Workflow for manual labeling

[sandrinebedard]

Identifing problematic vertebral labeling (QC)

• Send QC to participants.
• Sort QC by subject.
• Define the subject range that each participant will help with.
• Sort by Function (reverse order) to see vertebral labeling.
• Identify problematic labeling following this procedure.
  • If C2-3 is not aligned with the disc, mark as ❌.
  • Mark as ⚠️ for bad data quality (subjects to exclude) : most likely FOV that cuts C2-C3.
  • Here is an example of an adequate labeling ✅:
    
• If some segmentation errors are noticed, mark the segmentation of that subject as ❌ (you can search for the subject to find its segmentation).
• Download the lists by clicking on Download QC Fails and on Download Qc Artifacts (at the very bottom of the QC page).
• Send both lists to the lead team.
• Lead team will package the data (automatic labeling + associated data) and send to the respective participant .

Manually correcting vertebral labeling

• Download SCT v5.3, see here for installation instructions.
• Identify C1-C2, C2-C3 and C3-C4 by following this procedure. (A tutorial is included)
• Package the derivatives folder as a ZIP package and send it to the lead team.
• Lead team will quickly review and add to the git-annex centralized repository and add the YML list of subjects to exclude to exclude.yml.