Synapse-Aware Skeleton Generation

The skeleton benchmark and code for the 2019 MICCAIR paper Synapse-Aware Skeleton Generation for Neural Circuits [1]. For more information: https://www.rhoana.org/synapseaware.

Installation

git clone https://github.com/Rhoana/synapseaware.git .
cd synapseaware
conda create -n synapseaware_env python=3.7
conda install --file requirements.txt
cd connectome
python setup.py build_ext --inplace
cd ../isthmus 
python setup.py build_ext --inplace
cd ../teaser
python setup.py build_ext --inplace

Add the parent directory to this repository to your PYTHONPATH variable.

Meta Files

Each new dataset needs a meta file named meta/{PREFIX}.meta where {PREFIX} is a unique identifier for the dataset. All functions in this repository require as input this {PREFIX} identifier to find information such as resolution and grid size. An example meta file is provided in examples/meta/Fib25.meta.

Directory Structure

For each new {PREFIX}, you need to create the following directories:

baselines/teasers/{PREFIX}
baselines/topological-thinnings/{PREFIX}
connectomes/{PREFIX}
distances/{PREFIX}
segmentations/{PREFIX}
skeletons/{PREFIX}
somae/{PREFIX}
surfaces/{PREFIX} (optional)
synapses/{PREFIX}
volumetric_somae/segmentations/{PREFIX} (optional)
volumetric_somae/surfaces/{PREFIX}
widths/{PREFIX}

If the segment does not contain the somae, a central point should be chosen as the soma and be copied to the somae and volumetric_somae/surfaces directories. The algorithm also assumes the following directories:

meta
running_times/refinement
running_times/skeletons
width-errors (optional)

The script examples/scripts/create-folders.py creates the intended directory structure.

Example Script

The script examples/scripts/connectome.py extracts the skeleton for an example from the FIB-25 dataset publicly available online [2]. A central point was chosen as the soma location since this particular segment does not contain the cell body.

Input Format

All of the point clouds have the format:

unsigned long: Z Grid Size
unsigned long: Y Grid Size
unsigned long: X Grid Size
unsigned long: number of points 
unsigned long *: points in linear index

The file utilities/dataIO.py contains example input/output operations for this file format. Use the following methods to convert between Cartesian coordinates and the linear index:

# iv is the linear index
# yres is the Y Grid Size
# xres is the X Grid Size
def IndexToIndices(iv):
	iz = iv // (yres * xres)
	iy = (iv - iz * yres * xres) // xres
	ix = iv % xres
	return iz, iy, ix

def IndicesToIndex(ix, iy, iz):
	return iz * yres * xres + iy * xres + ix

Citations

[1] Matejek, B., Wei, D., Wang, X., Zhao, J., Palágyi, K., and Pfister, H., Synapse-Aware Skeleton Generation for Neural Circuits. In *International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. YYY-YYY. Springer, Cham, 2019.

[2] Takemura, S.Y., Xu, C.S., Lu, Z., Rivlin, P.K., Parag, T., Olbris, D.J., Plaza, S., Zhao, T., Katz, W.T., Umayam, L. and Weaver, C., 2015. Synaptic circuits and their variations within different columns in the visual system of Drosophila. Proceedings of the National Academy of Sciences, 112(44), pp.13711-13716.