LumVBCanSeg: A Lumbar Vertebral Body Cancellous Bone Segmentation Dataset

To address the lumbar vertebral body cancellous bone segmentation problem, we propose a new dataset called LumVBCanSeg. The entire dataset consists of 185 lumbar CT scans acquired using multiple CT scanners, including the two manufacturers(Philips, Siemens). All data were obtained at ShengJing Hospital of China Medical University. Cases involving vertebral fractures, metal implants, bone tumors and foreign materials were omitted. All data were resampled to an isotropic resolution of 1mm × 1mm × 1mm. Each lumbar 3D CT image has a corresponding segmentation mask annotated at voxel level using the medical image annotation package Pair. Annotations cover five lumbar vertebral body cancellous bones from L1 to L5, with labels ranging from 1 to 5. All annotations were performed by 3 physicians followed by further refinement, dismissal or approval by one physician with more than 30 years of experience in lumbar imaging.为解决腰椎椎体松质骨分割难题，我们构建了一款名为LumVBCanSeg的新型数据集。该数据集共包含185例腰椎CT扫描影像，采集设备涵盖飞利浦(Philips)、西门子(Siemens)两大品牌的多款CT扫描仪。所有数据均采集自中国医科大学附属盛京医院。本数据集剔除了包含椎体骨折、金属植入物、骨肿瘤及异物的病例。所有数据均被重采样至各向同性的1mm×1mm×1mm分辨率。每一例腰椎三维CT影像均配有基于医学影像标注工具Pair完成的体素级分割掩码。标注覆盖L1至L5共5个腰椎椎体的松质骨，标签取值范围为1至5。所有标注工作先由3名医师完成，随后由1名拥有30余年腰椎影像诊断经验的医师进行细化修正、驳回或最终确认。