Python code for converting sharp-kernel reconstructed CT images to soft-kernel images

This is a companion code for the paper

• Kernel Conversion for Robust Quantitative Measurements of Archived Chest Computed Tomography Using Deep Learning-Based Image-to-Image Translation by Naoya Tanabe, Shizuo Kaji, Hiroshi Shima, Yusuke Shiraishi, Tsuyoshi Oguma, Susumu Sato, Toyohiro Hirai, Frontiers in Artificial Intelligence Medicine and Public Health, DOI: 10.3389/frai.2021.769557

The code is based on

• Image-to-image translation by CNNs trained on paired data which is described in the 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

Requirements

• python 3: Anaconda is recommended
• chainer >= 7.2.0, chainercv, opecv, pydicom: install them by

pip install -U git+https://github.com/chainer/chainer.git
pip install -U chainercv opencv-contrib-python pydicom

(optional, but recommended) To use GPU, you have to install CUDA and CuPy.

• CUDA: follow the instruction here
• CuPy: follow the instruction here

How to use

• Arrange DICOM files into the directory structure as in the following example: (phantom images are included as demo)

sharp_kernel
+-- patient1
|   +-- patient1_001.dcm
|   +-- patient1_002.dcm
|   +-- ...
+-- patient2
|   +-- patient2_001.dcm
|   +-- patient2_002.dcm
|   +-- ...

• Execute the following commands from the terminal/command prompt:

    python convert.py -a args_partial -R sharp_kernel -o partial
    python convert.py -a args_full -R sharp_kernel -o full
    python dicom_overlay.py -i0 sharp_kernel -i1 full -i2 partial -o converted

• The first line creates a (temporary) directory named partial which contains converted images by the partial model (that is, CT values are cropped to -300 to 300 HU).
• The second line creates a (temporary) directory named full which contains converted images by the full model.
• The third line creates a directory named converted which contains the final converted images obtained by fusing the above two.

Licence

MIT Licence