Dear Marco,
Indeed computation of source power is nonlinear so the baseline
contributions will not cancel out. I think the best solution would be
to use high-pass filter rather than baseline correction and raise the
cutoff if necessary to remove drifts.
Best,
Vladimir
On Thu, Dec 6, 2012 at 10:50 AM, Marco Buiatti <[log in to unmask]> wrote:
> Vladimir and Urs, thanks for your replies.
>
> My reason for baseline correction is to minimise differences between
> data epochs belonging to different conditions which are not dependent
> on the experimental manipulation - e.g. low frequency drifts or
> residual artefacts - as these differences could spuriously bias my
> data analysis.
>
> Since these spurious differences increase if the baseline is taken
> further away from the relevant response, as far as I know, in the ERP
> literature it is common to remove the baseline just before the
> relevant stimulus, even in cases where the baseline might contain
> previously evoked responses - see priming studies focused on the N400.
>
> However, I seem to understand that this reasoning is not always valid
> when looking at differences in source space, at least in the case in
> which I first source-invert each condition separately, and in a second
> step I test the difference between the two conditions in source space.
> In this case, if I perform source inversion on EEG data
> baseline-corrected on a temporal window containing evoked potentials,
> for sure the single-condition source reconstruction will be distorted.
> Now, if source inversion were perfectly linear, this would not harm if
> I am only interested in comparing two conditions, but I am afraid
> source inversion is not completely linear right?
>
> Vladimir, I guess what you are suggesting is to perform source
> reconstruction on the data epoched without any baseline correction
> right? In this case, drifts would be only attenuated by the high-pass
> filter I have previously applied to continuous data, but source
> inversion should be more reliable.
>
> Thanks for telling me if this reasoning looks wrong to you.
>
> Best,
>
> Marco
>
>
> On 5 December 2012 22:50, Vladimir Litvak <[log in to unmask]> wrote:
>> Dear Marco,
>>
>> I don't quite understand at what point you need that baseline and why you don't compare the activation images to each other without any baseline.
>>
>> Best,
>> Vladimir
>>
>>
>>
>> On 5 Dec 2012, at 16:41, Marco Buiatti <[log in to unmask]> wrote:
>>
>>> Dear SPM Masters,
>>>
>>> Here's a more general and synthetic version of my previous question.
>>>
>>> My doubt: what is the optimal baseline for EEG source reconstruction,
>>> and specifically:
>>> 1) Is it problematic to set the baseline in a period that contains
>>> (sensory) evoked responses? Is there a risk of suppressing the sources
>>> of those evoked responses?
>>> 2) Otherwise, is it problematic to set a baseline much before the
>>> relevant stimulus (in my case, 800 ms before)? How do slow drifts in
>>> the data impact on source reconstruction?
>>>
>>> Thanks,
>>>
>>> Marco
>>>
>>>
>>> On 3 December 2012 11:59, Marco Buiatti <[log in to unmask]> wrote:
>>>> Dear SPM Masters,
>>>>
>>>> I have yet another doubt on EEG source reconstruction which might be
>>>> of general interest. I have a priming design in which the target
>>>> stimulus is preceded by a prime stimulus, presented either 800 ms
>>>> before (if it's auditory) or 360 ms before (if it's visual). In my
>>>> simplest case, I want to test the difference between two different
>>>> conditions, both comprising both visual and auditory primes. My
>>>> problem is to choose the most appropriate baseline for EEG source
>>>> reconstruction.
>>>>
>>>> At the sensor level, since I am testing a difference that should not
>>>> depend on the evoked response of the prime, it is common use to take
>>>> the baseline in the 100-200 ms window before the target presentation.
>>>>
>>>> For source reconstruction, is the fact that the pre-target baseline
>>>> contains the evoked responses of the prime problematic? Could it
>>>> suppress the same kind of responses and/or generate "phantom
>>>> responses"?
>>>> Should I choose a pre-prime baseline (e.g. 800 ms before the target),
>>>> at the risk of having more variability due to low frequency drifts?
>>>> Data have been high-pass filtered at 0.3 Hz before epoching, in this
>>>> case I might increase the high-pass filter cut-off of the continuous
>>>> data.
>>>>
>>>> Thanks a lot for your feedback,
>>>>
>>>> Best,
>>>>
>>>> Marco
>>>>
>>>> --
>>>> Marco Buiatti, PhD
>>>>
>>>> CEA/DSV/I2BM / NeuroSpin
>>>> INSERM U992 - Cognitive Neuroimaging Unit
>>>> Bāt 145 - Point Courrier 156
>>>> Gif sur Yvette F-91191 FRANCE
>>>> Ph: +33(0)169.08.65.21
>>>> Fax: +33(0)169.08.79.73
>>>> E-mail: [log in to unmask]
>>>> http://www.unicog.org/pm/pmwiki.php/Main/MarcoBuiatti
>>>>
>>>> ***********************************************
>>>
>>>
>>>
>>> --
>>> Marco Buiatti, PhD
>>>
>>> CEA/DSV/I2BM / NeuroSpin
>>> INSERM U992 - Cognitive Neuroimaging Unit
>>> Bāt 145 - Point Courrier 156
>>> Gif sur Yvette F-91191 FRANCE
>>> Ph: +33(0)169.08.65.21
>>> Fax: +33(0)169.08.79.73
>>> E-mail: [log in to unmask]
>>> http://www.unicog.org/pm/pmwiki.php/Main/MarcoBuiatti
>>>
>>> ***********************************************
>
>
>
> --
> Marco Buiatti, PhD
>
> CEA/DSV/I2BM / NeuroSpin
> INSERM U992 - Cognitive Neuroimaging Unit
> Bāt 145 - Point Courrier 156
> Gif sur Yvette F-91191 FRANCE
> Ph: +33(0)169.08.65.21
> Fax: +33(0)169.08.79.73
> E-mail: [log in to unmask]
> http://www.unicog.org/pm/pmwiki.php/Main/MarcoBuiatti
>
> ***********************************************
|