This dataset is mainly for brain image registration tasks (may be used for other tasks, e.g., segmentation...). We used the whole-brain segmentation masks to generate nonfinite-modality brain images and the trained improved 3D VAE to generate the deformation vector field. We can use the dataset to augment real datasets. Due to the limitation of upload files, the data is part of our whole dataset. If you need the whole dataset, please use git issues to contact with me or you can use the code to generate your dataset.

Now the paper is accepted (Computers in Biology and Medicine, IF: 6.7) and available online.

https://doi.org/10.1016/j.compbiomed.2022.105780

scipy==1.6.2
numpy==1.20.2
pandas==1.2.3
torch==1.7.1
matplotlib==3.4.1
SimpleITK==2.0.2

You can simply run the Demo.py. (Make sure you have already installed these packages)

python Demo.py

• datashow.pdf show some generated images
• Dataset folders are masks and generated images (parts due to limitation)
• Data images folder show some slices of images
• Demo.py show how to generate images and DVFs
• DVFGenerator.py is for DVF generation
• ImageGenerator.py is for image generation
• utils.py contains tool method
• vanilla_vae.py is the VAE model

• https://github.com/AntixK/PyTorch-VAE
• https://github.com/voxelmorph/voxelmorph
• https://github.com/airlab-unibas/airlab

@article{HE2022105780,
title = {Nonfinite-modality data augmentation for brain image registration},
journal = {Computers in Biology and Medicine},
volume = {147},
pages = {105780},
year = {2022},
issn = {0010-4825},
doi = {https://doi.org/10.1016/j.compbiomed.2022.105780},
url = {https://www.sciencedirect.com/science/article/pii/S0010482522005479},
author = {Yuanbo He and Aoyu Wang and Shuai Li and Yikang Yang and Aimin Hao},
keywords = {Nonfinite-modality, Data augmentation, Improved 3D VAE, Brain image registration},
}