Dear Sun,
2010/5/1 Sun Delin <[log in to unmask]>:
> Dear Vladimir,
>
> Thanks for your reply. I am so happy to tell you that I have found a way to get the significant result: first, get the group inversion result of baseline (in my study, it is -200~0ms before face stimuli onset); second, caculate the contrast between each condition and baseline by spm_imcalc (simply 'i1-i2'); third, put these new contrasts into 2nd level analysis. Attached are spmT images for P1 (70~129ms after face stimuli onset) and N170 (130~179ms after face stimuli onset). Significant activity could be seen by XJview at bilaterial occipital areas for P1 and bilaterial fusiform for N170 (dorminent in right side when I checked it by press 'goto global maxima' and then 'goto nearest suprathreshold channel'. However, the image seems left-right reversed). I greatly appreciate your talented method for group inversion, and I think with your help I could get both precise temporal (ms) and spatial (cm?) information about brain activity in ERP studies now.
I'm not familiar wit XJview, but if you open all the images together
using the CheckReg button in SPM you should see whether some are
inverted. That shouldn't happen at least as long as you only use SPM
for you processing and from what you sent me it doesn't seem that some
images are inverted relative to others.
> I still have some questions:
> 1. why should I make the new contrasts by subtracting group inversion results of baseline from those of conditions? In my opion, the averaged scalp results could be viewed as the difference betwwen conditions and baseline already, so why should I do that contrast again in group inversion results?
If you are comparing between conditions you don't necessarily need to
subtract the baseline, however you seem to have found in practice that
it's a good thing to do. In general comparing baseline with activation
and doing statistical tests just on that might not be a good idea
because you know in advance that the null hypothesis is not true so
you might find many non-specific activations. The reason you might
have to do that sometimes is related to your second question.
> 2. what is the meaning of the values from group inversion? Are they intensities of dipoles? What about the polarities?
>
The values you get from the group inversion are source power
normalized to the mean over all sources. So directionality information
is not preserved and the values are non-negative. That's why it
doesn't make sense to do a single-sample t-test on them. Your argument
holds for sensor-level data where indeed you can do a single-sample
t-test in SPM without subtracting anything.
Best,
Vladimir
> Best regards,
>
> Sun Delin
>
> 2010-05-01
>
> ======= At 2010-04-30, 19:08:10 you wrote: =======
>
>>Dear Sun,
>>
>>In the case of inversion results the [1 1 1 1] contrast is not very
>>meaningful because the images are always different from zero. So you
>>can show your average beta image but there is no statistically
>>meaningful question you can ask about it. However, it would be nice to
>>get the peak of the F statistic in the right place, which doesn't seem
>>to happen. I have a similar example from my own data that I'm planning
>>to look into and I'll let you know if I find the problem or need your
>>data. Do you get meaningful results from your other contrasts?
>>
>>Vladimir
>>
>>
>>
>>2010/4/30 Sun Delin <[log in to unmask]>:
>>> Dear Vladimir,
>>>
>>> ? I am sorry that I have to send you the email directly because the email size exceeds 500K limit of SPM mailing list. I met a problem when I was doing 2nd level analysis on ERP group inversion data. I am trying to locate the source of the P1 component in a 2*2 factorial designed visual experiment. Averaged data for each condition of each subject were included in the group inversion processing, and the smoothed output images (FWHM = 8mm) were involved in the 2nd level group analysis of full factorial design. The result of group inversion is really encouraging that I could find significant activation at occipital regions during 70-129ms after visual stimuli onset (an example for one condition in an individual subject could be seen in 'sw_sbj214-1_4.nii' attached). The beta images after 2nd level analysis are also beautiful (an example please see 'beta_0004.hdr/img' attached). However, the results of t or F test (please see 'spmF_0001.hdr/img' attached. I would like to see t!
>>> he P1 component in all conditions, therefore the contrast is [1 1 1 1]) seem greatly different from traditional SPM results. I have tried normalization (proportion scaling or ANCOVA) but found little help. Would you please have a look and give me some advice?
>>> ? BTW, is there any difference between fMRI and EEG in the 2nd level analysis? It seems that estimation of fMRI could provide some main and interaction results automatically.
>>>
>>>
>>> Bests,
>>> Sun Delin
>
> = = = = = = = = = = = = = = = = = = = =
>
>
>
>
|