Comparing Linear and Nonlinear Finite Element Models of Vertebral Strength Across the Thoracolumbar Spine: A Benchmark from Density-Calibrated Computed Tomography

This dataset supports the publication Comparing Linear and Nonlinear Finite Element Models of Vertebral Strength Across the Thoracolumbar Spine: A Benchmark from Density-Calibrated Computed Tomography (Walle, Matheson, Boyd; 2025, https://doi.org/10.1101/2025.04.19.649449). This dataset provides calibrated, resampled, and re-aligned CT images and segmentation masks for all non-fractured vertebrae from the VerSe'19 dataset, prepared for benchmarking finite element (FE) modeling of the thoracolumbar spine. The original VerSe 2019 dataset includes 160 CT scans of 141 patients. In this dataset, each vertebra is provided as a cropped NIfTI CT image and corresponding segmentation mask that separates cortical and trabecular compartments, vertebral body and processes, as well as anterior and posterior intervertebral disks for applying boundary conditions in FE analysis. Vertebrae have been individually reoriented such that the superior and inferior endplates of the vertebral body are approximately aligned with the axial plane, facilitating standardized model setup and loading conditions. Folder structure: nii_files/└── Subject/ ├── im/ │ ├── vertebra_20_im.nii.gz │ └── vertebra_21_im.nii.gz └── seg/ ├── vertebra_20_seg.nii.gz └── vertebra_21_seg.nii.gz calibration_files/└── Subject/ └── subject.txt The number in each filename (e.g., vertebra_20_im.nii.gz) corresponds to the original VerSe’19 vertebral label: 1–7 for C1–C7, 8–19 for T1–T12, and 20–25 for L1–L6. Labels 26 and 27 (sacrum and coccyx) as well as 28 (a 13th thoracic vertebra, T13). Segmentation labels: Trabecular bone of the vertebral body Cortical bone of the vertebral body Trabecular bone of the spinous process Cortical bone of the spinous process Posterior intervertebral disk Anterior intervertebral disk All segmentations were generated using 3D nnU-Net (v2) models trained on publicly available datasets. Vertebral substructure segmentation was based on VerSe’19 (https://doi.org/10.1148/ryai.2020190138), and reference tissue segmentation models were trained using TotalSegmentator (https://doi.org/10.5281/zenodo.10047292). The corresponding model weights are shared separately: vertebral substructures (https://doi.org/10.5281/zenodo.15238176) and reference tissues (https://doi.org/10.5281/zenodo.15238423). Calibration logs are provided to enable phantomless internal calibration of the original VerSe’19 scans (available at https://osf.io/nqjyw/) using the Ogo toolkit (https://github.com/Bonelab/Ogo), supporting recalibration for additional tissue compartments such as muscle or adipose tissue. While finite element outputs are not included in this deposit, they are tabulated and visualized in the associated GitHub repository (https://github.com/Bonelab/spineFE-benchmark), which also provides examples to generate all required models. Original data attribution: VerSe’19 dataset: https://osf.io/nqjyw/ Sekuboyina A et al. VerSe: A Vertebrae Labelling and Segmentation Benchmark for Multi-detector CT Images, Med Image Anal. 2021. https://doi.org/10.1016/j.media.2021.102166 Löffler M et al. A Vertebral Segmentation Dataset with Fracture Grading, Radiology: Artificial Intelligence. 2020. https://doi.org/10.1148/ryai.2020190138 Liebl H, Schinz D et al. A Computed Tomography Vertebral Segmentation Dataset with Anatomical Variations and Multi-Vendor Scanner Data. Preprint: https://arxiv.org/pdf/2103.06360.pdf本数据集支持下述论文的发表：《胸腰椎脊柱椎体强度的线性与非线性有限元（Finite Element, FE）模型对比：基于密度校准计算机断层扫描（Computed Tomography, CT）的基准测试》（Walle、Matheson、Boyd；2025，https://doi.org/10.1101/2025.04.19.649449）。 本数据集源自VerSe'19数据集，包含所有未骨折椎体的校准、重采样与重配准CT图像及分割掩码，专为胸腰椎脊柱有限元建模基准测试制备。原始VerSe 2019数据集包含141名患者的160次CT扫描。本数据集中，每个椎体均以裁剪后的NIfTI格式CT图像及对应分割掩码提供，掩码可区分皮质骨与松质骨区域、椎体及椎突，以及用于有限元分析边界条件设置的前、后椎间盘。所有椎体均经过单独重定向，使椎体的上、下终板大致与轴状面对齐，便于标准化的模型搭建与加载条件设置。 文件夹结构如下： nii_files/ └── 受试者（Subject）目录/ ├── im/ │ ├── vertebra_20_im.nii.gz │ └── vertebra_21_im.nii.gz └── seg/ ├── vertebra_20_seg.nii.gz └── vertebra_21_seg.nii.gz calibration_files/ └── 受试者（Subject）目录/ └── subject.txt 每个文件名中的数字（例如vertebra_20_im.nii.gz）对应VerSe’19原始椎体标签：1~7对应C1~C7（颈椎1~7），8~19对应T1~T12（胸椎1~12），20~25对应L1~L6（腰椎1~6）。标签26、27分别对应骶骨与尾骨，标签28对应第13胸椎（T13）。 分割标签定义如下： 1. 椎体松质骨 2. 椎体皮质骨 3. 棘突松质骨 4. 棘突皮质骨 5. 后椎间盘 6. 前椎间盘 所有分割掩码均基于公开数据集训练的3D nnU-Net（v2）模型生成。椎体亚结构分割基于VerSe’19数据集（https://doi.org/10.1148/ryai.2020190138），参考组织分割模型则通过TotalSegmentator工具训练（https://doi.org/10.5281/zenodo.10047292）。相关模型权重已单独共享：椎体亚结构模型权重（https://doi.org/10.5281/zenodo.15238176）与参考组织模型权重（https://doi.org/10.5281/zenodo.15238423）。 本数据集提供校准日志，可通过Ogo工具包（https://github.com/Bonelab/Ogo）对原始VerSe’19扫描数据（可从https://osf.io/nqjyw/获取）实现无体模内部校准，支持对肌肉、脂肪组织等额外组织区域进行重新校准。尽管本数据集未包含有限元分析输出结果，但相关结果已在关联的GitHub仓库（https://github.com/Bonelab/spineFE-benchmark）中以表格与可视化形式呈现，该仓库同时提供了生成所有所需模型的示例代码。 原始数据集归属说明： 1. VerSe’19数据集：https://osf.io/nqjyw/ 2. Sekuboyina A等. VerSe：多探测器CT图像的椎体标记与分割基准测试[J]. 医学图像分析, 2021. https://doi.org/10.1016/j.media.2021.102166 3. Löffler M等. 带骨折分级的椎体分割数据集[J]. 放射学：人工智能, 2020. https://doi.org/10.1148/ryai.2020190138 4. Liebl H、Schinz D等. 带解剖变异与多厂商扫描设备数据的CT椎体分割数据集[EB/OL]. 预印本: https://arxiv.org/pdf/2103.06360.pdf