PLAn-7T is an MR image segmentation method optimized for processing ultra-high field (7T) images. It is built upon the nnU-Net framework and uses transfer learning to enable effective training with a relatively small (<10) number of annotated examples.
This repository is a companion to the PLAn 7T paper:
Donnay, C.*, Dieckhaus, H.*, Tsagkas, H., Gaitán, M.I., Beck, E.S., Mullins, W., Reich, D.S., Nair, G.
Pseudo-Label Assisted nnU-Net (PLAn) Enables Automatic Segmentation of 7T MRI From a Single Acquisition.
Front. Neuroimaging, 01 December 2023; Sec. Brain Imaging Methods; Volume 2 - 2023. https://doi.org/10.3389/fnimg.2023.1252261.
These details are also found in install_PLAn7T.txt.
To install PLAn 7T for training or inference, you first need to install the necessary python packages:
conda create --name PLAn_7T python=3.8 cudatoolkit=11.3 pytorch torchvision torchaudio -c pytorch
You next need to install the base nnU-Net package. NOTE: PLAn 7T uses nnU-Net v1, so please be sure to install the correct version.
conda activate PLAn_7T
git clone https://github.com/MIC-DKFZ/nnUNet.git
cd nnUNet
git checkout nnunetv1
pip install -e .
Set your nnU-Net environment variables as directed here.
Obtain the PLAn files from this repository:
git clone https://github.com/hdieckhaus/PLAn-7T.git
Finally, add the custom Trainer file nnUNetTrainerV2Transfer_125epochs_0p0001.py into the nnunet/training/network_training folder
See the PLAn7T_inference.sh script. Accepted input formats are .nii or .nii.gz files.
See the PLAn7T_training.sh script. Accepted input formats are .nii or .nii.gz files.