Scholarlink

Tooth CBCT instance segmentation is a foundational step in advancing digital dental health systems, with the precision of tooth segmentation playing a critical role in accurate medical diagnosis. Deep learning based methods face significant challenges in accurately locating and classifying teeth in 3D dental images, especially given the complex anatomical structures present in CBCT scans. Additionally, manually identifying and labeling each tooth from these scans is an extremely time-consuming and burdensome task for medical professionals. To address these issues, in this paper, we propose a two-stage semi-supervised method for tooth instance segmentation, along with innovative data preprocessing. Our method achieves a distinguished fourth place on the validation set of the “MICCAI STS 2024 Challenge Task 2”, outperforming other mainstream semi-supervised methods. Code for this implementation is made available at https://github.com/magic-fortune/MICCAI2024 .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Two-Stage Approach for 3D Dental Tooth Instance Segmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yuhan Chen,
  • Chunshi Wang,
  • Bin Zhao

摘要

Tooth CBCT instance segmentation is a foundational step in advancing digital dental health systems, with the precision of tooth segmentation playing a critical role in accurate medical diagnosis. Deep learning based methods face significant challenges in accurately locating and classifying teeth in 3D dental images, especially given the complex anatomical structures present in CBCT scans. Additionally, manually identifying and labeling each tooth from these scans is an extremely time-consuming and burdensome task for medical professionals. To address these issues, in this paper, we propose a two-stage semi-supervised method for tooth instance segmentation, along with innovative data preprocessing. Our method achieves a distinguished fourth place on the validation set of the “MICCAI STS 2024 Challenge Task 2”, outperforming other mainstream semi-supervised methods. Code for this implementation is made available at https://github.com/magic-fortune/MICCAI2024 .