Spinal levels

In-vivo analysis of spinal levels from MRI images.

Measure the distance between spinal level midpoints along the spinal cord:

• Cervical: C2–C8, based on the spinal rootlets segmentation
• Lumbar: L1–S1, based on manually labeled spinal levels

All scripts require SCT (they use the SCT Python environment).

Scripts

Script	What it does
01_run_batch_rootlets_spinal_levels.sh	Cervical, per subject (run with sct_run_batch): spinal cord segmentation, PMJ detection, rootlets segmentation, and spinal levels with their distances from the PMJ. Manual labels in derivatives/labels are used when available.
02_compute_cervical_midpoints_distance.py	Cervical: aggregates the results across subjects, computes the C2–C8 midpoints distance and plots it.
config_01_run_batch_rootlets_spinal_levels.json	sct_run_batch config listing the spine-generic subjects with manual rootlets and spinal cord segmentations.

Spinal levels for the cervical analysis are computed by 02a_rootlets_to_spinal_levels.py, called from the batch script.

Cervical: C2–C8

Spinal levels are obtained from the intersection of the rootlets and the spinal cord segmentation. The midpoint of each level is the average of the distances of its start and end from the pontomedullary junction (PMJ). The C2–C8 distance is then distance_from_pmj_midpoint of C8 minus distance_from_pmj_midpoint of C2.

1. Download the data (the images are stored with git-annex):

cd ~/data/data.neuro.polymtl.ca/data-multi-subject
git annex get sub-*/anat/*_T2w.nii.gz derivatives/labels/sub-*/anat/*_T2w_label-*

2. Run the per-subject processing (edit path_data and path_output in the config if needed):

sct_run_batch -config config_01_run_batch_rootlets_spinal_levels.json

3. Aggregate the results:

python 02_compute_cervical_midpoints_distance.py -i <path_output>/data_processed

Outputs per subject (in data_processed/<subject>/anat/):

• *_label-rootlets_dseg_spinal_levels.nii.gz: spinal levels projected on the spinal cord
• *_label-rootlets_dseg_pmj_distance.csv: per-level start, end and midpoint distances from the PMJ

Outputs across subjects (in the folder passed to -i):

• spinal_levels_spine-generic.csv: per-level distances for all subjects
• spinal_levels_midpoints_distance_2-8_spine-generic.csv: C2–C8 midpoints distance (mm) per subject
• figure_spinal_levels_midpoints_distance_2-8.png: C2–C8 midpoints distance per subject

Lumbar Balgrist dataset

Scripts for the lumbar Balgrist dataset are in lumbar-balgrist/. Input: manually labeled spinal cord segmentations (sub-*/*_spinallevels_dseg.nii.gz) with one value per spinal level: 1=L1, 2=L2, 3=L3, 4=L4, 5=L5, 6=S1, 7=S2.

Script	What it does
lumbar-balgrist/lumbar_length_count_slices.py	Computes the lumbar spinal cord length by counting axial slices (-metric top-bottom: top of L1 to bottom of L5, default; -metric midpoints: L1–S1 midpoints distance) and plots it per subject.
lumbar-balgrist/lumbar_length_along_centerline.py	Same lengths, but measured along the spinal cord centerline (SCT get_centerline, ParamCenterline: bspline, smooth=30, fitted to the L1–S2 segmentation) to account for the cord curvature; also saves the fitted centerlines for QC.
lumbar-balgrist/correlation_length_height.py	Plots the correlation between the length and subject height.

python lumbar-balgrist/lumbar_length_count_slices.py -i <folder with sub-*/*_spinallevels_dseg.nii.gz> -o <output folder>
python lumbar-balgrist/correlation_length_height.py -i <output folder>/lumbar_length_L1-L5.csv

Outputs (in the folder passed to -o, default ~/data/spinal-levels/lumbar_rootlets_Balgrist/results_and_figures): lumbar_<metric>.csv (length per subject, and the min slice, max slice and midpoint slice of each spinal level) and figure_lumbar_<metric>.png.

Lumbar EPFL dataset

Scripts for the lumbar EPFL dataset (~/data/spinal-levels/lumbar_mri_Sergio/spinalcord_masks) are in lumbar-epfl/. Input: spinal cord masks with the personalized spinal levels saved one level per file (sub-*/ses-*/anat/*_label-SC_desc-<level>_mask.nii.gz, desc-L1 ... desc-S2), and participants.tsv (sex and height) one level above the masks folder.

Script	What it does
lumbar-epfl/lumbar_length_count_slices.py	Computes the lumbar spinal cord length (-metric top-bottom: top of L1 to bottom of L5, default; -metric midpoints: L1–S1 midpoints distance), plots it per subject and its correlation with height.
lumbar-epfl/lumbar_length_along_centerline.py	Same lengths, but measured along the spinal cord centerline (SCT get_centerline, ParamCenterline: bspline, smooth=30) to account for the cord curvature; also saves the fitted centerlines for QC.

python lumbar-epfl/lumbar_length_count_slices.py -i ~/data/spinal-levels/lumbar_mri_Sergio/spinalcord_masks

Outputs (in the current working directory): lumbar_<metric>_masks.csv, figure_lumbar_<metric>_masks.png and figure_correlation_lumbar_<metric>_masks_height.png.

Lumbar Balgrist + EPFL combined

lumbar-combined/combine_balgrist_epfl.py combines the L1–L5 length along the centerline of both datasets (CSV files from the two lumbar_length_along_centerline.py scripts) into a per-subject figure (stratified by sex, sorted by height) and a correlation with height (all subjects pooled). Sex is shown by color and the dataset by marker.

python lumbar-combined/combine_balgrist_epfl.py

Outputs (in the folder passed to -o, default ~/data/spinal-levels/results_and_figures_balgrist_epfl): lumbar_length_L1-L5_balgrist_epfl.csv, figure_lumbar_length_L1-L5_balgrist_epfl.png and figure_correlation_lumbar_length_L1-L5_balgrist_epfl_height.png.