Neonatal parcellation

Hi. I need to register an atlas with relative parcellation to a 3dT1 neonatal brain image in order to then extract DKI and DTI metrics over WM tracts. Registration using flirt and fnirt should be ok. My question now is which is the best way to calculate metrics over ROIs, that is extracting DKI and DTI maps just in the 122 tracts included in this JHU neonatal atlas.
Thanks Rosella

Hi @ros,

The generally accepted best approach (which we should probably document somewhere; more on that soon) is:

  1. Register subject image to template;

  2. Invert transformation;

  3. Apply inverted transformation (with nearest-neighbour interpolation if an index image) to obtain atlas parcellation image(s) in individual subject space;

  4. Compute mean values of quantitative metric images within parcels.

    If your atlas provides a single parcellation image consisting of integer labels, you can produce a binary mask image of any particular parcel of interest using the -eq operator in mrcalc.

    The mask corresponding to the voxels within which you wish to calculate the mean is then provided to the -mask option of mrstats, so that calculation of statistics of the input image intensities are constrained to only those voxels. You can even use “-output mean” in your mrstats call to instruct it to only output the mean, which simplifies the collation of data across many subjects.


1 Like