Generative Adversarial Network for Medical Imaging

MI-GAN

##Data

Copy your data into the two sub-directories DRIVE and STARE inside the directory Data. To save space, we have removed any image data from many of the sub-directories however, we have kept the empty directory there so that you may still be able to use the same path for a sub-directory. This is to help you to retain the hierarchy of the folders.

##Credits

Contribution from Talha Iqbal and Hazrat Ali

The following code is associated to the paper

Talha Iqbal and Hazrat Ali, Generative Adversarial Network for Medical Images (MI-GAN), Journal of Medical System, November 2018, 42:231. Doi: 10.1007/s10916-018-1072-9

Avaiable at: [https://link.springer.com/article/10.1007/s10916-018-1072-9]

PDF arxiv link: [https://arxiv.org/abs/1810.00551]

#How to use this code.

Package Dependency

• tensorflow 1.16.0
• python 3.5.3
• numpy 1.14.2
• matplotlib 2.0.2
• pillow 5.0.0
• scikit-image 0.13.0
• scikit-learn 0.19.0
• scipy 0.19.0

Download Data

Original data file strucure was modified for convenience by Jaemin Son.
Download data from here and copy data file in the same directory with codes file as following Directory Hierarchy.

Directory Hierarchy

.
+-- codes
¦   +-- dataset.py
¦   +-- evaluation.py
¦   +-- main.py
¦   +-- model.py
¦   +-- solver.py
¦   +-- TensorFlow_utils.py
¦   +-- utils.py
+-- data
¦   +-- DRIVE
¦   +-- STARE
+-- evaluation (get after running evaluation.py)
¦   +-- DRIVE
¦   +-- STARE
+-- results
¦   +-- DRIVE
¦   +-- STARE

codes: source codes
data: original data. File hierarchy is modified for convenience.
evaluation: quantitative and qualitative evaluation. (get after running evaluation.py)
results: results of other methods. These image files are retrieved from here.

Training

Move to codes folder and run main.py

python main.py --train_interval=<int> --ratio_gan2seg=<int> --gpu_index=<int> --discriminator=[pixel|patch1|patch2|image] --batch_size=<int> --dataset=[DRIVE|STARE] --is_test=False

• models will be saved in './codes/{}/model_{}_{}_{}'.format(dataset, disriminator, train_interval, batch_size)' folder, e.g., './codes/STARE/model_image_100_1' folder.
• smapled images will be saved in './codes/{}/sample__{}_{}_{}'.format(dataset, disriminator, train_interval, batch_size)', e.g., './codes/STARE/sample_image_100_1' folder.

Arguments

train_interval: training interval between discriminator and generator, default: 1
ratio_gan2seg: ratio of gan loss to seg loss, default: 10
gpu_index: gpu index, default: 0
discriminator: type of discriminator [pixel|patch1|patch2|image], default: image
batch_size: batch size, default: 1
dataset: dataset name [DRIVE|STARE], default: STARE
is_test: set mode, default: False

learning_rate: initial learning rate for Adam, default: 2e-4
beta1: momentum term of Adam, default: 0.5
iters: number of iterations, default: 50000
print_freq: print loss information frequency, default: 100
eval_freq: evaluation frequency on validation data, default: 500
sample_freq: sample generated image frequency, default: 200

checkpoint_dir: models are saved here, default: './checkpoints'
sample_dir: sampled images are saved here, default: './sample'
test_dir: test images are saved here, default: './test'

Test

python main.py --is_test=True --discriminator=[pixel|patch1|patch2|image] --batch_size=<int> --dataset=[DRIVE|STARE]

• Outputs of inferece are generated in 'seg_result_{}_{}_{}'.format(discriminator, train_interval, batch_size) folder, e.g., './codes/STARE/seg_result_image_100_1' folder.
• Make sure model already trained with defined dataset, discriminator, training interval, and batch size.

Evaluation

Note: Copy predicted vessel images to the ./results/[DRIVE|STARE]/V-GAN folder

python evaluation.py

Results are generated in evaluation folder. Hierarchy of the folder is

.
+-- DRIVE
¦   +-- comparison
¦   +-- measures
¦   +-- vessels
+-- STARE
    +-- comparison
    +-- measures
    +-- vessels

comparison: difference maps between V-GAN and gold standard
measures: AUC_ROC and AUC_PR curves
vessels: vessels superimposed on segmented masks
Area Under the Curve (AUC), Precision and Recall (PR), Receiver Operating Characteristic (ROC)

#Support

We would be offer any possible support as long as it does not have conflict with our own work committements. 

#Disclaimer

In no event shall we be liable to any party for direct, indirect, special, incidental or consequential damages, including lost profis, arising out of the use of this software code and its documentation.