fnndsc/ep-subdivide-mnc-methods repository overview

⁠Subdivide MINC Volume

This repository contains several wrapper scripts for increasing the resolution of MINC volumes by subdividing voxels along their edges and performing interpolation.

It can be ran as a ChRIS ds-plugin, in which case it applies several methods of subdivision to every MINC file found in its input directory, writing outputs to a given output directory. These results are quantified and the aggragate statistics are written to summary.json.

⁠Installation

ep-subdivide-mnc-methods is a ChRIS⁠ plugin, meaning it can run from either within ChRIS or the command-line.

⁠Using Apptainer

To get started with local command-line usage, use Apptainer⁠ (a.k.a. Singularity) to run ep-subdivide-mnc-methods as a container:

singularity exec docker://fnndsc/ep-subdivide-mnc-methods subdivide [--args values...] input/ output/

To print its available options, run:

singularity exec docker://fnndsc/ep-subdivide-mnc-methods subdivide --help

⁠Scripts

⁠subdivide_minc.py

Wrapper around mincresample (part of minc tools⁠).

⁠Example of subdivide_minc.py

./subdivide_minc.py --divisions=8 input_volume.mnc output_subdivided_linear.mnc
# or
./subdivide_minc.py --divisions= --options=-cubic input_volume.mnc output_subdivided_cubic.mnc

Flags can be passed directly to mincresample using --options=.... See man mincresample⁠ for options.

⁠Understanding Interpolation Options

mincresample interpolates values along voxel boundaries to smoothen the output. The options are -trilinear (default for registration step in CIVET), -tricubic, -nearest_neighbour, and -sinc.

⁠External Reading

• https://maidens.github.io/jekyll/update/2016/08/10/An-illustrated-guide-to-interpolation-methods.html⁠
• https://graphicdesign.stackexchange.com/questions/26385/difference-between-none-linear-cubic-and-sinclanczos3-interpolation-in-image⁠

⁠Notes for Surface Extraction

• nearest-neighbor has no blurring
• sinc and cubic are similar
• linear and cubic are smooth
• linear might be preferable over cubic for surface extraction because tight fitting around voxels is the cause of quality problem, not smoothness of mask edge

⁠subdivide_nibabel.py

Increases the resolution of a volume without interpolation by computing the Kronecker product⁠.

⁠Install dependencies for subdivide_nibabel.py

pip install numpy nibabel h5py
# or
conda install numpy nibabel h5py

Optional dependency on nii2mnc for output to MINC file format.

⁠Example of subdivide_nibabel.py

./subdivide_nibabel.py --divisions 4 input_volume.nii subdivided_output.nii

Both MINC and NIFTI file formats are supported.

⁠Development

Instructions for developers.

⁠Local Testing

Run tests using pytest inside a container.

docker build -t localhost/fnndsc/ep-subdivide-mnc-methods:dev --build-arg extras_require=dev .
docker run --rm -it -u "$(id -u):$(id -g)" \
  -v "$PWD/examples:/examples:ro" \
  -v "$PWD:/app:ro" -w /app \
  localhost/fnndsc/ep-subdivide-mnc-methods:dev \
  pytest -v -o cache_dir=/tmp/pytest