Awesome

RED_CNN

Implementation of Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)

There is several things different from the original paper.

The input image patch(64x64 size) is extracted randomly from the 512x512 size image. --> Original : Extract patches at regular intervals from the entire image.
use Adam optimizer

DATASET

The 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge by Mayo Clinic
(I can't share this data, you should ask at the URL below if you want)
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

Check the arguments.

run python prep.py to convert 'dicom file' to 'numpy array'
run python main.py --load_mode=0 to training. If the available memory(RAM) is more than 10GB, it is faster to run --load_mode=1.
run python main.py --mode='test' --test_iters=100000 to test.

RESULT