Hi,
There isn't really any way to "account" for differences in data quality and we
strongly recommend that the same sequences be used for any group
analysis (structural, fmri or diffusion). If each dataset corresponds to a
different group then there is no easy way to separate the effect of group
and the effect of dataset.
So at this point I think your options are:
1 - check *very, very* carefully all the segmentations and registrations to see if
you can identify any problems which might be driving the "off" results. If
there are then see if you can fix them with different processing options and
try the analysis again.
2 - alternatively, a rather nasty thing to try and reduce the effects of the different resolutions
(although it wouldn't account for any sequence changes and only partially for SNR
changes) would be to resample all of your images into a common "compromise"
resolution to begin with - say 1.3 by 1.3 by 1.8. This obvious degrades the
resolution by a reasonable amount, but has the effect of inducing a fair amount
of interpolation-related effects in both datasets which *might* be enough to make
them comparable. This isn't really a very good solution at all and I would really
only recommend it if the data was *extremely* valuable and could not be acquired
again (e.g. very rare patients or very old longitudinal data). There are considerable
methodological problems still associated with this idea (e.g. there is no guarantee
that the interpolation-related effects won't be a strong source of dataset bias still)
and therefore the interpretation would need to be very careful and you would have
difficulty convincing most reviewers that it was reasonable. So I would only do this
to get a rough feeling for what is in this data or as a last resort for extremely valuable
data.
Sorry I do not have better news, but this kind of acquisition really is a huge problem.
All the best,
Mark
On 21 Oct 2011, at 23:49, Aditya Kumar Kasinadhuni wrote:
> Hi all,
>
> I was trying to run a VBM analysis but apparently my structural datasets for one of the groups was acquired at an anisotropic resolution whereas the other was isotropic.
>
> Dataset 1 = 0.47x0.47x1.3
> Dataset 2 = 1 x 1 x 1
>
> My end results were completely off and I was wondering if this had anything to do with my completely erroneous results. I was not sure if FSL accounted for this in its algorithms while computing the differences. I know it registers the data and accounts for the intensities accordingly, but is this is a totally different case to begin with?
>
> Looking forward for your help.
>
> Thank you.
>
|