Skip to content

Latest commit

 

History

History
147 lines (114 loc) · 9.05 KB

VerSe.md

File metadata and controls

147 lines (114 loc) · 9.05 KB

VerSe

Dataset Information

The VerSe (Vertebrae Segmentation) dataset is a large-scale, multi-device, multi-center CT image spine segmentation dataset, encompassing scans from 355 patients totaling 374 scans. This dataset combines data from the VerSe19 and VerSe20 challenges at MICCAI in 2019 and 2020. Specifically, VerSe19 included 160 scans from 141 patients, while VerSe20 involved 319 scans from 300 patients. However, due to the overlap of 86 patients in both datasets, the total number of unique patients in VerSe is not simply the sum of the two datasets.

VerSe has categorized these 374 scans into 141 for training, 120 for validation, and 113 for testing, with all scans and annotations publicly available. Within the 26 vertebral annotation categories, in addition to the conventional 24 categories C1-C7, T1-T12, and L1-L5, the dataset uniquely annotates the rarer T13 and L6 vertebrae. Additionally, vertebrae only partially visible at the top or bottom of a scan were not annotated.

VerSe Patients Scans Scan split Vertebrae (Cer/Tho/Lum)
2019 141 160 80/40/40 1725 (220/884/621)
2020 300 319 113/103/103 4141 (581/2255/1305)
Total 355 374 141/120/113 4505 (611/2387/1507)

Dataset Meta Information

Dimensions Modality Task Type Anatomical Structures Anatomical Area Number of Categories Data Volume File Format
3D CT Segmentation Vertebrae Vertebrae 26 141 for training, 120 for validation, 113 for test .nii.gz

Resolution Details

Dataset Statistics spacing (mm) size
min (0.19, 0.19, 0.40) (103, 144, 34)
median (0.92, 0.92, 1.0) (512, 512, 292)
max (1.67, 1.67, 5.0) (960, 2048, 2023)

Label Information Statistics

Label Vertebrae Cases Percentage Max Volume (cm³) Min Volume (cm³) Median Volume (cm³)
1 C1 (Cervical 1) 76 20.32% 25 7.82 14.24
2 C2 (Cervical 2) 76 20.32% 28.23 11.73 18.38
3 C3 (Cervical 3) 80 21.39% 20.66 8.26 12.98
4 C4 (Cervical 4) 81 21.66% 19.69 7.57 12.51
5 C5 (Cervical 5) 90 24.06% 19.61 8.01 13.08
6 C6 (Cervical 6) 98 26.20% 22.2 8.59 17.5
7 C7 (Cervical 7) 128 34.49% 26.97 10.19 17.34
8 T1 (Thoracic 1) 180 48.13% 35.99 12.15 22.64
9 T2 (Thoracic 2) 198 52.94% 36.18 13.23 22.61
10 T3 (Thoracic 3) 182 48.66% 35.57 7.37 21.72
11 T4 (Thoracic 4) 174 46.52% 39.59 12.75 22.34
12 T5 (Thoracic 5) 165 44.12% 39.86 13.71 24.04
13 T6 (Thoracic 6) 160 42.78% 43.19 15.04 26.11
14 T7 (Thoracic 7) 166 44.39% 46.05 17.03 29.78
15 T8 (Thoracic 8) 176 47.06% 49.81 17.72 32.2
16 T9 (Thoracic 9) 215 57.49% 59.08 19.04 34.85
17 T10 (Thoracic 10) 249 66.58% 61.32 13.79 38.72
18 T11 (Thoracic 11) 268 71.66% 66.97 22.54 41.98
19 T12 (Thoracic 12) 272 72.73% 74.24 23.19 46.4
20 L1 (Lumbar 1) 298 79.68% 82.81 28.54 51.46
21 L2 (Lumbar 2) 297 79.41% 87.59 30.93 56.54
22 L3 (Lumbar 3) 295 78.88% 99.59 27.32 62.5
23 L4 (Lumbar 4) 292 78.07% 108.07 37.8 64.66
24 L5 (Lumbar 5) 276 73.80% 106.12 38 63.76
25 L6 (Lumbar 6) 50 13.37% 101.18 47.52 70.55
26 Sacrum 0 0.00% 0 0 0
27 Coccyx 0 0.00% 0 0 0
28 T13 (Thoracic 13) 6 1.60% 66.24 31.41 57.06

Label 26 27 are reserved but not provided in VerSe. Volume unit is cm^3.

Visualization

Official Visualization.

ITK-SNAP Visualization.

File Structure

The VerSe dataset includes six main folders, corresponding to the training, validation, and testing data for both VerSe19 and VerSe20. Each main folder contains two subdirectories: rawdata and derivatives. The rawdata subdirectory is dedicated to storing the original CT images ending in .nii.gz, while the derivatives subdirectory contains the segmentation masks, centroid coordinates stored in .json format, and .png reconstructed images for preview.

For example, the folder structure for the VerSe20 training data is as follows:

dataset-01training
├── rawdata
│   └── sub-gl003
│       └── sub-gl003_dir-ax_ct.nii.gz
└── derivatives
    └── sub-gl003
        ├── sub-gl003_dir-ax_seg-subreg_ctd.json
        ├── sub-gl003_dir-ax_seg-vert_msk.nii.gz
        └── sub-gl003_dir-ax_seg-vert_snp.png

Authors and Institutions

Anjany Sekuboyina, Informatics & Klinikum rechts der Isar, Technical University of Munich.

Source Information

Official Website: https://github.com/anjany/verse

Download Link: https://github.com/anjany/verse

Article Address: https://linkinghub.elsevier.com/retrieve/pii/S1361841521002127, https://pubs.rsna.org/doi/10.1148/ryai.2020190138, https://arxiv.org/pdf/2103.06360.pdf

Publication Date: 2020-05

Citation

@article{Sekuboyina2021,
  title={VerSe: A Vertebrae Labelling and Segmentation Benchmark for Multi-detector CT Images},
  author={Sekuboyina, Anjany and Husseini, Malek E and Bayat, Amirhossein and Löffler, Maximilian and Liebl, Hans and Li, Hongwei and Tetteh, Giles and Kukačka, Jan and Payer, Christian and Štern, Darko and others},
  journal={Medical image analysis},
  volume={71},
  pages={102166},
  year={2021},
  publisher={Elsevier}
}

@article{Loffler2020,
  title={A Vertebral Segmentation Dataset with Fracture Grading},
  author={L{\"o}ffler, Maximilian T and Sekuboyina, Anjany and Jacob, Alina and Grau, Anna-Lena and Scharr, Andreas and El Husseini, Malek and Kallweit, Mareike and Zimmer, Claus and Baum, Thomas and Kirschke, Jan S},
  journal={Radiology: Artificial Intelligence},
  volume={2},
  number={4},
  pages={e190138},
  year={2020},
  publisher={Radiological Society of North America}
}

@article{Liebl2021,
  title={A Computed Tomography Vertebral Segmentation Dataset with Anatomical Variations and Multi-Vendor Scanner Data},
  author={Liebl, Hans and Schinz, David and Sekuboyina, Anjany and Malagutti, Luca and L{\"o}ffler, Maximilian T and Bayat, Amirhossein and El Husseini, Malek and Tetteh, Giles and Grau, Katharina and Niederreiter, Eva and others},
  journal={arXiv preprint arXiv:2103.06360},
  year={2021}
}

Original introduction article is here.