complete QC plan for ndmg paper

[jovo]

for each brain, we want:

(ingested) coregistered images of:

• (not co-registered raw
  • MPRAGE
  • DTI
  • fMRI
• "base data"
  • MPRAGE
  • DTI
  • fMRI
• derived data
  • tensor map
  • displacement field (if we use ndreg)
  • shrinkage field (if we ndreg)
  • fibers (ideally ingested)

multivariate statistics

• dataset_variance
• adjacency matrix
• eric figs (fmri) (@vikramc1 to populate)
  • Mean Figure: plot for each voxel's mean signal intensity for
    each voxel in the corrected mri's timecourse
  • STdev Figure: plot of the standard deviation of each voxel's
    corrected timecourse
  • SNR Figure: plot of signal to noise ratio for each voxel in the corrected timecourse
  • Slice-wise intensity Figure: averages intensity over slices and plots each slice as a line for corrected mri
  • motion parameters figures: 1 figure for rotational, 1 figure for translational motion params, 1 for displacement, calculated with fsl's mcflirt
• loss vs. iteration for registration (if we use ndreg)

scalar statistics

• everything from eric (@vikramc1 populate this)
  TODO: Maybe we should just calculate the distributions instead of statistics on distributions

  • Hellinger Distance
  • Average MSE after regsitration
  • fmri
    • Signal Statistics
      • Signal Mean
      • Signal Stdev
      • Number of Voxels
      • Average SNR per voxel
    • Motion statistics
      • Absolute Translational Statistics
        • Max absolute motion
        • Mean absolute motion
        • Number of absolute motions > 1mm
        • Number of absolute motions > 5mm
        • Mean absolute (x, y, z) motion
        • Std absolute (x, y, z) position
        • Max absolute (x, y, z) motion
      • Relative translational statistics
        • Max relative motion
        • Mean relative motion
        • Number of relative motions > 1mm
        • Number of relative motions > 5mm
        • Mean relative (x, y, z) motion
        • Std relative (x, y, z) motion
        • Max relative (x, y, z) motion
      • Absolute rotational statistics
        • Max absolute rotation
        • Mean absolute rotation
        • Mean absolute (x, y, z) rotation
        • Std absolute (x, y, z) rotation
        • Max absolute (x, y, z) rotation
      • Relative rotation statistics
        • Max relative rotation
        • Mean relative (x, y, z) rotation
        • Std relative (x, y, z) rotation
        • Max relative (x, y, z) rotation

• MI loss (if we do ndreg, or something else if we don't do ndreg)

• nnz from DTI

qualitative assessments (0/1 score for):

• Each qc image will be rated 0 (bad) or 1 (good) for each subject in all datasets in a csv file

  • Criteria for labelling an image as good as follows:
    • Registration (based on images):
      • Registered brain orientation matches atlas
      • No substatntial misalignment
      • No missing brain regions
    • Tensors (based on images):
      • Brain not cut off substantially
      • brain is registered in correct orientation
    • Fibers (based on images):
      • No fibers leave the brain
      • Fibers are in correct (expected) orientation for the most part
    • Adjacency matrix (based on adjacency matrix):
      • Has hemispheres
      • not all 1 color (no edges or only edges)
  • Automatic scalars will also be recorded in the same csv file for each subject in all datasets

• csv rows are subjects, columns are scalar values

• html page with all the multivariate plots

[jovo]

@gkiar

[jovo]

this is done, right @ebridge2

[ebridge2]

this is/has been done for a while for fMRI; I will be adapting dMRI to have similar QA for raw, preprocessing, and registration tomorrow.

[jovo]

@dmannan see this comment, and this:
https://github.com/neurodata/ndmg/blob/merge-dmri-fmri/analysis/statistics.md