Hi Jake
In principle, you can compare any voxel-wise data across subjects
using randomise, as long as you can ensure that the data has the same
"meaning" across subjects, and that you can exchange subjects without
changing the covariance structure (which is the case here).
In the case of fdt_paths (from probtrackx), the data at each voxel is
the probability that the path of least hindrance to diffusion from
your seed passes through that voxel. So you have to ensure that you
are seeding from the same region across subjects in order for your
voxel-wise "quantity" to be comparable across subjects. (it can
sometimes be tricky to decide whether a region is the same across
subjects... need to consider both anatomy and function...).
In terms of the interpretation of your results, it worth noting that,
since you are comparing probabilities for the location of a tract, the
results should be understood as morphological rather than
microstructural. For the latter, it is best to use TBSS.
Cheers,
Saad.
On 1 Jun 2009, at 14:17, John Kuster wrote:
> Hello Steve,
>
> We used ROI's as seed regions in probtrack, and took the output for
> each
> person into MNI space by using transforms previously created (flirt
> and
> fnirt for each person's FA map to the FA template). Then, with the
> transformed probtrack data I used randomise to perform a contrast
> analysis
> between our patients and controls. Does this sound appropiate?
> Would there
> be another way to compare probtrack data between groups? If this is
> a valid
> method, we were wondering if you had any tips on how to relate the
> prob-tracking to the statistical output from randomise?
>
> Thank you!!
> Jake
>
> On 5/30/09 5:26 AM, "Steve Smith" <[log in to unmask]> wrote:
>
>> Not sure what you mean, in particular what processing you applied to
>> the probtrack outputs to get them into a common space.....?
>>
>>
>> On 29 May 2009, at 19:57, John Kuster wrote:
>>
>>> Hello Steve,
>>>
>>> Thanks for your explanation. Could you please explain how this
>>> applies to
>>> probtrack output? We have used randomise to compare probtrack
>>> outputs from a
>>> ROI in subjects and controls and are unsure of how to correct in
>>> this
>>> case...
>>>
>>> Thank you!!
>>>
>>>
>>> ---------------------------- Original Message
>>> ----------------------------
>>>> Subject: Re: [FSL] stastical corrections?
>>>> From: "Steve Smith" <[log in to unmask]>
>>>> Date: Tue, May 19, 2009 6:09 am
>>>> To: [log in to unmask]
>>>> --------------------------------------------------------------------------
>>>>
>>>> Hi,
>>>>
>>>> There are various practical reasons why TBSS samples the data into
>>>> 1x1x1mm standard space, primarily because having the mean-FA
>>>> skeleton
>>>> at this resolution works much better in practice than at lower
>>>> resolutions. We don't generally recommend doing a "VBM-style"
>>>> voxelwise analysis (i.e. using the first stage of the TBSS
>>>> processing
>>>> only) for the reasons outlined in the original TBSS paper -
>>>> however,
>>>> you're right, if we _did_, then working at 2x2x2mm would probably
>>>> be
>>>> just as good.
>>>>
>>>> You're right that upsampling the data complicates the multiple
>>>> comparison correction - but at any given resolution, and taking
>>>> into
>>>> account intrinsic data smoothness, it is important to get all these
>>>> issues taking into account. Hence both Gaussian random field
>>>> theory
>>>> inference explicitly corrects for effective smoothness for you, and
>>>> permutation-based inference, with FWE thresholding, implicitly also
>>>> takes all these things into account, to give valid multiple
>>>> comparison
>>>> correction. If you take either of these approaches in general when
>>>> done correctly you are ok.
>>>>
>>>> Cheers.
>>>>
>>>
>>>
>>> On 5/18/09 1:32 PM, "John Kuster" <[log in to unmask]> wrote:
>>>
>>>> I apologize if this has already been posted to the list, but I did
>>>> not see
>>>> it come through successfully. Thanks!
>>>>
>>>>
>>>>
>>>> Dear FSLers,
>>>>
>>>> We have a question regarding correction for multiple comparisons
>>>> in a
>>>> diffusion tensor imaging study:
>>>>
>>>> We have conducted a DTI study in which we have a priori areas of
>>>> interest
>>>> in our voxel-based analysis. We've pulled out the areas of interest
>>>> and
>>>> are only evaluating them initially, so the correction for multiple
>>>> comparisons should be based on the number of voxels considered,
>>>> with
>>>> smoothing/clustering criteria taken into account.
>>>>
>>>> So far, this is all fine. However, the problem arises when I
>>>> realize that
>>>> the registrations that FSL uses to put DTI data into standardized
>>>> space
>>>> (using an MNI template), takes 2x2x2 data and registers it into a
>>>> 1x1x1
>>>> volume. The voxel-based contrast is done on the registered 1x1x1
>>>> images,
>>>> and thus, eight times the number of comparisons are made than would
>>>> have
>>>> been made on the original data. So the question is: How do we
>>>> correct for
>>>> multiple comparisons? It doesn't make sense to me to "increase"
>>>> resolution and the number of comparisons without actually adding
>>>> real
>>>> information.
>>>>
>>>> Does anyone have any suggestions how to deal with this situation?
>>>>
>>>> Thank you!
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>
>>
>> ---------------------------------------------------------------------------
>> Stephen M. Smith, Professor of Biomedical Engineering
>> Associate Director, Oxford University FMRIB Centre
>>
>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>> +44 (0) 1865 222726 (fax 222717)
>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>> ---------------------------------------------------------------------------
>>
>>
>
--
Saad Jbabdi
Oxford University FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
+44 (0) 1865 222545 (fax 717)
www.fmrib.ox.ac.uk/~saad
|