation by improving the nnU - Net framework and promote the technological progress in this field.

Scaling nnU-Net for CBCT Segmentation

TeethGNN: Semantic 3D Teeth Segmentation with Graph Neural Networks

[Tooth segmentation and identification on cone-beam computed tomography with convolutional neural network based on spatial embedding information]

Fully automatic segmentation of craniomaxillofacial CT scans for computer-assisted orthognathic surgery planning using the nnU-Net framework

MRCM‐UCTransNet: Automatic and Accurate 3D Tooth Segmentation Network From Cone‐Beam CT Images

FDNet: Feature Decoupled Segmentation Network for Tooth CBCT Image

Teeth Localization and Lesion Segmentation in CBCT Images using SpatialConfiguration-Net and U-Net

Improving performance of deep learning models using 3.5D U-Net via majority voting for tooth segmentation on cone beam computed tomography

Automatic Dental Root CBCT Image Segmentation Based on CNN and Level Set Method

DentalSegmentator: robust deep learning-based CBCT image segmentation

Center-Sensitive and Boundary-Aware Tooth Instance Segmentation and Classification from Cone-Beam CT.

CTooth: A Fully Annotated 3D Dataset and Benchmark for Tooth Volume Segmentation on Cone Beam Computed Tomography Images

Inferior Alveolar Nerve Segmentation in CBCT images using Connectivity-Based Selective Re-training

Multi-scale Volumetric ConvNet with Nested Residual Connections for Segmentation of Anterior Cranial Base.

Refined Individual Tooth Segmentation from Cone Beam CT Images

Refined tooth and pulp segmentation using U-Net in CBCT image

STSN-Net: Simultaneous Tooth Segmentation and Numbering Method in Crowded Environments with Deep Learning

Root Canal Segmentation in CBCT Images by 3D U-Net with Global and Local Combination Loss

DenUnet: enhancing dental image segmentation through edge and body fusion

A Multi-Stage Framework for 3D Individual Tooth Segmentation in Dental CBCT

Tooth Instance Segmentation on Panoramic Dental Radiographs Using U-Nets and Morphological Processing