Refactoring

[schuhschuh]

Refactor Christian's MALPEM package to make it a MIRTK extension package which uses MIRTK commands instead of deprecated IRTK binaries.

MIRTK replacements

• seg_maths (NiftySeg): mirtk.calculate_element_wise
  • -bin: use (current image>0) to binarise
    • seg_maths image.nii -bin mask.nii
    • mirtk.calculate_element_wise('image.nii', '-threshold-le', 0, '-pad', 0, '-set', 1, '-out', 'mask.nii')
  • -thr
  • -uthr
  • -mul
  • -add
  • -sub
  • -div
• cl_create4D: mirtk.combine_images
• cl_pairwiseSymDOF: mirtk.average_dofs
• cl_apply_mask
  • mirtk.calculate_element_wise('scan.nii', '-mask', 'mask.nii', '-pad', -1, '-out', brain.nii')
• convert: mirtk.convert_image
• cl_extract_frame: mirtk.extract_image_volume
• cl_averageDOFs: mirtk.average_dofs
• dofinvert: mirtk.invert_dof
• dofcombine: mirtk.compose_dofs
• ireg: mirtk.register
• transformation: mirtk.transform_image
• display: Use mirtk.rendering.screenshots module for report generation.

Custom tools

• cl_compute_volume: Consider adding this to core MIRTK applications
• cl_gaussian_fusion
• cl_malpem
• cl_malpem4D
• pincram-0.2.3_ireg.sh (see below)
• scale.sh (see below)

External dependencies

• N4 (also used by Draw-EM package, see here... though binary shouldn't be in repo)

[schuhschuh]

PINCRAM script:

• ireg: mirtk register
• seg_maths operations: -div, -thr, -bin: mirtk calculate-element-wise
• dofcombine: mirtk compose-dofs
• transformation: mirtk transform-image
• headertool
  • -origin <x> <y> <z>
  • mirtk edit-image a.nii.gz b.nii.gz -origin <x> <y> <z>
• convert to -float: mirtk convert-image in.nii out.nii -float
• dilation: mirtk dilate-image
• erosion: mirtk erode-image
• subtract: mirtk calculate-element-wise a.nii -sub b.nii -out a_minus_b.nii
• padding
• labelStats
• evaluation: evaluation emasked-$thislevel.nii.gz $srctr -Tp 0 -mask emargin-$thislevel-dil.nii.gz -linear | grep NMI | cut -d ' ' -f 2

[schuhschuh]

scale.sh:

• seg_stats (NiftySeg) options (see here):
  • -t 1: Only estimate statistics if voxel is larger than 1
  • -R: The robust range (assuming 2% outliers on both sides) of all voxels
  • mirtk calculate-element-wise img.nii -threshold 1 -pct 2 98 -d ' '
• seg_maths (NiftySeg) operations: -thr, -bin, -mul, -uthr, -add

$maths_ms $input_ms -thr $max_ms -bin -mul 255 $peaks_ms

fac_ms=$(echo "scale=5; 255/$max_ms" | /usr/bin/bc)
echo $fac_ms

$maths_ms $input_ms -mul $fac_ms -uthr 255 -add $peaks_ms $output_ms
$maths_ms $output_ms -thr 257 -bin -mul 255 $peaks_ms
$maths_ms $output_ms -uthr 256 -add $peaks_ms $output_ms

$maths_ms $input_ms -uthr 0 -mul -1 -bin -mul -1 $tmp_ms
$maths_ms $output_ms -thr 0 -add $tmp_ms $output_ms

• cl_erosion (5 iterations):
  • Input is grey-scale image
  • Binarise by setting values >0 to one and zero otherwise
  • Erode this binary mask
  • Set all input grey-scale values outside of mask to zero, keep other values unchanged

[schuhschuh]

myProbabilisticAtlas adds SwapImages function, which is also in Draw-EM's implementation.

[schuhschuh]

To-Do list

1. Remove proot script.
2. Refactor PINCRAM Bash script:
   • Move common functions directly inside pincram script.
   • Use mirtk calculate-element-wise instead of seg_maths.
   • Any replacement for labelStats available? How about mirtk evaluate-overlap?
     • Usage: labelStats ref.nii.gz assess.nii.gz -q | cut -d ',' -f 1
     • -q: Quiet summary, CSV output: OR,Dice (mean values over all labels)
     • Add -mean option or similar to evaluate-overlap command.
3. Refactor C++ implementation of required custom commands:
   • cl_compute_volume --> mirtk malpem-compute-volume
   • cl_gaussian_fusion --> mirtk malpem-gaussian-fusion
   • cl_malpem --> mirtk malpem-refinement
   • cl_malpem4D --> mirtk malpem-longitudinal-refinement
4. Refactor MALPEM Python modules:
   • Execute MIRTK commands using mirtk.run, more conveniently as:
     • import mirtk
     • mirtk.run('<command>', args=[], opts={}) --> mirtk.<command>(*args, **kwargs).
   • Rename malpem.mytools to malpem.tools and update all references.
   • malpem.bias_correction
   • malpem.brain_extraction
   • malpem.intensity_normalise
     • Implement scale functionality using MIRTK tools directly in malpem.intensity_normalise.robust_rescale function.
   • malpem.label_fusion
   • malpem.label_refinement
   • malpem.registration
   • malpem.report
     • Use mirtk.rendering.screenshots module.
5. Update installer script.
6. Update README and LICENSE.
7. Include package in main MIRTK repository as optional extension package (Git submodule).