CTSpine1K

为了推进脊柱图像分析的研究，我们在此展示了一个大规模的综合数据集：CTSpine1K。为了构建一个复制实际外观变化的综合脊柱数据集，我们从以下四个开源资源中挑选出 CTSpine1K，总计 1,005 个不同外观变化的 CT 卷（超过 500,000 个标记切片和超过 11,000 个椎骨）。
*科隆。该子数据集来自与 CT 结肠造影试验相关的 CT COLONOGRAPHY 数据集12。我们为我们的数据集随机选择每个患者的两个位置之一（我们打开用于选择它们的代码，dicom2nii.py），它们具有相似的信息。有 825 次 CT 扫描，采用医学数字成像和通信 (DICOM) 格式。
*HNSCC-3DCT-RT。该子数据集包含使用西门子 16 层 CT 扫描仪和头部和--颈部鳞状细胞癌 (HNSCC) 患者 13。这些图像采用 DICOM 格式。
*默沙东 T10。该子数据集来自第十届医学分割十项全能14。为了获得更多包含脊柱的切片，我们选择了由 201 个案例组成的 task03_liver 数据集。这些图像采用神经影像信息学技术倡议 (NIfTI) 格式 (https://nifti.nimh.nih.gov/nifti-1)。
*新冠肺炎。该子数据集包含来自 632 名 COVID-19 感染患者的非增强胸部 CT。这些图像是在爆发环境中的护理点从经逆转录聚合酶链反应 (RT-PCR) 确认存在 SARS-CoV-215 的患者那里获取的。我们选择 40 次扫描，图像以 NIfTI 格式存储。
我们将所有 DICOM 图像重新格式化为 NIfTI，以简化数据处理和去识别图像，满足贡献站点的机构审查委员会 (IRB) 政策。这些子数据集的更多细节可以在 12-15 中找到。所有现有的子数据集都在知识共享许可 CC-BY-NC-SA 下，我们将保持许可不变。需要注意的是，对于子数据集 task03_liver 和子数据集 COVID-19，我们只从中选择了一部分案例，并且在所有这些数据源中，我们排除了那些质量非常低的案例。

To advance research on spinal image analysis, we hereby present a large-scale comprehensive dataset: CTSpine1K. To construct a comprehensive spinal dataset that replicates real-world appearance variations, we curated CTSpine1K from four open-source resources, totaling 1,005 distinct CT volumes (over 500,000 annotated slices and more than 11,000 vertebrae).
* Cologne. This sub-dataset is derived from the CT COLONOGRAPHY dataset12 associated with a CT colonography trial. We randomly selected one of two positions per patient for our dataset (we have released the code dicom2nii.py used for this selection), which carry similar clinical information. It contains 825 CT scans in Digital Imaging and Communications in Medicine (DICOM) format.
* HNSCC-3DCT-RT. This sub-dataset includes 13 patients with head and neck squamous cell carcinoma (HNSCC) scanned using a Siemens 16-slice CT scanner. These images are stored in DICOM format.
* Merck T10. This sub-dataset originates from the 10th Medical Segmentation Decathlon14. To obtain more slices containing the spine, we selected the task03_liver dataset consisting of 201 cases. These images are stored in the Neuroimaging Informatics Technology Initiative (NIfTI) format (https://nifti.nimh.nih.gov/nifti-1).
* COVID-19. This sub-dataset contains non-contrast chest CT scans from 632 COVID-19-infected patients. These images were acquired at point-of-care settings during the outbreak from patients with RT-PCR-confirmed SARS-CoV-2 infection15. We selected 40 scans, and the images are stored in NIfTI format.
We reformat all DICOM images to NIfTI to simplify data processing and de-identify the images, complying with the institutional review board (IRB) policies of the contributing sites. More details of these sub-datasets can be found in references 12–15. All existing sub-datasets are licensed under the Creative Commons CC-BY-NC-SA license, and we will retain this license unchanged. It is worth noting that for the task03_liver and COVID-19 sub-datasets, we only selected a subset of cases from them, and we excluded those with very low quality across all these data sources.

搜集汇总

背景与挑战

背景概述

CTSpine1K是一个大规模脊柱CT图像数据集，专为椎骨分割研究设计，包含来自四个开源资源的1,005个CT卷，覆盖超过500,000个标记切片和11,000个椎骨，具有多样外观变化。数据集整合了DICOM和NIfTI格式，经过去识别处理，适用于医疗智慧脊柱分析任务，由中国科学院等机构于2021年发布。