*works also on CT!
You can select older versions with --dataset_id [ID]
| ID | NAME | Version | Modality | Resolution | Link |
|---|---|---|---|---|---|
| 100 | VIBESegmentator | current | MRI/CT | axial 1.41 mm, S=3mm | Paper |
| 99 | VIBESegmentator | current (sagittal resolution) | MRI/CT | sagittal 1.41 mm, R=3mm | Paper |
| 12 | VIBESeg-CT-iso-0.8mm | current | CT | iso 0.8 mm | |
| 278 | Splits the body in 11 regions | current | NAKO VIBE-only | iso 4 mm | |
| 282 | Torso Water-Fat-swap detection¹ | current (use script from linked repo) | multi-echo-Vibe VIBE | axial 1.41 mm, S=3mm | preprint; Other Repo |
| 538 | All Feet bone CT | current (crop to feet required) | CT | iso 0.8 mm | |
| 540 | Pankreas | current (crop to Torso) | VIBE | iso 1.0 mm |
All spine segmentation models for VIBE, CT and T2w have been moved to SPINEPS
¹ Use run_water_fat_swap_detection.py; If you only want the swaps segmented, else look in the other repo
| ID | NAME | Version | Modality | Resolution | Link |
|---|---|---|---|---|---|
| 98 | VIBESegmentator | (deprecated-spinal channel bugged in CT) | MRI | axial 1.41 mm, S=3mm | |
| 80 | VIBESegmentator | preprint (deprecated) | MRI | axial 1.41 mm, S=3mm | |
| 85-87 | VIBESegmentator | early preprint (deprecated) | MRI | axial 1.41 mm, S=3mm | |
| 520 | CT Bone segmentations | preliminary | CT | iso 0.8 mm | |
| 511 | Spine Vertebra instance | Fast, but may produce off-by-one errors. Use SPINEPS if this method fails. | VIBE-only | axial 1.41 mm, S=3mm | See also |
| 512 | Spine Vertebra subregion | incorporated in SPINEPS | VIBE-only | axial 1.41 mm, S=3mm | See also |
All outdated/deprecated models are still available, we just no longer recommend them, as their should be better alternatives.
- Python 3.10 or higher.
- Tested on Ubuntu and Windows.
- One of the following:
- Nvidia-GPU with 8 GB of RAM or more.
- A mps device (--ddevice mps)
- A strong CPU; This is usually very slow. (--ddevice cpu)
- Open a command line (search for Terminal on Mac or cmd on Windows in your OS search).
- Ensure you have Anaconda (Python) installed.
- Navigate to the folder where you want to download the script using
cd [FOLDER PATH].
# Run this commands by coping in to the Terminal
# Recommended: make a virtual Python environment (example shows Anaconda)
conda create -n "VIBESegmentator" python=3.11.0
conda activate VIBESegmentator
# Install PyTorch that works with your GPU (follow instructions at https://pytorch.org/get-started/locally/)
pip install torch torchvision torchaudio
# Install required Python packages
pip install TPTBox # use ==0.8.0 or ==0.7.6 if you encounter issues with newer versions.
pip install nnunetv2
# Download the scripts (they will be downloaded to your current folder)
git clone https://github.com/robert-graf/VIBESegmentator.git
cd VIBESegmentator
It should take less than 30 minutes to install, excluding installing conda/python. The longest time takes to install pytroch.
Many packages are dropping support for Python 3.9, making it increasingly difficult for pip to automatically resolve compatible dependency versions.
We do not recommend using Python 3.9. Avoid it if you have the option to use a newer Python version.
# Install PyTorch compatible with your GPU
# Follow the instructions at https://pytorch.org/get-started/locally/
# pip install "dicom2nifti==2.0.1"
# pip install "acvl_utils==0.2"
# pip install "nnunetv2==2.5.2"
# pip install "TPTBox==0.7.6"Download the nnUNet weights automatically. They are put in [Path to this project]/nnUNet/nnUNet_results/.
# Run this commands in a Terminal where you navigate to the folder where `run_VIBESegmentator.py` is
conda activate VIBESegmentator
# Total Segmentation
python run_VIBESegmentator.py --img [IMAGE-PATH] --out_path [OUTPATH] --ddevice [cpu|cuda|mpu (optional)]
conda activate VIBESegmentator
#cd to the folder where you downloaded the GitHub files (the folder where the run_VIBESegmentator.py is located)Let's run some examples that were never seen during training. You can download all four examples from here
(Source: link Number s0284, s0286, s0287, s0288)
Unzip the file. For this example, we copied all files to a folder called 'example' where the run_VIBESegmentator.py is located.
python run_VIBESegmentator.py --img example/mri1.nii.gz --out_path example/seg1.nii.gz
python run_VIBESegmentator.py --img example/mri2.nii.gz --out_path example/seg2.nii.gz
python run_VIBESegmentator.py --img example/mri3.nii.gz --out_path example/seg3.nii.gz
python run_VIBESegmentator.py --img example/mri4.nii.gz --out_path example/seg4.nii.gzThe segmentation is stored in "example/segX.nii.gz." You can view the results with ITKSnap by first drag-and-drop the mriX.nii.gz and then drag-and-drop the segX.nii.gz (with X is a number from before) and click on "Open as Segmentation"
To run this script on your data, update the paths.
| ID | NAME |
|---|---|
| 1 | spleen |
| 2 | kidney_right |
| 3 | kidney_left |
| 4 | gallbladder |
| 5 | liver |
| 6 | stomach |
| 7 | pancreas |
| 8 | adrenal_gland_right |
| 9 | adrenal_gland_left |
| 10 | lung_upper_lobe_left |
| 11 | lung_lower_lobe_left |
| 12 | lung_upper_lobe_right |
| 13 | lung_middle_lobe_right |
| 14 | lung_lower_lobe_right |
| 15 | esophagus |
| 16 | trachea |
| 17 | thyroid_gland |
| 18 | intestine |
| 19 | duodenum |
| 20 | unused |
| 21 | urinary_bladder |
| 22 | prostate |
| 23 | sacrum |
| 24 | heart |
| 25 | aorta |
| 26 | pulmonary_vein |
| 27 | brachiocephalic_trunk |
| 28 | subclavian_artery_right |
| 29 | subclavian_artery_left |
| 30 | common_carotid_artery_right |
| 31 | common_carotid_artery_left |
| 32 | brachiocephalic_vein_left |
| 33 | brachiocephalic_vein_right |
| 34 | atrial_appendage_left |
| 35 | superior_vena_cava |
| 36 | inferior_vena_cava |
| 37 | portal_vein_and_splenic_vein |
| 38 | iliac_artery_left |
| 39 | iliac_artery_right |
| 40 | iliac_vena_left |
| 41 | iliac_vena_right |
| 42 | humerus_left |
| 43 | humerus_right |
| 44 | scapula_left |
| 45 | scapula_right |
| 46 | clavicula_left |
| 47 | clavicula_right |
| 48 | femur_left |
| 49 | femur_right |
| 50 | hip_left |
| 51 | hip_right |
| 52 | spinal_cord |
| 53 | gluteus_maximus_left |
| 54 | gluteus_maximus_right |
| 55 | gluteus_medius_left |
| 56 | gluteus_medius_right |
| 57 | gluteus_minimus_left |
| 58 | gluteus_minimus_right |
| 59 | autochthon_left |
| 60 | autochthon_right |
| 61 | iliopsoas_left |
| 62 | iliopsoas_right |
| 63 | sternum |
| 64 | costal_cartilages |
| 65 | subcutaneous_fat |
| 66 | muscle |
| 67 | inner_fat |
| 68 | IVD |
| 69 | vertebra_body |
| 70 | vertebra_posterior_elements |
| 71 | spinal_channel |
| 72 | bone_other |
Open access paper: doi:10.1007/s00330-025-12035-9
The related paper is available as preprint: arXiv:2406.00125
@article{graf2025vibesegmentator,
title={VIBESegmentator: full body MRI segmentation for the NAKO and UK Biobank},
author={Graf, Robert and Platzek, Paul and Riedel, Evamaria Olga and Ramsch{\"u}tz, Constanze and Starck, Sophie and M{\"o}ller, Hendrik K and Atad, Matan and V{\"o}lzke, Henry and B{\"u}low, Robin and Schmidt, Carsten Oliver and others},
journal={European Radiology},
pages={1--15},
year={2025},
publisher={Springer}
}
run_VIBESegmentator.py [...] --dataset_id 12
Labels see
from TPTBox.core.vert_constants import Full_Body_Instance
for i in Full_Body_Instance:
print(f"{i.value:4}\t: {i.name}")run_VIBESegmentator.py [...] --dataset_id 278
