Scholarlink

<p>3D tooth segmentation is a pivotal research area in computer vision and medical image processing, with applications spanning dental diagnosis, treatment planning, and orthodontic modeling . Existing methods often struggle to accurately model complex tooth geometries and capture fine-grained details, thereby limiting segmentation accuracy and robustness. To address these challenges, we propose GeoNet, a 3D tooth segmentation network enhanced by geometric features. GeoNet integrates multimodal geometric features, including 3D coordinates and normal vectors, to capture both local and global geometric properties of tooth point clouds. By designing an Adaptive Attention-Distance Fusion Module and a Point Relationship Fusion Module in the coordinate domain, and a Gaussian Channel Aggregation Model in the normal vector domain, GeoNet effectively models the overall geometry and detailed boundary information of teeth. Here we show that GeoNet achieves state-of-the-art performance on two publicly available datasets, with Dice Similarity Coefficients (DSC) of 93.95% and 88.17% on the Teeth3DS and TeethIOS datasets, respectively. Our work demonstrates the importance of geometric feature integration in 3D tooth segmentation, providing a robust solution with potential implications for digital oral healthcare. The code is available at <a href="https://github.com/khywc/GeoNet.git">https://github.com/khywc/GeoNet.git</a>.</p>

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

Geonet: enhanced 3D tooth segmentation via geometric feature integration

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Hong-an Li,
  • Chen Wu

摘要

3D tooth segmentation is a pivotal research area in computer vision and medical image processing, with applications spanning dental diagnosis, treatment planning, and orthodontic modeling . Existing methods often struggle to accurately model complex tooth geometries and capture fine-grained details, thereby limiting segmentation accuracy and robustness. To address these challenges, we propose GeoNet, a 3D tooth segmentation network enhanced by geometric features. GeoNet integrates multimodal geometric features, including 3D coordinates and normal vectors, to capture both local and global geometric properties of tooth point clouds. By designing an Adaptive Attention-Distance Fusion Module and a Point Relationship Fusion Module in the coordinate domain, and a Gaussian Channel Aggregation Model in the normal vector domain, GeoNet effectively models the overall geometry and detailed boundary information of teeth. Here we show that GeoNet achieves state-of-the-art performance on two publicly available datasets, with Dice Similarity Coefficients (DSC) of 93.95% and 88.17% on the Teeth3DS and TeethIOS datasets, respectively. Our work demonstrates the importance of geometric feature integration in 3D tooth segmentation, providing a robust solution with potential implications for digital oral healthcare. The code is available at https://github.com/khywc/GeoNet.git.