3dwarp -deoblique and dwi data need to correct bvals and bvecs?

AFNI Message Board
1

[b]Hello,

I am running am trying to run my dwi data through TRACULA and for two of my subjects I was unable to run the command orientLAS.
After using 3dinfo on the data set I found that the two subjects that didn’t run had a Data Axes Tilt: Oblique (45.000 deg. from plumb)[/b]

My data set

Data Axes Tilt: Oblique (43.300 deg. from plumb)
Data Axes Tilt: Oblique (38.100 deg. from plumb)
Data Axes Tilt: Oblique (42.100 deg. from plumb)
Data Axes Tilt: Oblique (32.400 deg. from plumb)
Data Axes Tilt: Oblique (41.700 deg. from plumb)
Data Axes Tilt: Oblique (42.300 deg. from plumb)
Data Axes Tilt: Oblique (45.000 deg. from plumb) (didn’t work)
Data Axes Tilt: Oblique (37.900 deg. from plumb)
Data Axes Tilt: Oblique (34.300 deg. from plumb)
Data Axes Tilt: Oblique (31.700 deg. from plumb)
Data Axes Tilt: Oblique (31.800 deg. from plumb)
Data Axes Tilt: Oblique (27.500 deg. from plumb)
Data Axes Tilt: Oblique (38.900 deg. from plumb)
Data Axes Tilt: Oblique (45.000 deg. from plumb) (didn’t work)
Data Axes Tilt: Plumb
Data Axes Tilt: Oblique (13.200 deg. from plumb)

When I applied 3wdarp -deoblique to the dwi images and then ran them through TRACULA and orientLAS for fsl it worked, however, now I am worried if this will cause an issue with eddy current correction. Do I need to modify the bvecs and bvals for the subjects where I use -deoblique on them before eddy current correction? Sorry, I am not super comfortable with how afni treats obliquity.

2

Hi-

That is a pretty extreme obliquity!

What is obliquity? There are coordinates defined in scanner: a grid where the z-axis runs along the bore, and then x- and y- planes are perpendicular to that. However, sometimes a person’s head isn’t aligned conveniently to that, and fitting a “shoe box” FOV would be more convenient if slices were made on a different grid-- that is, one defines a new grid whose x-y plane differs from that of the scanner. These new coordinates are called “oblique coordinates”.

In oblique coordinates, the brain slices are probably roughly aligned with the x-y plane (where, in the scanner coordinates, the standard brain planes are probably at an angle to the x-y coordinates).

For visualization purposes, the oblique coordinates are convenient. However, for relating to other data sets (which might not have been acquired obliquely) or likely with the DWI gradients (which I believe would be acquired in terms of scanner coordinates), using the scanner coordinates makes most sense.

How do we convert between the oblique coordinates and scanner coordinates? We use matrix transforms (yay!). The “oblique” DWI dataset that you have has information in its own header to transform itself into the scanner coordinates. If you run


3dWarp -deoblique -prefix DSET_SCANNER_COOR DSET

then DSET_SCANNER_COOR will show your dataset on a new grid, whose x-, y- and z-values match with the scanner. It will probably appear to be at a fairly steep angle to standard view planes (based on what appears to be your dset’s amount of oblique angulation). But there you have it.

Note that the price you pay for acquiring your data on oblique coordinates is that, when you return your data to the scanner coordinates, you re-grid your data, which is a process that inherently involves smoothing, because the dset values have to be interpolated.

Re. eddy current correction: how are you performing that? With AP-PA (blip up/down) correction, for example using DIFFPREP in TORTOISE (which woudl likely involve using the anatomical volume as a reference)? Or are you applying a field map, and if so, was this acquired on the same oblique coordinates?

The phase encode axis of a dset is defined along the acquired slices, so therefore in oblique coords.

–pt

3

Thank you for the thorough explanation! I am actually using a dataset where I do not have a field map or AP & PA acquisition so I am simply running the eddy_correct command in fsl.

4

Hi AFNI experts,

Thanks for your thorough explanation of obliquity. I am further looking for the answer to bvec correction.

This is my scenario:
I have DWI (AP-PA), T1, T2 data from macaque, all of which are oblique and 90 degrees rotated from stereotaxic orientation.
After running dcm2niix on T1/T2, I have deobliqued and rotated the anatomical data by 90 degrees, and axialized to NMTv2. All seem fine.
If I do the same to DWI (i.e. deoblique and rotate), then I should rotate the corresponding bvec using the same transformations, right?
I am following FATCAT DEMO2, so shortly the data will go into TORTOISE3 for DIFFPREP and DR_BUDDI.

Thanks for any hints.
Best,
Tahereh

5

For any other user landing here after a search, I found the following very old answers in the forum, and worked out my way from here:

https://afni.nimh.nih.gov/afni/community/board/read.php?1,52909,54477#msg-54477
https://afni.nimh.nih.gov/afni/community/board/read.php?1,52909,54478#msg-54478