Preprocessing and segmentation tool for MRI images.

This is an automated pipline for for MRI images preprocessing and segmentation(nnU-Net) for medical data. Trained and tested for multiple sclerosis lesions segmentation

#The preprocessing stage is optional(!) and consists of:

• DICOM to NIfTI conversion (if needed).
• Images registration.
• N4 Bias Field Correction
• Brain extruction (using https://github.com/MIC-DKFZ/HD-BET).

Requirments

nnU-Net requires a GPU! For inference, the GPU should have 4 GB of VRAM. For training nnU-Net models the GPU should have at least 10 GB (popular non-datacenter options are the RTX 2080ti, RTX 3080 or RTX 3090). Due to the use of automated mixed precision, fastest training times are achieved with the Volta architecture (Titan V, V100 GPUs) when installing pytorch the easy way. Since pytorch comes with cuDNN 7.6.5 and tensor core acceleration on Turing GPUs is not supported for 3D convolutions in this version, you will not get the best training speeds on Turing GPUs. You can remedy that by compiling pytorch from source (see here) using cuDNN 8.0.2 or newer. This will unlock Turing GPUs (RTX 2080ti, RTX 6000) for automated mixed precision training with 3D convolutions and make the training blistering fast as well. Note that future versions of pytorch may include cuDNN 8.0.2 or newer by default and compiling from source will not be necessary.

Installation

windows

In an anaconda prompt:

conda create -n bpt__nnunet3.8 python=3.8 (yes to all)
conda activate bpt__nnunet3.8
conda config --set ssl_verify False
cd <path to project>
pip install -r ./requirements.txt (yes to all)
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia (yes to all)

Linux

1. Go to https://github.com/SuperElastix/elastix/releases/tag/5.0.0 and download "elastix-5.0.0-linux.tar.bz2" and extract it wherever you like.
2. Open a new terminal and do vi ~/.bashrc
3. Add the following lines to the file:

export PATH=<path to extracted folder>/elastix-5.0.0-Linux/bin:$PATH
export LD_LIBRARY_PATH=<path to extracted folder>/elastix-5.0.0-Linux/lib:$LD_LIBRARY_PATH

4. Close the Terminal.
5. Open a new terminal and do:

cd <path to project>
virtualenv  bpt__nnunet3.8 -p python3.8
virtualenv -p /usr/bin/python3.8 bpt__nnu-net3.8
source bpt__nnunet3.8/bin/activate
pip install --trusted-host files.pythonhosted.org --trusted-host pypi.org --trusted-host pypi.python.org -r ./requirements.txt -vvv
pip install torch torchvision torchaudio 

Prepare the data

Before executing, make sure that the directory you are working on is orgenized as following:

<FOLDER FOR CASES>/
├── <FOLDER SUBJECT 1>
│   └── <FOLDER STUDY 1>
│       └── nnUNetTrainerV2__nnUNetPlansv2.1
│           ├── FLAIR (folder or nii)
│           ├── T1 (folder or nii)
│           ├── T1C (folder or nii)
│           ├── T2 (folder or nii)
│       └── <FOLDER STUDY 2>
├── <FOLDER SUBJECT 2>

This pipline can only work with FLAIR, T1, T1C, T2 modalities. You don't need to have all the modalities. Make sure that you have the needed modalities for registration (default is FLAIR), brain extruction(default is T1C), and inference (dependence on the model you chooce)! or the process will fail!!!

Run inference

For prediction, user must provide a path to a trained nnU-Net model.

set CUDA_VISIBLE_DEVICES=<number of gpu>
python <path to project>/gui-test.py 

For both preprocessing and segmentation, The output is saved to /BET_SCANS directory.