This work aims to explore a new approach to model the encephalic vasculature using the formalism of graphs that naturally fit the structure of blood vessels.
It is necessary to install:
- Keras with TensorFlow backend
- h5py
- scipy
- sklearn
- numpy
- tifffile
- argparse
- progressbar
To obtain the graph follow the procedure below:
- If necessary make_train_set.py (folder: augment-train-set) performs an augmenting operation over the train-set
- Use train_vascular.py to train the neural network
- Use predict.py to segment the images
- It is possibile to use overlap_img.py (folder: pre-process-denoising) to overlap the predicted image (It is possibile that this operation reduces the noise)
- Use prepare_tif_stack.py to filter and binarize the predicted images
- Use the Fiji's (Fiji is just ImageJ) plug-in to create the graph that models the predicted images. To run the plug-in move in the directory containing the launcher and the run (command for linux)
$ ./ImageJ-linux64 --ij2 --headless --run path/to/script 'img_file="path/to/image",ouput="path/to/output/folder"'
Fiji: https://fiji.sc/#
Headless mode: http://imagej.net/Scripting_Headless - Use cluster_meanshift.py (folder: post-process-denoising) to clean the graph from noises
NB use -h to see the script's helps