During classes of Medical images analaysis on Faculty of Electrotehnics of University of Ljubljana, I tried to make super resolution model of MRI (Magnetic resonance imaging) with deep learning. I used source code from github (https://github.com/krasserm/super-resolution), which was modified for my aplication. I used EDSR and SRGAN architectures for learning neural network. I lerned two models. One dataset was made artificial with high resolution image downgraded to get low resolution image pair. Second were real pairs of LR and HR images.
On this Drive there are all neccesary folders, code, images etc.: https://drive.google.com/drive/folders/1Kmn9bVfWjjyrB5lL8VHOV3vP3K134n3a?usp=sharing
real_images -> 3D MR images of 182 patients. Every subfolder have 4 3D images one is HR (high resolution) with space sampling of 1x1x1mm. The other three are LR images which have 3mm space sampling along one axis (sagital, transversal and coronal).
synthetic_images -> 28 T1 MR images, 28 FLAIR MR images, 36 LR T1 MR images in 8 T1 raw LR MR images.
dataset_MRI_real -> This floder contains HR-LR real pairs (5000 image pairs). There is also prepared numpy array in dataset_split sub-folder, for learning the model (train and test data).
dataset_MRI_synthetic -> the same as previous folder except there are synthetic made HR-LR pairs (2000 image pairs)
model -> contains some useful functions for SRGAN model and resolving LR images with trained model
validation -> contain statistic data of validation on both data groups real and synthetic (PSNR and SSIM metrics)
weights_SR -> weights of models I trained.
data_preparation_synthetic.ipynb: -> This notebook is used for extracting HR slices from 3D images stored in synthetic_images. It also contains making numpy array of data for training and testing.
REALdata_preparation.ipynb: -> This notebook is used for extracting HR-LR slices pairs from 3D images stored in dataset_MRI_real. It also contains making numpy array of data for training and testing.
training_REAL_DATA.ipynb: ->This notebook is used for training on prepared dataset stored in dataset_MRI_real. It contains both EDSR and SRGAN model.
training_synthetic.ipynb:->This notebook is used for training on prepared dataset stored in dataset_MRI_synthetic. It contains both EDSR and SRGAN model.
VALIDATION.ipynb: -> This notebook is used for validating on test data which are stored in dataset_MRI_real and dataset_MRI_synthetic.
unzip_zip_file.ipynb:-> Just useful notebook for zip/unzip files.
During my work I used Jupyter notebooks in Google Colab environment. So all my notebooks are prepared to run in Google Colab, they also include pip install for all necessary Python libraries that aren't installed by default. In req.txt file are all libraries that were installed in Google Colab. You don't need all of these but I copy all to the file to make sure.
The easiest way for runinng those scripts is to store all files and folders in one place like in my Google Drive (link above). All you have to watch out is if all directory in scripts are correct. If you do this, running those scripts should not be an issue.
