Scholarlink

<p>Lumbar vertebral substructure segmentation on CT is important for anatomical quantification, surgical planning, and spine imaging biomarker research, but publicly available voxel-level annotations of lumbar subregions remain limited. Here, we present LumbarSeg-6K, a large-scale, multi-source CT dataset comprising 5,952 cropped single-vertebra lumbar volumes derived from VerSe, the CT-COLON subset of CTSpine1K, and LumASe. Each vertebra was annotated for seven anatomical substructures: vertebral body, pedicle, lamina, superior articular process, inferior articular process, transverse process, and spinous process. All data were harmonized under a unified labeling protocol using semi-automated presegmentation, manual refinement, multi-tier expert review, and quality control. Annotation reliability was assessed through an inter-observer consistency analysis, and dataset usability was further evaluated using representative source-stratified segmentation experiments. LumbarSeg-6K is intended to support reproducible research on lumbar vertebral substructure segmentation, anatomical measurement, surgical-planning algorithms, and related CT-based spine applications.</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

A large-scale CT dataset for lumbar vertebral substructure segmentation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Fangzheng Xu,
  • Qingzhi Xiang,
  • Xiao Xia,
  • Jianwen Fu,
  • Fuping Li,
  • Longhao Yang,
  • Shaobo Cheng,
  • Yan Yu,
  • Liming Cheng

摘要

Lumbar vertebral substructure segmentation on CT is important for anatomical quantification, surgical planning, and spine imaging biomarker research, but publicly available voxel-level annotations of lumbar subregions remain limited. Here, we present LumbarSeg-6K, a large-scale, multi-source CT dataset comprising 5,952 cropped single-vertebra lumbar volumes derived from VerSe, the CT-COLON subset of CTSpine1K, and LumASe. Each vertebra was annotated for seven anatomical substructures: vertebral body, pedicle, lamina, superior articular process, inferior articular process, transverse process, and spinous process. All data were harmonized under a unified labeling protocol using semi-automated presegmentation, manual refinement, multi-tier expert review, and quality control. Annotation reliability was assessed through an inter-observer consistency analysis, and dataset usability was further evaluated using representative source-stratified segmentation experiments. LumbarSeg-6K is intended to support reproducible research on lumbar vertebral substructure segmentation, anatomical measurement, surgical-planning algorithms, and related CT-based spine applications.