Hi Mark,
Thank you so much, this is very helpful!
Best,
Liz
On Oct 2, 2013, at 3:03 AM, Mark Jenkinson wrote:
> Dear Liz,
>
> When you look at differences you will be sensitive to changes caused by values that are either both positive or both negative or mixed, so you are correct that you need to use contrast masking in order to separate out these results.
>
> The first level contrast does not restrict what is sent up to the higher level (no thresholding occurs to the data that is passed up) then it is easier to only use the [1] contrast, since otherwise it is easy to get confused by sign flips between the levels. So, based on using the [1] contrasts at the first level, it is easiest to use the following contrasts:
>
> c1: group1_negmean [ -1 0 ]
> c2: group2 negmean [ 0 -1 ]
> c3: group1 > group2 [ 1 -1 ]
> c4: group2 > group1 [-1 1 ]
>
> and then mask:
> a) contrast c3 by c1and c2 (with the Z>0 button selected)
> b) contrast c3 by c2 (with Z>0 button selected)
> c) contrast c4 by c1and c2 (with the Z>0 button selected)
> d) contrast c4 by c1 (with Z>0 button selected)
>
> These will then restrict the significant group differences to only those caused by either both groups having deactivations (a and c) or only the smaller group having deactivations (b and d). The usefulness of the latter is that one group may have near zero, and hence noisy, activations and so being stringent about then being one sign can throw away useful results just due to random noise. It really depends on your question of interest whether you want to be sure that both groups are "deactivating" or not.
>
> Also, you do not have to use the (Z>0) button, but if you don't then you are further restricting yourself to areas where you had enough statistical power to also know that the "deactivation" with respect to baseline was significant on its own. It is possible that you can get a significant difference result without this, and so I would tend to use the (Z>0) button, but if you want to be more stringent then that is fine too.
>
> I hope this makes things clearer for you.
> All the best,
> Mark
>
>
>
> On 1 Oct 2013, at 20:59, Elizabeth Woytowicz <[log in to unmask]> wrote:
>
>> Hi FSLers,
>>
>> I am trying to analyze functional data of 2 tasks between 2 different patient groups and am particularly interested in comparing the areas of deactivation. I think because these are negative values, I need to use contrast masking, but I’m not sure if this is correct or what I should be masking.
>>
>> So far, I had the analysis set up in FEAT as follows:
>>
>> At the first level, each task was run separately so each scan only had one EV. I set up a [1] and [-1] contrast to find the activation and deactivation, respectively.
>>
>> For the higher-level analysis, I set up the following contrasts of the same task between the 2 different groups:
>> c1: group1_mean [ 1 0 ]
>> c2: group2 mean [ 0 1 ]
>> c3: group1 > group2 [ 1 -1 ]
>> c4: group2 > group1 [-1 1 ]
>>
>> For the lower level [1] cope I think this is fine, but for the [-1] lower-level cope, I don’t think this is ok because I am interested in the negative values.
>>
>> How would you suggest setting up this analysis, or what steps am I missing, to correctly compare the deactivation between the groups?
>>
>> Any guidance would be greatly appreciated.
>>
>> Thanks,
>> Liz
>>
>>
>> Elizabeth J. Woytowicz, PhD Student
>> Department of Physical Therapy & Rehabilitation Science
>> University of Maryland, Baltimore School of Medicine
|