Scholarlink

This work presents the contribution of the OS team to the ToothFairy challenge organized during MICCAI 2023. The challenge focuses on creating a comprehensive 3D dataset of maxillofacial structures, including both sparse and partially dense labels, to enhance the efficiency of automated Inferior Alveolar Nerve (IAN) segmentation. However, sparse labels do not fully represent the IAN structure. To overcome the adverse effects associated with sparse labeling, we generate weak annotations using semi-supervised learning and refine them into strong annotations via manual labeling. Finally, model performance is improved by training with a combination of dense labels and strong annotations. Our method is quantitatively evaluated on the ToothFairy verification cases, achieving a Dice similarity coefficient (DSC) of 0.7354 and a 95% Hausdorff distance (HD95) of 7.85.

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

Weakly-Supervised Convolutional Neural Networks for Inferior Alveolar Nerve Segmentation in CBCT Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jae Hwan Han,
  • Wan Kim,
  • Hong-Gi Ahn

摘要

This work presents the contribution of the OS team to the ToothFairy challenge organized during MICCAI 2023. The challenge focuses on creating a comprehensive 3D dataset of maxillofacial structures, including both sparse and partially dense labels, to enhance the efficiency of automated Inferior Alveolar Nerve (IAN) segmentation. However, sparse labels do not fully represent the IAN structure. To overcome the adverse effects associated with sparse labeling, we generate weak annotations using semi-supervised learning and refine them into strong annotations via manual labeling. Finally, model performance is improved by training with a combination of dense labels and strong annotations. Our method is quantitatively evaluated on the ToothFairy verification cases, achieving a Dice similarity coefficient (DSC) of 0.7354 and a 95% Hausdorff distance (HD95) of 7.85.