Yes, it is ok to drop a handful of volumes. Dropping 2-3 volumes out of
your 210 total is not going to give you any bias. It is much better, in
terms of SNR, to drop just 2-3 volumes than to exclude an entire run
from the analysis (in which case you'll have differing SNR across
subjects). And replacing the problematic volume with the average of the
other two runs is entirely artificial.
cheers,
-Mike H.
On Fri, 2010-01-29 at 16:47 +0100, Alina Jurcoane wrote:
> Thanks everybody for the replies.
>
> To Mike, do you really think it's ok to simply discard some gradient
> directions?
> My gradients are not ordered to improve the spherical coverage of partial
> scans as in the paper below. I fear that dropping some volumes (gradients)
> would give me a bias towards some specific directions. Am I wrong? And even
> if I had grads as in that paper, I think their scheme is meant for
> prematurely interrupted scans rather than for some volumes discarded at
> "random".
>
> Cook PA, Symms M, Boulby PA, Alexander DC. Optimal acquisition orders of
> diffusion-weighted MRI measurements. J Magn Reson Imaging. 2007
> May;25(5):1051-8.
>
> Thanks,
> Alina
>
>
> --------------------------------------------------
> From: "Michael Harms" <[log in to unmask]>
> Sent: Friday, January 29, 2010 3:08 PM
> To: <[log in to unmask]>
> Subject: Re: [FSL] artifacts and dti
>
> I would just discard the problematic grad directions, so they do not
> contribute to the DTI fitting. No reason not to use the rest of the
> directions from that run, provided those data are truly ok.
>
> cheers,
> -Mike H.
>
> On Fri, 2010-01-29 at 13:38 +0100, Alina Jurcoane wrote:
> > Hi,
> >
> > I have three identical dti runs per subject, each with 70 timepoints (10b0
> > and 60 grad directions) resulting in 210 timepoints per subject.
> >
> > Suppose that in one run I have 2-3 timepoints in which some slices are
> > affected by artifacts (spikes, noise - see attached example). What is the
> > best way to proceed?
> > 1. Discard that run ending up with some subjects having 2 runs and some
> > subjects having 3 runs.
> > 2. Replace the artifact timepoint with the average of the same timepoint
> > from the other two runs.
> > 3. Discard the whole subject data alltogether.
> >
> > Thanks,
> > Alina
|