I have a data set (diff:1 b0 and 64 b1000, T1 and GRE filedmaps) containing 2 groups (OLD and YOUNG subjects) and I would like to do tractography between my ROIs (for example M1-M1 through CC) which I have in MNI space and correlate the tractography outcome with behavioral measures.
So far I have performed motion and eddy correction using mrtrix commands and epi distortion correction for all subjects.
I was wondering if there is any general pipeline that I can follow after performing these steps.
More specifically:
Do I need to perform dwi biasfield correction and intensity normalization? (Happily I could make ants package work, just want to make sure before applying to data set.)
At what stage should I register to T1?
I can provide (at least for the Young group) the T1 freesurferaparc+aseg.nii.gz files and BET T1 and b0 if they are necessary.
correlate the tractography outcome with behavioral measures
What kind of ‘outcomes’ are you looking to quantify? If these have anything to do with streamline count, you should be applying some flavour of SIFT, in which case bias field correction is very much recommended. It’s probably preferable for tracking anyway, so that the effect of the FOD cutoff threshold is more consistent throughout the field of view.
intensity normalization?
This depends on precisely how you intend on normalising ‘connection density’ between subjects. If you’ll be doing as most do and just use the same number of streamlines in each subject, then image intensity normalisation will have no effect. If you’re looking to do something more clever… well, I haven’t yet published how I think it should be done… so you’re on your own
(Happily I could make ants package work, just want to make sure before applying to data set.)
There’s also the option of using FSL to estimate the bias field in the dwibiascorrect script. Though we use ANTs ourselves, and would tend toward recommending that option based on our own experiences.
At what stage should I register to T1?
My personal preference is:
Register T1 → DWI, rather than DWI → T1. That way you don’t have to worry about gradient reorientation.
Perform the registration after DWI pre-processing, but before any T1 processing (that includes importing the subject into FreeSurfer). That way any processed T1 images are inherently aligned with the original T1 (and hence the DWIs), and you don’t have to worry about which images have been registered and which have not.
In fact I should have mentioned the parameters of interest such as Ave FA along the tracts.
So SIFT and intensity normalization are not necessary?
Indeed performing T1 -> DWI is very clever (was done in SPM12, coreg/estimate).
In fact I should have mentioned the parameters of interest such as Ave FA along the tracts.
So SIFT and intensity normalization are not necessary?
SIFT would still have some influence in this scenario, but not a huge amount. And I think it would be difficult to justify the computational expense and complexity, modelling crossing fibres in such a way only to reduce down to a mean FA. Intensity normalisation is still preferable for the sake of tracking as it influences the FOD amplitude threshold, though if your bias field is minimal then the effect on tracking will be minimal; the effect of a bias field on FA calculation should be minimal also, since it’s fitting an exponential decay rather than dealing with absolute values.
