Skip to content

Latest commit

 

History

History
188 lines (140 loc) · 12.2 KB

TotalSegmentator_MRI.md

File metadata and controls

188 lines (140 loc) · 12.2 KB

TotalSegmentator MRI

Dataset Information

TotalSegmentator MRI dataset includes 298 MR images and provides segmentation annotations for up to 56 different commonly used anatomical structures. Among these, 251 MR images are sourced from the Picture Archiving and Communication System (PACS) at the University Hospital Basel, collected between 2011 and 2023, while the remaining 47 MR images come from the Imaging Data Commons (IDC) platform to enhance diversity. The dataset is derived from random sampling in routine clinical practice, representing a real-world dataset that can be generalized to clinical applications. It encompasses various lesions, scanners, imaging sequences, and data from different medical institutions. Notably, although the official paper mentions 59 categories, only 56 classes are provided in the publicly available dataset, with slight discrepancies.

Magnetic Resonance Imaging (MRI) offers detailed anatomical images without using ionizing radiation, crucial for diagnosing various clinical conditions, including neurological disorders and musculoskeletal injuries. While the TotalSegmentator (CT) dataset has been widely used, MRI image segmentation remains challenging due to variations in imaging parameters and protocols across different sequences and body parts, impacting the universality and accuracy of algorithms. The TotalSegmentator MRI dataset expands the capability to handle various MRI images, aiming to develop an open-source and user-friendly segmentation model that can automatically and robustly segment major anatomical structures without relying on specific MR sequences.

Dataset Meta Information

Dimensions Modality Task Type Anatomical Structures Anatomical Area Number of Categories Data Volume File Format
3D MRI Segmentation 56 body structures Entire body 56 298 .nii.gz

Resolution Details

Dataset Statistics spacing (mm) size
min (0.20, 0.20, 0.20) (11,12,5)
median (1.03, 1.18, 3.0) (320, 260, 80)
max (6.5, 10.6, 28) (1092, 896, 1138)

Number of 2D slices in the dataset: 57,991

Label Information Statistics

Label Structure Cases Percentage Max Volume (cm³) Median Volume (cm³)
1 spleen 118 39.60% 1075.97 173.02
2 kidney_right 115 38.59% 335.82 118.39
3 kidney_left 120 40.27% 446.03 113.4
4 gallbladder 88 29.53% 110.95 14.03
5 liver 128 42.95% 3172.88 1181
6 stomach 123 41.28% 833.72 196.02
7 pancreas 108 36.24% 169.44 43.42
8 adrenal_gland_right 105 35.23% 7.75 1.89
9 adrenal_gland_left 104 35.26% 9.07 2.81
10 lung_left 139 46.64% 2577.92 531.88
11 lung_right 138 46.31% 3616.18 623.26
12 esophagus 120 40.27% 34.84 8.73
13 small_bowel 154 51.68% 1699.95 258.14
14 duodenum 117 39.26% 105.99 31.74
15 colon 179 60.07% 2837.21 263.84
16 urinary_bladder 64 21.48% 502.32 64.98
17 prostate 36 12.08% 108.13 17.98
18 sacrum 93 31.21% 226.63 112.92
19 vertebrae 195 65.44% 399.01 151.25
20 intervertebral_discs 193 64.77% 201.98 73.01
21 spinal_cord 197 66.11% 105.17 38.41
22 heart 116 38.93% 818.85 360.54
23 aorta 141 47.32% 226.72 55.46
24 inferior_vena_cava 136 45.64% 129.78 32.2
25 portal_vein_and_splenic_vein 113 37.92% 35.47 10.23
26 iliac_artery_left 110 36.91% 24.34 4.14
27 iliac_artery_right 113 37.92% 25.1 4.18
28 iliac_vena_left 106 35.57% 38.99 7.3
29 iliac_vena_right 108 36.24% 29.86 5.25
30 humerus_left 40 13.42% 167.43 31.31
31 humerus_right 41 13.76% 179.23 24.09
32 fibula 35 11.74% 50.67 9.41
33 tibia 38 12.75% 426.23 103.47
34 femur_left 88 29.53% 438 103.01
35 femur_right 88 29.53% 462.92 110.36
36 hip_left 118 39.60% 365.59 101.1
37 hip_right 119 39.93% 336.61 103.37
38 gluteus_maximus_left 84 28.19% 859.77 235.84
39 gluteus_maximus_right 83 27.85% 733.48 263.89
40 gluteus_medius_left 92 30.87% 459.73 127.54
41 gluteus_medius_right 92 30.87% 345.01 125.98
42 gluteus_minimus_left 73 24.50% 136.64 46.14
43 gluteus_minimus_right 72 24.16% 86.14 40.38
44 autochthon_left 178 59.73% 614.1 185.48
45 autochthon_right 177 59.40% 583.41 169.24
46 iliopsoas_left 166 55.70% 423.27 108.68
47 iliopsoas_right 170 57.05% 413.83 99.59
48 quadriceps_femoris_left 70 23.49% 2228.98 75.2
49 quadriceps_femoris_right 72 24.16% 2149.81 81.72
50 thigh_medial_compartment_left 63 21.14% 1163.8 165.98
51 thigh_medial_compartment_right 72 24.16% 1167.81 155.36
52 thigh_posterior_compartment_left 50 16.78% 657.73 61.61
53 thigh_posterior_compartment_right 52 17.45% 664.63 19.04
54 sartorius_left 69 23.15% 198.14 16.75
55 sartorius_right 68 22.82% 169 21.52
56 brain 17 5.70% 1783.46 1577.3

Visualization

Official Visualization.

Local ITK-SNAP Visualization.

File Structure

The dataset file structure is as follows: the folder contains 298 subdirectories (from s0001 to s0298). Each subdirectory contains an MRI image file (mri.nii.gz) and corresponding segmentation files for multiple anatomical structures. In the root directory, there is also a meta.csv file. The meta.csv file contains the following metadata categories for each MRI image: patient information (age, gender), institution information (institution, study type, manufacturer, scanner model), scan parameters (slice thickness, scan sequence, repetition time, echo time, magnetic field strength), and data division (data source, dataset partition).

TotalsegmentatorMRI_dataset_v100/
├── s0001/
│   ├── mri.nii.gz
│   └── segmentations/
│       ├── adrenal_gland_left.nii.gz
│       ├── adrenal_gland_right.nii.gz
│       ├── aorta.nii.gz
│       ├── autochthon_left.nii.gz
│       ├── autochthon_right.nii.gz
│       └── ...   # 省略其他类别
├── s0002/
├── s0003/
├── ...
├── s0298/
└── meta.csv

Authors and Institutions

Tugba Akinci D’Antonoli (University Hospital Basel, Switzerland)

Lucas K. Berger (University Hospital Basel, Switzerland)

Ashraya K. Indrakanti (University Hospital Basel, Switzerland)

Nathan Vishwanathan (University Hospital Basel, Switzerland)

Jakob Weiß (University Medical Center Freiburg, Germany)

Matthias Jung (University Medical Center Freiburg, Germany)

Zeynep Berkarda (University Medical Center Freiburg, Germany)

Alexander Rau (University Medical Center Freiburg, Germany)

Marco Reisert (University Medical Center Freiburg, Germany)

Thomas Küstner (University Hospital of Tuebingen, Germany)

Alexandra Walter (German Cancer Research Center, Germany; Karlsruhe Institute of Technology, Germany)

Elmar M. Merkle (University Hospital Basel, Switzerland)

Martin Segeroth (University Hospital Basel, Switzerland)

Joshy Cyriac (University Hospital Basel, Switzerland)

Shan Yang (University Hospital Basel, Switzerland)

Jakob Wasserthal (University Hospital Basel, Switzerland)

Source Information

Official Website: https://github.com/wasserth/TotalSegmentator

Download Link: https://zenodo.org/records/11367005

Article Address: https://arxiv.org/pdf/2405.19492

Publication Date: 2024-05

Citation

@misc{dantonoli2024totalsegmentator,
      title={TotalSegmentator MRI: Sequence-Independent Segmentation of 59 Anatomical Structures in MR images}, 
      author={Tugba Akinci D'Antonoli and Lucas K. Berger and Ashraya K. Indrakanti and Nathan Vishwanathan and Jakob Weiß and Matthias Jung and Zeynep Berkarda and Alexander Rau and Marco Reisert and Thomas Küstner and Alexandra Walter and Elmar M. Merkle and Martin Segeroth and Joshy Cyriac and Shan Yang and Jakob Wasserthal},
      year={2024},
      eprint={2405.19492},
      archivePrefix={arXiv},
      primaryClass={eess.IV}
}

Original introduction article is here.