Volume-to-volume translation by CNNs trained on paired data

written by Shizuo KAJI

CAUTION: This code is at an early beta stage!

This is an implementation of volume-to-volume translation using a paired volume dataset. It can be used for various tasks including

segmentation, denoising, super-resolution, modality-conversion, and reconstruction

The details can be found in our paper: "Overview of image-to-image translation using deep neural networks: denoising, super-resolution, modality-conversion, and reconstruction in medical imaging" by Shizuo Kaji and Satoshi Kida, Radiological Physics and Technology, Volume 12, Issue 3 (2019), pp 235--248, arXiv:1905.08603

If you use this software, please cite the above paper.

Background

This code is based on

Requirements

a modern GPU with a large memory (the crop_size should be reduced for GPUs with less than 20G memory)
python 3: Anaconda is recommended
chainer >= 7.2.0, cupy, chainerui, chainercv, opecv, pydicom, pynrrd: install them by

pip install cupy chainer chainerui chainercv opencv-contrib-python pydicom pynrrd

Note that with GeForce 30 RTX series, the installation of chainer and cupy can be a little tricky for now. You need CUDA >= 11.1 for these GPUs, and it is supported by CuPy >= v8. The latest version of Chainer v7.7.0 available on pip is not compatible with the latest version of CuPy. See here. You can install the latest Chainer directly from the github repository, which is compatible with the latest version of CuPy. For example, follow the following procedure:

Install CUDA 11.1
pip install cupy-cuda111
pip install -U git+https://github.com/chainer/chainer.git

You will see some warning messages, but you can ignore them.

Licence

MIT Licence

Training Data Preparation

data_cn4
+-- trainA
|   +-- patient1
|   |   +-- patient1_001.dcm
|   |   +-- patient1_002.dcm
|   |   +-- ...
|   +-- patient2
|   +-- ...
+-- trainB
|   +-- patient1
|   |   +-- patient1_001.dcm
|   |   +-- patient1_002.dcm
|   +-- patient2
|   +-- ...
+-- testA
|   +-- patient3
|   +-- patient4
+-- testB
|   +-- patient3
|   +-- patient4

The model learns to convert image domains from trainA to trainB. testA and testB are used for evaluation during training. Each directory, for example patient1, contains a series of DICOM files to form a volume.

Training

Example:

python train_cgan3d.py -R data_cn4 -l1 0 -l2 0 -ldice 10 -ldis 0 -ca -1200 100 -vf 100 -u concat -gu resize -rt 8 -e 1000 -cn 4 --class_weight 1 1 100 1000 -srf 2

Conversion with a trained model

python convert.py -R data_cn4/testA -a result/2021???/args