Dear Kambiz,
In addition to what Rik and Vladimir said, the results section display
should be different if SPM is running in the EEG modality (an extra
question will be asked for "Data Type" for which you should answer
"Scalp-Time").
The number of voxels listed in the table corresponds to the number of
voxels within the mask: you get 2723604 instead of 2768896 (64*64*676)
because some voxels do not contain any signal (NaN) from your
interpolated scalp data (at the "corners" of the scalp, 67 voxels at a
time).
You can find more information about Small Volume Correction in the SPM8
manual p.344:
http://www.fil.ion.ucl.ac.uk/spm/doc/manual.pdf
with a useful link to the Cambridge MEG wiki.
You will also find more information about factorial designs here:
http://d.gitelman.googlepages.com/conweights.pdf
Best regards,
Guillaume.
Vladimir Litvak wrote:
> Dear Kambiz,
>
> In addition to what Rik said, the presentation of your results in the
> figure seems a little strange. If you are using the latest SPM update
> you should select the Scalp x Time option when you present the results
> and then you won't get the brain outline in the background and it'll
> be possible to see the time dimension more clearly.
>
> Also in contrast to what you said there seems to be something
> significant in this figure with p<0.01 unc, although indeed nothing
> that passes the FWE correction.
>
> Best,
>
> Vladimir
>
> On Wed, Jan 13, 2010 at 3:39 PM, Kambiz Tavabi <[log in to unmask]> wrote:
>> Dear Forum,
>>
>> In trying to understand the results out put for a 2x2x2 MEEG ANOVA. I fail
>> to understand how the number of voxels is derived. Data was acquired at 1KHz
>> no filtering, downsampled to 600 Hz, epochs of 676 samples (200 ms baseline)
>> were filtered with a 1.5 Hz high-pass. The data was then baseline corrected
>> and scanned for artifacts using peak-to-peak algorithm with an amplitude
>> threshold of 3.5 pT, the data was averaged and converted to images by
>> interpolating to 64x64 image size. Images were smoothed using a (6,6,5)
>> kernel. We assume that the number of voxels in the results (see attached
>> PDF) is derived by 64x64x676?
>> Application of FWE or a conservative P value (<0.01) results in nothing, and
>> thresholding 0.01<p<0.05 results in everything significant. Any advise to
>> find a sensible middle ground in the results would be appreciated. We plan
>> to use a (6,6,7) kernel. But to improve the situation is it advisable to
>> reduce the number of corrections by smoothing to a smaller image, or using a
>> smaller epoch size?
>>
>> Thanks
>>
>
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|