Scholarlink

<p>Age-related macular degeneration (AMD) and diabetic macular edema (DME) are vision-threatening pathologies for which optical coherence tomography (OCT) provides high-resolution three-dimensional imaging, facilitating comprehensive diagnostic evaluation. Three-dimensional (3D) visualization and precise 3D segmentation of lesions enable accurate assessment of morphology, dimensions, and spatial relationships, thereby enhancing clinical analysis and disease management. However, the lack of a 3D dataset for AMD and DME significantly limits the reliability, robustness, and applicability of deep learning-based 3D segmentation techniques in this field. Here, we present an OCT dataset comprising 224 volumetric images, including 122 for AMD and 102 for DME, annotated with pigment epithelial detachment and intraretinal fluid. We propose a novel 3D segmentation network based on the BiFormer Block, which employs Bi-Level Routing Attention to capture local context and long-range dependencies. Experiments demonstrated that the proposed dataset and network will facilitate the exploration and validation of novel 3D segmentation methods for AMD and DME.</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

Comprehensive 3D Optical Coherence Tomography Dataset for AMD and DME: Facilitating Deep-Learning-Based 3D Segmentation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月10日

查看全文
  • Wenjing Huang,
  • Lang Qin,
  • Mingxin Xu,
  • Hao Zheng,
  • Yuxing Gan,
  • Shaoyu Pei,
  • Renxiong Wu,
  • Yong Liu,
  • Jie Zhong,
  • Guangming Ni

摘要

Age-related macular degeneration (AMD) and diabetic macular edema (DME) are vision-threatening pathologies for which optical coherence tomography (OCT) provides high-resolution three-dimensional imaging, facilitating comprehensive diagnostic evaluation. Three-dimensional (3D) visualization and precise 3D segmentation of lesions enable accurate assessment of morphology, dimensions, and spatial relationships, thereby enhancing clinical analysis and disease management. However, the lack of a 3D dataset for AMD and DME significantly limits the reliability, robustness, and applicability of deep learning-based 3D segmentation techniques in this field. Here, we present an OCT dataset comprising 224 volumetric images, including 122 for AMD and 102 for DME, annotated with pigment epithelial detachment and intraretinal fluid. We propose a novel 3D segmentation network based on the BiFormer Block, which employs Bi-Level Routing Attention to capture local context and long-range dependencies. Experiments demonstrated that the proposed dataset and network will facilitate the exploration and validation of novel 3D segmentation methods for AMD and DME.