Hi Reza,
Could you please advise how can I perform cluster-wise inferences
using FEAT? I checked the FEAT GUI and it contains both first level
and second level analysis. However, I just want to perform a
two-sample unpaired t-test with cluster-wise inference on my data. Is
there a way to bypass the first level analysis and can I use text-only
commends to perform the analysis?
Best,
Leo
On Tue, Aug 17, 2010 at 7:22 AM, Miguel Burgaleta
<[log in to unmask]> wrote:
> OK Reza, thanks!
>
> On Tue, Aug 17, 2010 at 10:04 AM, Reza Salimi <[log in to unmask]> wrote:
>>
>> Generally speaking, you have two tools for performing cluster-wise
>> inferences in FSL:
>> 1) FEAT for a parametric cluster inference (using random field theory)
>> 2) randomise for a nonparametric cluster inference
>> the latter is the one my former reply corresponds to, as TBSS analysis is
>> carried out using randomise.
>> If by "run randomise on the clustered image" you mean, when running
>> randomise the statistic of interest is cluster size and since you wanna
>> control for FWE you JUST take max_cluster_size, YES :)
>> Cheers
>> On Tue, Aug 17, 2010 at 2:52 PM, Miguel Burgaleta
>> <[log in to unmask]> wrote:
>>>
>>> Thanks Reza.
>>> So, in FSL lingo, I would run cluster on my t-map and then run randomise
>>> on the clustered image?
>>> Miguel
>>>
>>>
>>> On Tue, Aug 17, 2010 at 2:54 AM, Reza Salimi <[log in to unmask]> wrote:
>>>>
>>>> Hi Miguel,
>>>>
>>>> On Tue, Aug 17, 2010 at 4:04 AM, Miguel Burgaleta
>>>> <[log in to unmask]> wrote:
>>>>>
>>>>> Hi FSLers,
>>>>> I just noticed that, in the TBSS examples shown in the TBSS paper
>>>>> (Smith et al., 2006, NeuroImage), statistical analyses are conducted by
>>>>> "permutation-based inference on cluster size". Does this mean that the tmap
>>>>> is thresholded and converted into a mask that is then used to do the
>>>>> multiple comparisons correction?
>>>>
>>>> it is a standard non-parametric cluster-based inference,
>>>> i.e., you convert your t-stat image to a cluster image after a
>>>> thresholding, then use a number of permutation to see how the cluster sizes
>>>> are under the null. In each permutation you sample from the max_cluster_size
>>>> and form the null distribution of cluster size, which (as you used
>>>> max_cluster across the whole volume) is corrected for multiple comparison
>>>>
>>>>>
>>>>> If yes, is TFCE the alternative method to that arbitrary thresholding?
>>>>
>>>> yes
>>>>
>>>>>
>>>>> Thanks a lot for your clarification on this.
>>>>> Miguel
>>>>
>>>>
>>>> --
>>>> Reza Salimi-Khorshidi,
>>>> DPhil Candidate, FMRIB Centre of the University of Oxford (Linacre
>>>> College)
>>>> Associate Member, Oxford-Man Institute for Quantitative Finance,
>>>> University of Oxford
>>>>
>>>> Email: [log in to unmask]; Tel: +44 (0) 1865 222704; Fax: +44 (0)1865
>>>> 222717
>>>> Address: FMRIB Centre, John Radcliffe Hospital, Oxford OX3 9DU, UK
>>>>
>>>>
>>>
>>
>>
>>
>> --
>> Reza Salimi-Khorshidi,
>> DPhil Candidate, FMRIB Centre of the University of Oxford (Linacre
>> College)
>> Associate Member, Oxford-Man Institute for Quantitative Finance,
>> University of Oxford
>>
>> Email: [log in to unmask]; Tel: +44 (0) 1865 222704; Fax: +44 (0)1865
>> 222717
>> Address: FMRIB Centre, John Radcliffe Hospital, Oxford OX3 9DU, UK
>>
>>
>
>
|