(Python) utility to filter mgz volumes to per-voxel-value directories of jpg/png image slices

## Quick Overview

• Filters mgz volume files by voxel value to well organized directories of image files.

## Overview

mgz2imgslices is a simple Python utility that fiters “labels” from mgz volume files and saves each label set as slices of (by default) png files, organized into a series of directories, one per label set.

An mgz format file simply contains a 3D volume data structure of image values. Often these values are interpreted to be image intensities. Sometimes, however, they can be interpreted as label identifiers. Regardless of the interpretation, the volume image data is simply a number value in each voxel of the volume.

This script will scan across the input mgz volume, and for each voxel value create a new output directory. In that directory will be a set of numpy arrays (.npy files), one per slice of the original volume. These numpy arrays will only contain the voxel values in the original dataset that all had that particular voxel value.

In this manner, mgz2imgslices can also be thought of as a dynamic filter of an mgz volume file that filters each voxel value into its own output directory of .npy files.

## Dependencies

Make sure that the following dependencies are installed on your host system (or even better, a python3 virtual env):

• pfmisc : (a general miscellaneous module for color support, etc)
• nibabel : (to read NIfTI files)
• numpy : (to support large, multidimensional arrays and matrices)
• imageio : (interface to read and write image data)
• pandas : (data manipulation and analysis)
• re : (support for regular expressions)
• time : (support for various time related functions)

## Assumptions

This document assumes UNIX conventions and a bash shell. The script should work fine under Windows, but we have not actively tested on that platform – our dev envs are Linux Ubuntu and macOS.

### Python module

One method of installing this script and all of its dependencies is by fetching it from PyPI.

pip3 install mgz2imgslices

### Docker container

We also offer a docker container of mgz2imgslices as a ChRIS-conformant platform plugin here https://github.com/FNNDSC/pl-mgz2imgslices – please consult that page for information on running the dockerized container. The containerized version exposes a similar CLI and functionality as this module.

## How to Use

mgz2imgslices needs at a minimum the following required command line arguments:

• -i | --inputFile <inputFile>: The input .mgz file to convert.
• -d | --outputDir <outputDir>: The output directory. This in turn will contain several subdirectores, one per image voxel value in the input mgz file. Each of these sub directories will contain npy files, filtered to that voxel value.
• -o | --outputFileStem <outputFileStem> : The name of the output files within the label directories (numpy arrays and png/jpg images)

NOTE:

• The --lookupTable arg for this Python utility requires that you pass FreeSurferColorLUT.txt (or another LUT file of the same format) which should be present within the inputDir.
• If you are using the docker image (visit https://github.com/FNNDSC/pl-mgz2imgslices) to run this utility, you can use either __fs__ or __val__ to the --lookupTable argument.

## Examples

First, let’s create a directory, say devel wherever you feel like it. We will place some test data in this directory to process with this plugin.

cd ~/
mkdir devel
cd devel
export DEVEL=$(pwd) Now, we need to fetch sample MGZ data. ### Pull mgz data git clone https://github.com/FNNDSC/mgz_converter_dataset.git Make sure the mgz_converter_dataset directory is placed in the devel directory. • Make sure your current working directory is devel. At this juncture it should contain mgz_converter_dataset. • Create an output directory named results in devel. mkdir results && chmod 777 results #### EXAMPLE 1 • Run mgz2imgslices using the following command. Change the arguments according to your need. mgz2imgslices -I${DEVEL}/mgz_converter_dataset/100307/                              \
--inputFile aparc.a2009s+aseg.mgz                                       \
--outputDir \${DEVEL}/results/                                          \
--outputFileStem sample                                                \
--outputFileType png                                                   \
--saveImages
--label label                                                          \
--wholeVolume FullVolume                                               \
--lookupTable FreeSurferColorLUT.txt                                                  \
--skipLabelValueList 0,4,7

The skipLabelValueList will skip any voxels in the input mgz that have numerical values of, in this case, 0, 4, 7. Note that each output filtered directory will have a name prefix string of label and should appear something similar to:

results/label-Left-Cerebral-White-Matter/sample-000.png
...
results/label-Left-Cerebral-White-Matter/sample-00255.png

...
...

results/label-ctx_rh_S_temporal_transverse/sample-000.png
...
results/label-ctx_rh_S_temporal_transverse/sample-00255.png

## Command Line Arguments

ARGS

[-i|--inputFile  <inputFile>]
Input file to convert. Should be an mgz file.

[-o|--outputFileStem <outputFileStem>]
The output file stem to store image conversion. If this is specified
with an extension, this extension will be used to specify the
output file type.

[-t|--outputFileType <outputFileType>]
The output file type. If different to <outputFileStem> extension,
will override extension in favour of <outputFileType>.

Should be a pngonly.

[--saveImages]
If specified as True(boolean), will save the slices of the mgz file as
".png" image files along with the numpy files.

[--label <prefixForLabelDirectories>]
Prefixes the string <prefixForLabelDirectories> to each filtered
directory name. This is mostly for possible downstream processing,
allowing a subsequent operation to easily determine which of the output
directories correspond to labels.

[-n|--normalize]
If specified as True(boolean), will normalize the output image pixel values to
0 and 1, otherwise pixel image values will retain the value in
the original input volume.

[-l|--lookupTable <LUTfile>]
Need to pass a <LUTfile> (eg. FreeSurferColorLUT.txt)
to perform a looktup on the filtered voxel label values
according to the contents of the <LUTfile>. This <LUTfile> should
conform to the FreeSurfer lookup table format (documented elsewhere).

Note that the special <LUTfile> string __val__, __fs__ or __none__
can be passed only when running the docker image (fnndsc/pl-mgz2imageslices)
of this utility which effectively means "no <LUTfile>".
In this case, the numerical voxel values are used for output directory names.
This special string is really only useful for scripted cases of running
this application when modifying the CLI is more complex than simply setting
the <LUTfile> to __val__.

While running the docker image, you can also pass __fs__ which will use
the FreeSurferColorLUT.txt from within the docker container to perform a
looktup on the filtered voxel label values according to the contents of
the FreeSurferColorLUT.txt

[--skipAllLabels]
Skips all labels and converts only the whole mgz volume to png/jpg images.

[-s|--skipLabelValueList <ListOfLabelNumbersToSkip>]
If specified as a comma separated string of label numbers,
will not create directories of those label numbers.

[-f|--filterLabelValues <ListOfVoxelValuesToInclude>]
The logical inverse of the [skipLabelValueList] flag. If specified,
only filter the comma separated list of passed voxel values from the
input volume.

The detault value of "-1" implies all voxel values should be filtered.

[-w|--wholeVolume <wholeVolDirName>]
If specified, creates a diretory called <wholeVolDirName> (within the
outputdir) containing PNG/JPG images files of the entire input.

This effectively really creates a PNG/JPG conversion of the input
mgz file.

Values in the image files will be the same as the original voxel
values in the mgz, unless the [--normalize] flag is specified
in which case this creates a single-value mask of the input image.

[-h|--help]
If specified, show help message and exit.

[--json]
If specified, show json representation of app and exit.

[--man]
If specified, print (this) man page and exit.

[--meta]
If specified, print plugin meta np_data and exit.

[--savejson <DIR>]
If specified, save json representation file to DIR and exit.

[-v <level>|--verbosity <level>]
Verbosity level for app. Not used currently.

[--version]
If specified, print version number and exit.

[-y|--synopsis]
Show short synopsis.