Dear Muhammad,
On Fri, May 7, 2010 at 3:18 AM, MP <[log in to unmask]> wrote:
> Hi Vladimir,
> Thanks so much for attending to this issue.
> Yes, when I display the TF maps, they are displayed just like the file you
> have attached, but when I click on Info --> Channels --> 3D locations, it
> has a totally different channel layout? Which one should I believe?
The 3D locations are absolutely not relevant for this analysis. As
long as 2D locations are OK there is no reason to worry.
> The way I am doing the stats is:
> 1. I convert all the averaged TFs from each subject to an Image. This gives
> me 4 images for each subject (since I have 4 conditions).
> 2. Then I do a 2nd level Independent t-test to compare the conditions for
> all the subjects (as I shown in my previous email to you)
> 3. That comparison gives me the electrode location and the time that i have
> a significant difference at. (By the way, I only look at cluster-level
> FWE/FDR p-values, and if they are >0.05 I consider that result to be
> significant, is that correct for EEG data?)
Yes, using FWE is correct.
> 4. Then I use the script that you had sent me to extract the averaged power
> from the electrode and the time window that were significant in the previous
> code.
> D = spm_eeg_load('filename.mat');
>
> data = D.selectdata('electrode', time, {'Cond1', 'Cond2', 'Cond3',
> 'Cond4'});
> and here the I load the mrtf1_xxxxxxx.mat file that was created by the
> convert2image step in addition to creating the images for each condition.
> 5. I take the values and paste them in SPSS for further analysis.
> 6. Now, when I do a t-test on the conditions, which showed significant
> result in SPM, it doesnt give a significant result in SPSS, not even close
> (p>0.2). That is why i think that there may be something wrong with the
> channel registration which is confirmed by the 3D channel map.
What you can try is grand-average the mrtf1... files and then just
look at the grand average or use Other/Contrast to compute the
contrast of interest there and display it. The results should be
consistent with what you got from the images.
Best,
Vladimir
> Thanks
> -Muhammad
>
> On Thu, May 6, 2010 at 7:28 PM, Vladimir Litvak <[log in to unmask]>
> wrote:
>>
>> I looked at your example and I don't see a problem. You select a
>> stripe around the midline and that's what the result looks like (see
>> attached). If you can't figure out what is happening with your
>> statistics I can look at it again.
>>
>> Best,
>>
>> Vladimir
>>
>> On Thu, May 6, 2010 at 9:51 PM, Vladimir Litvak
>> <[log in to unmask]> wrote:
>> > On Thu, May 6, 2010 at 7:56 PM, MP <[log in to unmask]> wrote:
>> >> When I extract the average power from an electrode and the time
>> >> duration
>> >> from the stats, and say do a t-test outside of SPM (in SPSS) I don't
>> >> get a
>> >> significant difference. That's why I think it is because the electrodes
>> >> are
>> >> mis-registered in my TF data.
>> >
>> > That's indeed strange but if the electrode wrong that could explain it.
>> >
>> >>
>> >> On a related note, how can I find out exactly which electrode the stats
>> >> are
>> >> pointing to? Currently I am going by just eye-balling the glass brain.
>> >
>> > Right click on the glass brain and you should get a pop-up menu with
>> > 'show labels' option. I'll look at your file.
>> >
>> > Vladimir
>> >
>> >>
>> >> Thanks
>> >>
>> >> On Thu, May 6, 2010 at 2:50 PM, Vladimir Litvak
>> >> <[log in to unmask]>
>> >> wrote:
>> >>>
>> >>> I meant what's wrong with the statistics?
>> >>>
>> >>> Vladimir
>> >>>
>> >>> On Thu, May 6, 2010 at 7:39 PM, MP <[log in to unmask]> wrote:
>> >>> > If you look at the electrode positions on the .MAT file i just sent
>> >>> > you
>> >>> > with
>> >>> > the .DAT file I sent from the yousendit, you will see the electrode
>> >>> > position
>> >>> > is totally wrong.
>> >>> >
>> >>> > On Thu, May 6, 2010 at 2:33 PM, Vladimir Litvak
>> >>> > <[log in to unmask]>
>> >>> > wrote:
>> >>> >>
>> >>> >> The picture looks very well. So why do you think this is wrong?
>> >>> >>
>> >>> >> Vladimir
>> >>> >>
>> >>> >> On Thu, May 6, 2010 at 7:27 PM, MP <[log in to unmask]> wrote:
>> >>> >> > Hi Vladimir,
>> >>> >> >
>> >>> >> > I just sent you the .DAT part of the file that I feed in
>> >>> >> > spm_eeg_tf,
>> >>> >> > and
>> >>> >> > attached is the .MAT part of it (because I could not attach both
>> >>> >> > files
>> >>> >> > at
>> >>> >> > the same time in yousendit). Also I have attached what I meant by
>> >>> >> > 'apparently plausible statistical results'.
>> >>> >> >
>> >>> >> > Thanks
>> >>> >> >
>> >>> >> > -Muhammad
>> >>> >> >
>> >>> >> > On Thu, May 6, 2010 at 4:22 AM, Vladimir Litvak
>> >>> >> > <[log in to unmask]>
>> >>> >> > wrote:
>> >>> >> >>
>> >>> >> >> Dear Muhammad,
>> >>> >> >>
>> >>> >> >>
>> >>> >> >> On Wed, May 5, 2010 at 3:09 PM, MP <[log in to unmask]> wrote:
>> >>> >> >> > Hi Vladimir and other SPM EEG experts,
>> >>> >> >> > I am averaging my ERP data both in time and time/frequency
>> >>> >> >> > domains
>> >>> >> >> > and I
>> >>> >> >> > am
>> >>> >> >> > using SPM's default electrode locations for 64 (10-20)
>> >>> >> >> > electrodes.
>> >>> >> >> > In
>> >>> >> >> > my
>> >>> >> >> > time domain averages, the electrode locations look fine, while
>> >>> >> >> > in
>> >>> >> >> > my
>> >>> >> >> > time-freq data, all the electrode (I am selecting 24 of the 64
>> >>> >> >> > electrodes
>> >>> >> >> > for the TF decomposition and a frequency window of 3 - 30 Hz)
>> >>> >> >> > seem
>> >>> >> >> > to
>> >>> >> >> > be
>> >>> >> >> > located on the front of the head.
>> >>> >> >> > The section of my script that does TF decomp is as follows:
>> >>> >> >> > ...
>> >>> >> >> > S = [];
>> >>> >> >> > S.D = <filename>;
>> >>> >> >> > S.tf.frequencies = [3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
>> >>> >> >> > 19 20
>> >>> >> >> > 21
>> >>> >> >> > 22
>> >>> >> >> > 23
>> >>> >> >> > 24 25 26 27 28 29 30];
>> >>> >> >> > S.tf.Mfactor = 5;
>> >>> >> >> > S.tf.channels = [1 2 3 8 10 12 17 19 21 26 28 30 36 38 49 46
>> >>> >> >> > 48 50
>> >>> >> >> > 54
>> >>> >> >> > 56
>> >>> >> >> > 58
>> >>> >> >> > 60 61 62];
>> >>> >> >> > S.tf.phase = 0;
>> >>> >> >> > S.tf.pow - 1;
>> >>> >> >> > D = spm_eeg_tf(S);
>> >>> >> >> > ...
>> >>> >> >>
>> >>> >> >> I think I'll need a data example to understand what you mean.
>> >>> >> >> Could
>> >>> >> >> you send me the file that you use as input to spm_eeg_tf via
>> >>> >> >> yousendit?
>> >>> >> >>
>> >>> >> >> >
>> >>> >> >> > However, when I convert these TF averages to images and do
>> >>> >> >> > stats
>> >>> >> >> > on
>> >>> >> >> > them, I
>> >>> >> >> > see results around the right (original) electrode sites, but
>> >>> >> >> > when
>> >>> >> >> > i
>> >>> >> >> > extract
>> >>> >> >> > the power for that electrode and do stats outside of SPM, the
>> >>> >> >> > results
>> >>> >> >> > don't
>> >>> >> >> > correspond to the SPM results. Is it due to mis-registration
>> >>> >> >> > of
>> >>> >> >> > electrode
>> >>> >> >> > during the TF decomp step?
>> >>> >> >> >
>> >>> >> >>
>> >>> >> >> Here it's also not quite clear what kind of differences you get.
>> >>> >> >> Please provide more details.
>> >>> >> >>
>> >>> >> >> Vladimir
>> >>> >> >
>> >>> >> >
>> >>> >
>> >>> >
>> >>
>> >>
>> >
>
>
|