ivadomed / canproco

Code for preprocessing the CanProCo brain and spinal cord dataset
MIT License
4 stars 1 forks source link

Discuss looking into tissue bridges in the case of spinal cord MS lesions #100

Open plbenveniste opened 4 months ago

plbenveniste commented 4 months ago

This is related to PR 4489.

A metric which could be interesting to compute in the case of MS lesions in the spinal cord is the size of the tissue bridges. It measures the distance/surface of the healthy spinal at the level of a lesion. It could be seen as the surface where neural information can still circulate.

The following image from Jan describes the measure of a tissue bridge:

image

valosekj commented 3 months ago

Trying on sub-cal119_ses-M0_STIR, which has 2 lesions:

input image ![Kapture 2024-07-08 at 13 30 49](https://github.com/ivadomed/canproco/assets/39456460/bcbf9286-8691-42d8-834e-72ad0af508cc)

Code:

# Pull the code for the tissue bridges
cd $SCT_DIR
git fetch
git checkout jv/compute_midsagittal_tissue_bridges

# Run sct_analyze_lesion
cd ~/data/data.neuro.polymtl.ca/canproco/derivatives/labels/sub-cal119/ses-M0/anat
sct_analyze_lesion -m sub-cal119_ses-M0_STIR_lesion-manual.nii.gz -s sub-cal119_ses-M0_STIR_seg-manual.nii.gz

Output:

...
Lesion count = 2

Measures on lesion #1...
  (S-I) length : 5.01 mm
  Max. equivalent diameter : 3.97 mm
  Maximum axial damage ratio : 0.17
  Midsagittal slice of the spinal cord: 4
  Sagittal slice 3, Minimum dorsal tissue bridge width: 2.73 mm (axial slice 195)
  Sagittal slice 3, Minimum ventral tissue bridge width: 0.0 mm (axial slice 194)
  Sagittal slice 3, Total tissue bridge width: 2.73 mm
  Volume : 44.1 mm^3

Measures on lesion #2...
  (S-I) length : 7.84 mm
  Max. equivalent diameter : 3.25 mm
  Maximum axial damage ratio : 0.13
  Midsagittal slice of the spinal cord: 4
  Sagittal slice 5, Minimum dorsal tissue bridge width: 0.0 mm (axial slice 221)
  Sagittal slice 5, Minimum ventral tissue bridge width: 2.75 mm (axial slice 220)
  Sagittal slice 5, Total tissue bridge width: 2.75 mm
  Volume : 55.86 mm^3
...

Output PNG image produced by sct_analyze_lesion:

image

One interesting thing to notice (which does not have an effect on the computation of the bridges, though) is that the lesion mask is not 100% within the cord mask. From the anatomical perspective, the lesion should be part of the cord. This should be taken into account when training models, for example, by ensuring that the lesions are part of the cord, as done in SCIseg here.