 |
 |
How are you dealing with the fact that you are introducing both linear and nonlinear distortions of the data to the template in relation to the bvecs that have nothing to do with the EPI distortions? Because the bvecs are in undistorted space, you are actually correcting the data relative to the bvecs when you do a standard (i.e. field map, phase up/phase down) or FNIRT based distortion correction. I would think you would only want to use your method on the data after you have calculated the diffusion parameters, though perhaps it would not matter if you don’t care about the orientations of the tensors…
Peace,
Matt.
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Siddharth Srivastava
Sent: Monday, April 19, 2010 3:58 PMSubject: Re: [FSL] AW: [FSL] EPI distortion of DTI
I adopt a similar approach, but instead of spatially normalizing directly to the T1, i have done one
of the following on various instances:
1) get FA for the original dataset after usual pre-processing steps, affine normalize+ warp the FA
to the FA template in MNI, and than warp the entire 4D data used to calculate this FA, to the MNI,
then recalculate the tensors from the warped data.
2) Directly warp the DTI data to the MNI space (affine + warps) using the EPI or T2 template in the
MNI space as the reference image, and a contrast matched image from the 4D set as the floating
image.
Both these approaches tend to start with (floating, reference) pair with similar contrast prior to
spatial normalization.
In case the purpose is to do VBM both of these techniques work well, and are able to correct for
most of the distortions. I have not observed any major shifts in the usual magnitudes of FA, RD etc, and
the V1*FA maps look quite all right. The disadvantage is that in case one needs to go to a different reference
space, a part of the calculations have to repeated.
sid.
On Mon, Apr 19, 2010 at 1:24 PM, Jesper Andersson <[log in to unmask]> wrote:
Hi Matt (and Andreas),
I have used this on a few datasets, and was satisfied that the correction
produced fewer problems than the original distortion.
I think the phrase "fewer problems than the original distortion" is important here.
When fnirt was developed I didn't think it would be possible to use for this purpose, but when people reported that they had tried it with some success I did some tests too. As Matt says, it produces results that at least visually are better than the often quite horrific distortions one sees in diffusion images. However, the contrast is quite different in FA and T1 images, and in particular the high intensity "tracts" in the FA images are typically limited to the centres of the white matter areas. This means that one can often see that fnirt attempts to "widen" these tracts, which I think might be quite bad in many applications.
It would be _much_ better to perform a proper distortion correction based either on a fieldmap or on blip-up-blip-down data (for which software is now being tested in house), and therefore I would like to discourage people from thinking that they can skip collecting those extra data and simply use fnirt instead.
Jesper