Skip to content

Latest commit

 

History

History
30 lines (20 loc) · 2.47 KB

recommended_usage.rst

File metadata and controls

30 lines (20 loc) · 2.47 KB

Recommended Usage

Guidelines are provisional, as they are evolving, but the following are unlikely to change

General suggestions:

  • For the best results from outlier detection process, it is recommended to divide list of IDs into known groups (healthy, disease1, disease2, young, old etc) based on non-imaging parameters (such as clinical diagnosis, age etc), to perform the QC process independently on each group.
  • Be generous in the number of slices you use to review in each view (even if they appear small in the collage), as you have the ability to zoom-in anywhere you please for detailed inspection.
  • Routinely toggle overlays to ensure composite overlays are not affecting your perception of GM/WB boundaries in scans with unusual intensity distributions (low or high contrast, dark or too bright etc).

For Freesurfer outputs:

  • Inspect the quality of raw T1 MRI scans first, using visualqc, for presence of any artefacts, such as motion, ringing, ghosting, and anything else.
  • Install and run Freesufer, on ALL subjects in your dataset.
  • Follow the troubleshooting guide by the Freesurfer team, that includes atleast the following checks. These slides are a fantastic start to get an idea of what to focus on.
    • Review the accuracy of white and pial surfaces (this is the default), and identify subjects for further inspection (errors in the preceding steps of the pipeline)
    • Review the segmentation of white matter is accurate (overlay wm.mgz on T1.mgz) for each subject, and identify those to be rerun or to be corrected for minor errors.
    • Review the accuracy of skull-stripping for each subject, and identify the subjects that need to be rerun with special flags (for major errors), or corrected manually (for minor errors).

Alignment checks (Registration quality)

  • When comparing across modalities (e.g. EPI to T1, or PET to T1), big dissimilarities in intensity distributions might (if PET distribtuion too narrow, while T1w has a broad distribution) produce useless composites. In such cases, edge overlay or animation could work.
  • You could even attempt to rescale them beforehand as well, whichever is suitable for your application.