One should always correct EPI data for geometric distortion coming from
the b0 image. For some applications (e.g. looking at beta maps or
variance maps) it is important to correct for b1 intensity bias fields so
that the numbers mean the same thing across the whole brain. For
z-statistic images this doesnıt make any difference because they are a
ratio in which the bias field cancels out.
The latest version of the HCPıs minimal preprocessing pipelines allows
both b0 distortion correction and b1 intensity correction of fMRI data if
you have acquired phase reversed spin echo EPI images that match the fMRI
image geometries in the same imaging session.
Peace,
Matt.
On 5/24/16, 12:05 PM, "FSL - FMRIB's Software Library on behalf of Michael
Riedel" <[log in to unmask] on behalf of [log in to unmask]> wrote:
>Hello,
>
>I was hoping to get some expert opinions on correcting EPI datasets using
>estimated bias field maps from FAST (or any other segmentation software).
>The article "Effects of image contrast on functional MRI image
>registration," by Gonzalez-Castillo et al., 2013 examines the impact of
>correcting EPI datasets on motion correction algorithms. Surely acquired
>field maps would be better suited for correcting the EPI datasets, but in
>the event those weren't acquired, I saw a couple of previous posts on the
>forum:
>
>https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;d7dc8031.1605 and
>https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=fsl;4281da36.131
>
>that speak to a simple multiplication or division of the EPI dataset by
>the estimated bias field. First, is it a correct assumption that you
>would divide the EPI dataset by the estimated bias field, such that
>voxels exhibiting higher intensities would be reduced? Second, is this a
>common practice in data pre-processing? Related to this question, in the
>above mentioned article, bias field estimation is performed on the 6th
>EPI volume. Is it advantageous to take that approach, or use the bias
>field estimated by FAST and transform that into native functional space?
>Or should they essentially be identical?
>
>My last question is related to the impact of correcting the EPI dataset
>on beta-estimates after statistical modeling. Is there any reason to
>believe the distribution of beta's would change, i.e., increased
>accuracy, because of this technique?
>
>Thanks in advance for any insight,
>
>Michael Riedel
|