panakino / 3dbpconv

tensorflow implementation for 3D BPConvNet

3D BPConvNet

3D parallel MRI reconstruction for accelerated MRI. In our paper ISBI 2018, we reconstructed 320x320x256x8 volume using 3D BPConvNet. It takes sub 10 seconds. We also used wavelet transform from built-in of Matlab 2016, however, in this repo, we do not provide wavelet transform. The codes only cover the network (after wavelet decomposition and before wavelet synthesis).

Whole codes are forked from https://github.com/jakeret/tf_unet.

Training configuration

• Tensorflow 1.1.0
• 2 GPUs (TITAN X pascal arch.)
• MacOS X 10.12.6
• Python 2.7.12
• We used Knee dataset from http://mridata.org.
• 17 subj for training (x8 with augmentation)/ 3 subj for testing
• Data format : NCXYZ (batch x channel x X x Y x Z)

illustration

Commands

Before starting,

pip install pillow matplotlib h5py

To start training a model for 3D BPConvNet:

python main.py --lr=1e-4 --output_path='logs/' --data_path='data_path/*.h5' --test_path='test_path/*.h5' --features_root=32 --layers=5 --is_training=True

To deploy trained model:

python main.py --lr=1e-4 --output_path='logs/' --data_path='data_path/*.h5' --test_path='test_path/*.h5' --features_root=32 --layers=5 --is_training=False

You may find more details in main.py.

Contact

[e-mail]