QuickNAT_tensorflow_v2

Tool: QuickNAT: Segmenting MRI Neuroanatomy in 20 seconds

If you use this code please cite:

Guha Roy, A., Conjeti, S., Navab, N., and Wachinger, C. 2018. QuickNAT: Segmenting MRI Neuroanatomy in 20 seconds. Accepted for publication at NeuroImage.

Link to paper: https://arxiv.org/abs/1801.04161

• Tensorflow Implementation is contributed by IshmeetKaur

• Pytorch Implementation is contributed by Shayan Ahmad Siddiqui

You can request the OASIS 30 dataset with manual labels in MICCAI 2012 Grand Challenge on Multi-Atlas Labeling

Changes

• Add the matlab code create_wholedataset.m to build imdb structure for any nifti formats dataset.
• Provide the code to remap the labels of OASIS dataset in order to reduce the total number of classes.
• Provide the code to calculate the class weights for the weighted_cross_entrophy.
• Add more comments on the original tensorflow code to help understand.

Simple to run!

Run in cluster

• Request a interative CHPC GPU node

qsub -I -l nodes=1:ppn=1:gpus=1:V100,walltime=1:00:00

• Load the module for CUDA, Singularity and run tensorflow

module load cuda-8.0
module load singularity-2.4.2
singularity exec --nv /export/tensorflow-1.7.0/test/ubuntu_tf_gpu python3 /home/caelyn/QuickNAT_tensorflow/training.py

Run in local machine

• After activating Tensorflow environment, simply run

python3 training.py 

• Testing

python3 testing.py

Troubleshooting

• The error was shown as 'Unable to create file (file locking disabled on this file system, use HDF5_USE_FILE_LOCKING environment variable to override)' can be solved by adding the flag:

export HDF5_USE_FILE_LOCKING=FALSE