Vascu Synth is a software for simulating 3D images of blood vessels. They are produced as a series of png-images in a folder. The images can be used, for example, as test images for 3D image analysis or to generate datasets for use in deep learning. This repository contains a clone of the software together with installation instrucations and scripts that simplify usage.

I am not the author of the software. Please refer to the following links for the official documentation.

Code was originally posted at
https://github.com/midas-journal/midas-journal-794.
The project website is http://vascusynth.cs.sfu.ca.
The following research papers provide additional information:
http://dx.doi.org/10.1016/j.compmedimag.2010.06.002
http://dx.doi.org/10380/3260

'Sources' contains the vascu synth C++ sources cloned from the link above. The only change made to them is to output png-images instead of jpg and to add an extra digit to the filenames to possibly allow stacks with more than 1000 images.

'example_config' contains example config files and a call that can be used for testing after compiling the sources.

'python_scripts' contains scripts to

• generate pseudo-random valid config files (vascu synth is quite picky with the format).
• split config files for cheap parallelization (see section on 'run_split' below)
• (combine the pngs and view 3D files (requires a self-written viewer not included here))

'run_split': The software uses only a single CPU and slowly filled up our system's memory while running. For cheap parallelization and as a workaround for the memory leak, generating a dataset was split into several calls to Vascu Synth. 'run_split' contains 5 shell scripts that can be called from different terminals (or screens) to generate a dataset of up to 2000 images using 5 CPUs and clearing memory every 10 images on each CPU. They will work most efficiently, if datasets are generated in multiples of 200 images. This folder also contains a script that automates the calls to the 5 shell scripts and to python scripts.

('matlab_scripts' contains matlab-scripts for generating a dataset that can be used for deconvolution. This makes use of a number of toolboxes that may not be available on every system.)

Compiling VascuSynth is described below. This requires a C-compiler, cmake and the Insight Toolkit (ITK). Note that Vascu Synth is from 2011 and may not support the newest versions of these.

The recommended pipeline for generation of a dataset of pseudo-random objects is running the generate_dataset script. The shell scripts have been tested in bash. The config generators called from the scripts requires Python (tested with Python 3.6) and some python-packages.

(The script to view the objects in python requires an additional self-written viewer not included here. The matlab-scripts require matlab and the following

• https://github.com/sziem/matlab_tools
• Toolboxes from our research group: https://nanoimaging.de/
• DipImage http://www.diplib.org/ Please contact us, if you are interested in using these scripts.)

1. run ./generate_dataset.sh SEED N_IMAGES(200 by default)
   (--> continue with step 6 to use the matlab scripts)

1. change seed (and possibly other parameters) in generate_vascu_config.py
2. run generate_vascu_config.py
3. run split_and_run.sh
4. wait for run split_and_run.sh to finish
5. rename CHANGE_NAME-folder, eg. to the number of the random seed used
   (--> continue with step 6 to use the matlab scripts))

6. run matlab_scripts/create_simulated_data_pairs.m to generate pairs of ground truth samples (obj) and simulated microscopic images (img)

Installation instruction and official requirements can be found in
http://dx.doi.org/10380/3260

We were able to compile successfully with cmake-3.10 and Insight Toolkit (ITK) 4.13 (with compatibility to ITK 3 set in CMake), but there were some issues. Particularly note the modified CMakeLists.txt.

After compilation, you may test with the configuration provided in 'example_config'.