Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?



Biomedisa ( is a free and easy-to-use open-source online platform for segmenting large volumetric images, e.g. CT and MRI scans, at Heidelberg University and the Heidelberg Institute for Theoretical Studies (HITS). The segmentation is based on a smart interpolation of sparsely pre-segmented slices taking into account the complete underlying image data. It can be used in addition to segmentation tools like Amira, ImageJ/Fiji and MITK. Biomedisa finds its root in the projects ASTOR and NOVA funded by the Federal Ministry of Education and Research (BMBF). If you are using Biomedisa for your research please cite: Lösel, P.D. et al. Introducing Biomedisa as an open-source online platform for biomedical image segmentation. Nat. Commun. 11, 5577 (2020).

Hardware requirements

  • One or more NVIDIA GPUs with compute capability 3.0 or higher or an Intel CPU.
  • 32 GB RAM or more (depending on the size of the image data).

Software requirements

Installation (command-line-only)

Full installation (GUI)

Interpolation examples

Small example

Download the tumor test example from the gallery or directly as follows:

# Brain tumor
wget --no-check-certificate -O ~/Downloads/tumor.tif
wget --no-check-certificate -O ~/Downloads/labels.tumor.tif

Run the Biomedisa interpolation. The result will be saved in Downloads as final.tumor.tif.

# Ubuntu
python3 ~/git/biomedisa/demo/ ~/Downloads/tumor.tif ~/Downloads/labels.tumor.tif

# Windows
python git\biomedisa\demo\ Downloads\tumor.tif Downloads\labels.tumor.tif

Further examples

Download further examples from the gallery or directly as follows:

# Trigonopterus
wget --no-check-certificate -O ~/Downloads/trigonopterus.tif
wget --no-check-certificate -O ~/Downloads/

# Mineralized wasp
wget --no-check-certificate -O ~/Downloads/NMB_F2875.tif
wget --no-check-certificate -O ~/Downloads/labels.NMB_F2875.tif

Run the segmentation using e.g. 4 GPUs.

# Ubuntu
mpiexec -np 4 python3 ~/git/biomedisa/demo/ ~/Downloads/NMB_F2875.tif ~/Downloads/labels.NMB_F2875.tif

# Windows
mpiexec -np 4 python -u git\biomedisa\demo\ Downloads\NMB_F2875.tif Downloads\labels.NMB_F2875.tif

Obtain uncertainty and smoothing as optional results.

python3 ~/git/biomedisa/demo/ ~/Downloads/tumor.tif ~/Downloads/labels.tumor.tif --uncertainty --smooth 100

Use pre-segmentation with different orientations (not exclusively xy-plane).

python3 ~/git/biomedisa/demo/ 'path_to_image' 'path_to_labels' --allaxis

For more information, type

python3 ~/git/biomedisa/demo/ --help

Memory Error

If memory errors (either GPU or host memory) occur, you can start the segmentation as follows:

python3 ~/git/biomedisa/demo/ 'path_to_image' 'path_to_labels' -np 4 -sz 2 -sy 2 -sx 2

Where -n is the number of GPUs and each axis (x,y and z) is divided into two overlapping parts. The volume is thus divided into 2*2*2=8 subvolumes. These are segmented separately and then reassembled.

AI example

Automatic segmentation using a trained network

Download a trained neural network and a test image from the gallery or directly as follows:

wget --no-check-certificate -O ~/Downloads/heart.h5
wget --no-check-certificate -O ~/Downloads/testing_axial_crop_pat13.nii.gz

Use the trained neural network to predict the result of the test image with a batch size of 6 batches. The result will be saved in Downloads as final.testing_axial_crop_pat13.tif.

# Ubuntu
python3 ~/git/biomedisa/demo/ ~/Downloads/testing_axial_crop_pat13.nii.gz ~/Downloads/heart.h5 --predict -bs 6

# Windows
python git\biomedisa\demo\ Downloads\testing_axial_crop_pat13.nii.gz Downloads\heart.h5 -p -bs 6

Train a neural network for automatic segmentation

To train the neural network yourself, download and extract the training data from the gallery or directly as follows:

wget --no-check-certificate -O ~/Downloads/training_heart.tar
wget --no-check-certificate -O ~/Downloads/training_heart_labels.tar
cd ~/Downloads
tar -xf training_heart.tar
tar -xf training_heart_labels.tar

Train a neural network with 200 epochs and batch size (-bs) of 24. The result will be saved in Downloads as heart.h5. If you have a single GPU or low memory, reduce the batch size to 6.

# Ubuntu
python3 ~/git/biomedisa/demo/ ~/Downloads/training_heart ~/Downloads/training_heart_labels --train --epochs 200 -bs 24

# Windows
python git\biomedisa\demo\ Downloads\training_heart Downloads\training_heart_labels --train --epochs 200 -bs 24

Validate the network during training

Specify directories containing validation images and validation labels.

python3 ~/git/biomedisa/demo/ ~/Downloads/training_heart ~/Downloads/training_heart_labels --train --val_images ~/Downloads/validation_images --val_labels ~/Downloads/validation_labels

Split your data into 80% training data and 20% validation data and use early stopping if there is no improvement within 10 epochs.

python3 ~/git/biomedisa/demo/ ~/Downloads/training_heart ~/Downloads/training_heart_labels --train --validation_split 0.8 --early_stopping 10

Automatic cropping

Both the training and inference data should be cropped to the region of interest for best performance. As an alternative to manual cropping, you can use Biomedisa's AI-based automatic cropping. After training, auto cropping is automatically applied to your inference data.

python3 ~/git/biomedisa/demo/ 'path_to_images' 'path_to_labels' --train --crop_data

Get help

For more information, type

python3 ~/git/biomedisa/demo/ --help

Update Biomedisa

If you have used git clone, change to the Biomedisa directory and make a pull request.

cd git/biomedisa
git pull

If you have installed the full version of Biomedisa (including MySQL database), you also need to update the database.

python3 migrate

If you have installed an Apache Server, you need to restart the server.

sudo service apache2 restart


For the versions available, see the tags on this repository.


  • Philipp D. Lösel

See also the list of contributors who participated in this project.


Frequently asked questions can be found at:


If you use the package or the online platform, please cite the following paper.

Lösel, P.D. et al. Introducing Biomedisa as an open-source online platform for biomedical image segmentation. Nat. Commun. 11, 5577 (2020).


This project is covered under the EUROPEAN UNION PUBLIC LICENCE v. 1.2 (EUPL).


The Biomedical Image Segmentation App ( is an open-source online platform for segmentation of 3D images, e.g. CT and MRI scans, @UniHeidelberg @HITStudies.







No packages published