<– main page

---

You can download the Singularity container here: link.

While you can find the docker here: link.

Testing tutorial

Steps for singularity (similar to docker):

1. Download the container and the code from Github

   wget https://osf.io/download/v8g9x/ && mv index.html LOD_Brain.simg

   git clone https://github.com/rockNroll87q/LOD-Brain

2. Prepare your testing data. As an example, I download a volume from a public repository (link):

   wget https://s3.amazonaws.com/openneuro.org/ds002207/sub-01/anat/sub-01_T1w.nii.gz

   wget https://s3.amazonaws.com/openneuro.org/ds002207/sub-02/anat/sub-02_T1w.nii.gz

3. Run the container:

    singularity exec --cleanenv --nv \
    -B {PATH_TO_YOUR_TESTING_DATA}:/data/ \
    -B ./LOD-Brain/src:/source/ \
    -B ./LOD-Brain/trained_model/:/model/ \
    -B {WHERE_YOU_WANT_TO_SAVE_THE_OUTPUT}:/output/ \
    LOD_Brain.simg python /source/inference.py --vol_in=/data/
   

   Example:

    singularity exec --cleanenv --nv \
    -B ./:/data/ \
    -B ./LOD-Brain/src:/source/ \
    -B ./LOD-Brain/trained_model/:/model/ \
    -B ./:/output/ \
    LOD_Brain.simg python /source/inference.py --vol_in=/data/
   

   Example (docker):

    docker run -it --rm  \
    -v ~/test_lod_brain/input/T1w_256/:/data/ \
    -v ~/test_lod_brain/LOD-Brain/src/:/source/ \
    -v ~/test_lod_brain/LOD-Brain/trained_model/:/model/ \
    -v ~/test_lod_brain/output/:/output/ \
    rocknroll87q/cerebrum3t-multisite:latest python /source/inference.py --vol_in=/data/
   

   In vol_in you can pass a folder, for example vol_in=/data/, or a volume like vol_in=/data/sub-01_T1w.nii.gz

Training

To train a network from scratch, use the following example:

1. Build the singularity image using docker hub.

    singularity build ./LOD_brain.simg docker://rocknroll87q/cerebrum3t-multisite:latest
   

2. Prepare your csv.

3. Lunch the training (below the default values):

    singularity shell --cleanenv --nv \
     -B ~/training_data/:/LOD_Brain/data/ \
     -B ~/src/:/source/ \
     -B ~/output/:/output/ \
    LOD_brain.simg bash
   	
    export WANDB_DIR='/output/multi_data/'
    cd /source/
   
    python ./main.py --network.num_levels=2 --network.num_initial_filter=8 --network.num_blocks_per_level=3 --network.conv_block=Plain --network.downsampling_factor=4 --training.loss=per_channel_dice_loss --network.conv_repetition=False --network.bn=GN --data.Filename_csv=dataset_short_training_a+a_LODbrain.csv  --data.Path_in_csv=/LOD_Brain/data/analysis/csv/segmentation/ --network.num_classes=7
   

   Please note that you must set up a wandb account and modify the project name in line 65 of main.py.

FAQ

A. If you get the error ModuleNotFoundError: No module named 'tensorflow_addons'

$ singularity shell --cleanenv --nv LOD_Brain.simg
$ python -m pip install tensorflow-addons