Scholarlink

<p>Automatic tooth segmentation has improved the accuracy and efficiency of clinical diagnosis. It also provides a technical basis for patient-specific treatment planning. Although great progress has been made in this field, existing methods still have difficulty in accurately segmenting tooth structures in dental CBCT images with metal artifacts and blurred boundaries. To address these challenges, this paper proposes RFFA-Net, a unified encoder-decoder network for 3D tooth segmentation in dental CBCT images. First, a Dense Mamba (DM) module is introduced into the encoder. It combines recursive filtering with dense connections to enhance feature extraction and reduce the interference of metal artifacts. Second, a Dynamic Multi-scale Feature Fusion (DMFF) module is designed at the network bottleneck. It adaptively aggregates multi-scale features from the encoder to better capture the irregular shapes of teeth. Finally, an Interleaved Channel-Spatial Attention (ICSA) module is introduced to address blurred tooth boundaries. It uses a gating mechanism to dynamically adjust the weights of channel and spatial attention, which helps fuse global semantic information with local details. Experiments were conducted on two commonly used dental CBCT datasets. The results show that RFFA-Net outperforms existing state-of-the-art methods in overall segmentation performance, demonstrating its effectiveness and robustness. It also achieves more stable and accurate segmentation results in CBCT images with metal artifacts and blurred boundaries.</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

RFFA-Net: Recursive filtering and feature aggregation driven network for tooth segmentation in CBCT image

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月25日

查看全文
  • Yu Ao,
  • Hongze Han,
  • Yuqin Li,
  • Yu Miao,
  • Weili Shi

摘要

Automatic tooth segmentation has improved the accuracy and efficiency of clinical diagnosis. It also provides a technical basis for patient-specific treatment planning. Although great progress has been made in this field, existing methods still have difficulty in accurately segmenting tooth structures in dental CBCT images with metal artifacts and blurred boundaries. To address these challenges, this paper proposes RFFA-Net, a unified encoder-decoder network for 3D tooth segmentation in dental CBCT images. First, a Dense Mamba (DM) module is introduced into the encoder. It combines recursive filtering with dense connections to enhance feature extraction and reduce the interference of metal artifacts. Second, a Dynamic Multi-scale Feature Fusion (DMFF) module is designed at the network bottleneck. It adaptively aggregates multi-scale features from the encoder to better capture the irregular shapes of teeth. Finally, an Interleaved Channel-Spatial Attention (ICSA) module is introduced to address blurred tooth boundaries. It uses a gating mechanism to dynamically adjust the weights of channel and spatial attention, which helps fuse global semantic information with local details. Experiments were conducted on two commonly used dental CBCT datasets. The results show that RFFA-Net outperforms existing state-of-the-art methods in overall segmentation performance, demonstrating its effectiveness and robustness. It also achieves more stable and accurate segmentation results in CBCT images with metal artifacts and blurred boundaries.