Normalization strategy

[sandrinebedard]

Description

The main goal is to create a database of healthy individuals, across age/sex of spinal cord morphometrics. The database would be in the PAM50 space, to do slice-wise comparisons with DCM subjects (and possibly also other pathologies).

Registration to PAM50

• Use all the disc labels (i.e., more than 2)
• Scale in S-I direction
• No rotation slice-wise (will biais torsion)
• Apply the transformation to the spinal cord segmentation to bring it in the PAM50 space
  • Note : warping segmentation in PAM50 space can results in partial segmentation at edge slices.
  • How do we assess the quality of the registration? discs alignement?

Possible issue:

• warping segmentation in PAM50 space resulting in partial segmentation at edge slices

Data included in the database

• .csv files from sct_process_segmentation
• No angle correction since we are in the straithened space
• rows: slices, column: metric (for different converage Dealing with different coverage)
• Include compression ratio and torsion (to be added to sct_process_segmentation)
• Include dice score comparision for triangular compression (see issue)
• Include demographics (sex, age, weight, height)

Dealing with different coverages:

• option 1:
  Build the CSV files so that they ALL have the same number of lines, and if info is missing (because low coverage), replace metric with ‘nan’.
• option 2:
  Have CSV files with variable number of lines (only have lines where segmentation exists)

Usage with DCM

• Resampling will be important if important anisotropy
• We need the discs to regsiter to PAM50 --> can be difficult, but can get it from other images
• Ensure that we don't lose sensitivity of the compression metrics in the straightened space

Datasets of healthy individuals to include:

• spine-generic
• Which contrast?

Steps

1. Select datasets to use of healthy individuals.
2. Create script to register to PAM50 template, run sct_process_segmentation to output .csv files (see section Registration to PAM50 for the specifics).
3. Validate the registration and results.
4. Apply the normalization on DCM patients and test the compression detection.

[sandrinebedard]

The updated version of the normalization strategy is now in the draft manuscript: https://docs.google.com/document/d/1zVYHBG-cepmZ7oDyr3iqArdQrbHNTgT7lUmWeRFj4Ng/edit#heading=h.ecl8nn3h89a8