Manual correction strategy for adding Whole-Spine

[sandrinebedard]

Description

This issue presents the workflow and method to manually correct the whole-spine T2w spinal cord segmentation (cross-ref with problematic GT in the dataset issue)

Running inference

👉 Link to code

1. Install contrast-agnostic M5' model:

sct_deepseg -install seg_sc_contrast_agnostic -custom-url https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/releases/download/v3/model_contrast_agnostic_20250115-2211.zip

2. Launch processing (Example command):

sct_run_batch -path-data ~/datasets/whole-spine/ -path-out ~/process_results/whole-spine-norm/processed_whole-spine-norm_2025-01-21_t1t2_v2.1 -script scripts/process_data_whole-spine.sh -jobs 15

Manual correction

1. Install manual_correction package

Example command:

python manual_correction.py -config /Users/sandrinebedard/processed_data/whole-spine-seg/qc_fail_whole-spine_t2w_Sandrine.yml -path-img /Users/sandrinebedard/processed_data/whole-spine-seg/data2corrrect_Sandrine

What to look for?

1. Top of the cord

Erase or correct (if the slice is above C1 vertebrae, okay to erase):

2. Go through all axial slices and correct obvious over/under segmentation

3. Lumbar enlargement

Correct over segmentation of the rootlets:

4. Conus medullaris

4.1. Find the tip of the conus medullaris on the sagittal and coronal slice

4.2. Validate that it is the mid-sagittal slice on the axial slice

4.3. Erase what is below this point

4.4. Correct for over segmentation of the tip of the conus medullaris to ensure cone shape across slices

Sending the data

You will see under data2correct_<your name> a derivatives folder with a qc folder, send this to Sandrine

[sandrinebedard]

@valosekj & @SomeoneInParticular Here are the instrcuctions, you can find the qc_fail file and data2_correct and QC report here:
~/duke/projects/dcm-normalization/normalization-2025/processed_whole-spine-norm_2025-01-21_t1t2_5prime-v2/data2correct_<your_name>.zip

~/duke/projects/dcm-normalization/normalization-2025/processed_whole-spine-norm_2025-01-21_t1t2_5prime-v2/qc_fial_whole-spine_t2w_<your_name>_<your_name>.yml

[sandrinebedard]

Tip of the conus medullaris:

I would end the segmentation at the tip of the conus medullaris, at the limit of where the fillium terminale starts, as shown here:

Some links

Refs:

• https://radiopaedia.org/articles/conus-medullaris
• https://www.researchgate.net/publication/286625306_Prevention_of_diagnostic_errors_in_position_of_conus_medullaris_in_adult_patients/figures?lo=1)