ldct_nonlocal

Official Source Code for the paper Noise Conscious Training of Non Local Neural Network Powered by Self Attentive Spectral Normalized Markovian Patch GAN for Low Dose CT Denoising.

Dependencies:

Pytorch

Pydicom

DATASET

The 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge by Mayo Clinic
https://www.aapm.org/GrandChallenge/LowDoseCT/

The data_path should look like:

data_path
├── L067
│   ├── quarter_3mm
│   │       ├── L067_QD_3_1.CT.0004.0001 ~ .IMA
│   │       ├── L067_QD_3_1.CT.0004.0002 ~ .IMA
│   │       └── ...
│   └── full_3mm
│           ├── L067_FD_3_1.CT.0004.0001 ~ .IMA
│           ├── L067_FD_3_1.CT.0004.0002 ~ .IMA
│           └── ...
├── L096
│   ├── quarter_3mm
│   │       └── ...
│   └── full_3mm
│           └── ...      
...
│
└── L506
    ├── quarter_3mm
    │       └── ...
    └── full_3mm
            └── ...     

USE

use the prep.py file to generate the traning and test data.

update the data path in train_model.py and test_model.py

run the file train_model.py to train

Acknowledgments

• Our code architecture is inspired by JiahuiYu and SSinyu.