minc_keras is a code base that was developed during a hackathon to facillitate the implementation of deep learning models for brain imaging with the Keras package. It is also used as a hands-on teaching tool for the presentations listed below.
Presentations were created in collaboration with the MCIN lab and NeuroTechX. NeuroTechX is a non-profit organization whose mission is to facilitate the advancement of neurotechnology by providing key resources and learning opportunities, and by being leaders in local and worldwide technological initiatives. Their 3 pillars are “Community”, “Education”, and “Professional Development”.
Create / Log-in to Google account
Go to https://colab.research.google.com
Download and load: https://www.dropbox.com/s/8uw13lbwbf83c0d/NeuroTech_MTL_28_8_18.ipynb?dl=0
Install docker on your OS: https://docs.docker.com/install/#cloud
docker pull tffunck/neurotech:latest
wget https://bootstrap.pypa.io/get-pip.py (Or go to the link and download manually)
python3 get-pip.py
pip3 install pandas numpy scipy h5py matplotlib tensorflow keras
git clone https://github.com/tfunck/minc_keras
Data should be organized in the BIDS format (http://bids.neuroimaging.io/). While the code in this repository is in theory supports HDF5 files, at the moment only the MINC format is supported. Nifti support will be provided in future releases.
data/output/
data/output/sub-01/sub-01_task-01_ses-01_T1w_anat_rsl.mnc
data/output/sub-01/sub-01_task-01_ses-01_variant-seg_rsl.mnc
data/output/sub-02/sub-02_task-01_ses-01_T1w_anat_rsl.mnc
data/output/sub-02/sub-02_task-01_ses-01_variant-seg_rsl.mnc
python3 minc_keras/minc_keras.py --source /path/to/your/data/ --target /path/to/desired/output --epochs --input-str "string that identifies input files" --label-str "string that identifies labeled files" --predict
python3 minc_keras/minc_keras.py --source minc_keras/data/output/ --target . --epochs 5 --input-str "T1w_anat" --label-str "seg" --predict 1
Thomas Funck (thomas.funck@mail.mcgill.ca)
Paul Lemaitre
Andrew Doyle