Thanks a lot Mark that clears things up
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Mark Jenkinson [[log in to unmask]]
Sent: Saturday, July 13, 2013 12:47 PM
To: [log in to unmask]
Subject: Re: [FSL] AW: [FSL] Motion artefact correction: Melodic ICA and motion parameters?
Dear Helen,
If there is correlation between regressors of interest (including signals from seed regions) and the motion regressors then they can change the results. It can occur in either direction, more or less "activation", depending on the amount and sign of the correlation.
As for ordering - spatial smoothing can occur either before or after, as it is a linear operation. For temporal filtering you should either use unfiltered motion regressors applied to unfiltered data, or use filtered motion regressors applied to filtered data.
All the best,
Mark
On 13 Jul 2013, at 10:16, Helen Sawaya <[log in to unmask]>
wrote:
> Dear Mark
>
> I'm cutting in because I also have a concern about adding the motion parameters as regressors. I am regressing them out using fsl_regfilt. Does it matter at which point in the analysis they are regressed? For example is it okay if we regress them out after the functional images have been smoothed but before they have been high pass filtered?
>
> Also, is it possible that regressing out the 6 motion parameters would actually add some spurious correlations? I am trying to understand why after regressing motion, I see some added correlation between my seed and regions in the CSF for a few subjects, but less so when i did not regress motion. (I regressed out signal from CSF)
>
> Thank you!
> Helen
>
> ________________________________________
> From: FSL - FMRIB's Software Library [[log in to unmask]] On Behalf Of Mark Jenkinson [[log in to unmask]]
> Sent: Saturday, July 13, 2013 11:20 AM
> To: [log in to unmask]
> Subject: Re: [FSL] AW: [FSL] Motion artefact correction: Melodic ICA and motion parameters?
>
> Hi,
> It isn't quite that simple. The motion regressors may still have explanatory power, even after ICA denoising. It depends on how exact the denoising was. Also, there may be correlations between your motion regressors and regressors of interest, and in this case you can't say that there will be no effect of adding the motion regressors. Therefore it depends on both the quality of the denoising and the correlation to regressors of interest, and there is no simple answer.
> All the best,
> Mark
>
>
> "Baeuchl, Christian" <[log in to unmask]> wrote:
>
>
> Hi Mark,
>
> thanks for your quick and helpful response!
>
> I just have one follow-up question. If I understand you correctly the worst thing that the motion parameters in my GLM could do is to reduce the degrees of freedom in the model without having any additional explanatory value (in a case where all motion-related artifacts have been removed with MELODIC). Regressors of no interest, like the motion parameters, wouldn't distort my statistics if they don't explain any variance in the data. Is that correct?
>
> Best,
> Christian
>
> -----Ursprüngliche Nachricht-----
> Von: FSL - FMRIB's Software Library [mailto:[log in to unmask]] Im Auftrag von Mark Jenkinson
> Gesendet: Dienstag, 9. Juli 2013 23:32
> An: [log in to unmask]
> Betreff: Re: [FSL] Motion artefact correction: Melodic ICA and motion parameters?
>
> Hi,
>
> This looks sensible to me.
> The addition of the motion parameters will soak up any residual motion changes that weren't taken out by the ICA denoising. So it is still a reasonable option.
>
> All the best,
> Mark
>
>
> On 9 Jul 2013, at 17:24, Christian Bäuchl <[log in to unmask]> wrote:
>
>> Dear FSL experts,
>>
>> I mainly use SPM8 for preprocessing and for analyzing my functional data, but for artifact detection I would also like to use MELODIC ICA as a part of my preprocessing pipeline.
>>
>> In my dataset I have subjects who show substantial movement during scanning. In order to minimize those movement related artifacts as much as possible I want to denoise my data using MELODIC and also use the motion parameters as derived from the realignment step.
>> I'm quite unsure though if the order AND the choice of steps within my processing- and analysis pipeline is correct/valid.
>>
>> A short outline of my steps for a single subject:
>>
>> [in SPM8:]
>> Slice timing correction - Realignment of scans(motion correction)
>>
>> [in FSL 5.0:]
>> Convert realigned 3D images into 4D - perform high-pass filtering and BET on data (no further preprocessing) - calculate IC's, check for artifacts and remove artifactual IC's from data - convert 4D back into 3D images
>>
>> [in SPM8:]
>> Normalization of functional to MNI - smoothing the data - perform statistical analysis including the motion parameters from the realignment step into the GLM
>>
>> I don't know if it makes sense to include the motion parameters in the GLM after I've already used MELODIC to remove motion (and other) artifacts. In a previous post Mark Woolrich mentioned that using MELODIC in conjunction with motion parameters in the GLM is okay (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;6732ebcd.0911).
>>
>> But I don't know if this is also valid if I follow the preprocessing and analysis steps as described above. If the motion parameters are derived BEFORE I correct for motion artifacts using MELODIC, aren't they useless if they are subsequently fed into the GLM for statistical analysis?
>>
>> Any help would be much appreciated,
>> Christian
|