Dear Mugundhu,
On 15 Dec 2020, at 17:52, Mugundhu R wrote:
> Dear Christian,
>
> Thank you for your explanation. I tried with 8mm, 6mm, and 4mm
> smoothness
> with the TFCE approach. The 8 mm smoothness produced a result with a
> huge
> cluster as I explained in the previous mail. I used the slice overly
> option
> to display the selected cluster, but still, it picked up many regions
> since
> the cluster was huge. Then I tried with 4mm smoothness, it produced
> the smaller cluster with less number of voxels. For 6 mm smoothness, I
> found a reasonably good result. I decided to use the result with 6 mm
> smoothness. I appreciate your suggestions if any.
The FWHM of 6mm are also the default and personally I only change that
value quite rarely if I expect very large or very small effects (i.e.
Matched Filter Theorem).
Best
Christian
>
> Thank you very much for this wonderful toolbox.
>
> Mugundhu
>
> On Mon, Dec 7, 2020 at 4:48 PM Christian Gaser
> <[log in to unmask]>
> wrote:
>
>> Dear Mugundhu,
>>
>> On 7 Dec 2020, at 4:21, Mugundhu R wrote:
>>
>>> Dear Christian,
>>>
>>> Thank you for your suggestion and paper link. It was quite useful.
>>> With 3 permutation values of 5000, 10000, & 20000, I did the
>>> analysis
>>> again. Similar findings were shown from all the analyses. Please
>>> view
>>> the
>>> screenshot attached.
>>
>> Because the TFCE toolbox is using tail approximation the expected
>> differences over 5000 permutations should be rather small as you also
>> found. With 20000 permutations your are simply on the safe side for
>> most
>> studies.
>>
>>> In the result, there is a massive cluster. Should I
>>> just report the peak coordinates that I found in the cluster or all
>>> of
>>> the
>>> regions that were found in the cluster? I attempted 8 mm smoothness
>>> and
>>> then 4 mm smoothness, but getting similar results with changes in
>>> the
>>> numbers of voxels. Support me on how to report this, please.
>> If you use the slice overlay tool in CAT12 there is an option to
>> print
>> atlas information for the resulting clusters. This indicates the
>> percentage of which a cluster consists of several anatomical regions
>> defined in the atlases. This is also the preferred method to report
>> results.
>>
>> Best,
>>
>> Christian
>>>
>>> Regards
>>> Mugundhu
>>>
>>> On Fri, Dec 4, 2020 at 9:25 PM Christian Gaser
>>> <[log in to unmask]>
>>> wrote:
>>>
>>>> Dear Mugundhu,
>>>>
>>>> On Tue, 1 Dec 2020 13:33:16 +0530, Mugundhu RB
>>>> <[log in to unmask]>
>>>> wrote:
>>>>
>>>>> Dear Experts,
>>>>> I am analyzing my structural MRI data. I have 3 groups, A (n=28),
>>>>> B
>>>>> (n=28,
>>>>> C (n=27) and 4 co-variates. I used the Full Factorial Design model
>>>>> in
>>>> CAT12
>>>>> for statistical analysis. It has generated F contrast
>>>>> automatically.
>>>>> From
>>>>> the F test (main effect of ANOVA) result, I have selected 8
>>>>> regions
>>>>> to
>>>>> create a single union mask. I used this mask for SVC in TFCE
>>>>> analysis.
>>>> This might be a typical case of double-dipping:
>>>> https://dx.doi.org/10.1038%2Fnn.2303
>>>>
>>>> Small volume correction is thought to limit your statistics to
>>>> those
>>>> regions where you have a clear prior anatomical hypothesis. It is
>>>> not
>>>> appropriate to select the SVC area based on a statistical test in
>>>> the
>>>> same
>>>> design.
>>>>
>>>>> For group comparison, example;
>>>>> For A>B I used the contrast 1 -1 0 0 0 0 0
>>>>> For B>C I used the contrast 0 1 -1 0 0 0 0
>>>>> I kept TFCE permutation as 5000
>>>>>
>>>>> I have a few questions.
>>>>> For group comparison, the contrasts which I mentioned here are
>>>>> correct?
>>>> If your thirst three columns represent the three groups, yes.
>>>> Please
>>>> also
>>>> check the CAT12 manual for defining contrasts.
>>>>
>>>>> TFCE with 5000 permutations is sufficient for 3 group comparison
>>>>> or
>>>>> should
>>>>> I increase the number of permutation?
>>>> 5000 permutations are a good starting point to get some first
>>>> results.
>>>> This is also mentioned in the help text (In order to obtain
>>>> reliable
>>>> estimates you need about 5000-10000 permutations.)
>>>> For your final analysis, a higher number of 20000 iterations might
>>>> be
>>>> better to be on the safe side.
>>>>
>>>>> Should I do any additional statistics corrections?
>>>> If you finished the permutations there are also maps saved for FDR-
>>>> and
>>>> FWE-correction. Please check the online help of the TFCE toolbox
>>>> for
>>>> the
>>>> naming convention of the result files.
>>>>
>>>> Best,
>>>>
>>>> Christian
>>>>
>>>>>
>>>>> Please advise me.
>>>>> Thank you
>>>>> Mugundhu
>>>>>
>>>>
>>>>
>>>>
>>
>>
>>
|