My experience is that large amplitude non-rapid motions must be solved in a separate way than the artifacts from rapid motions. For example, Lemieux, et al. (2007) uses 24 motion regressors for motion and adds null regressors for artifacts. Similarly, ArtRepair has separate functions for motion regression and artifact "repair and deweight". As you suggest, it is plausible to mix and match, e.g. use ArtRepair motion regression and add null regressors for artifacts, as long as both problems are addressed.
The motion regression method in ArtRepair is intended to be more robust for large motions and possible motion-induced activations than using motion regressors. After some in-house successes, we made the software available. However, the relative performance of these methods in all situations is still unknown. Feedback will be very helpful to make further improvements!
Best regards,
Paul
----- Original Message -----
From: "Vy Dinh" <[log in to unmask]>
To: "Paul Mazaika" <[log in to unmask]>, [log in to unmask]
Sent: Wednesday, December 8, 2010 2:34:44 PM
Subject: RE: [SPM] Use of ArtRepair-large differences
Hi Paul & Marko,
Do you propose using ArtRepair to address the issue of motion-induced activation as an alternative (or addition) to nulling "bad" scans during 1st level model estimation? Also, according to the ArtRepair manual, the motion adjustment performed by art_motionregress only applies the regression value obtained from low motion to all scans. To address scans with high motion, we would need to use the repair functions?
Best,
Vy T.U. Dinh
Research Assistant, Neurological Sciences
Rush University Medical Center
Phone: (312) 563-3853
Fax: (312) 563-4660
Email: [log in to unmask]
________________________________________
From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Paul Mazaika [[log in to unmask]]
Sent: Wednesday, December 08, 2010 12:18 PM
To: [log in to unmask]
Subject: Re: [SPM] Use of ArtRepair-large differences
Nicole,
I agree with Chris that the rapid motion artifacts could explain most of the difference.
Another risk in this data is the total range of motion (4 degrees, 2 mm) may be too large to be fully corrected by six motion regressors. Using 24 motion regressors (Lund, 2005) may be better. I suggest trying the art_motionregress program, use it before the artifact repair, and then not use the motion regressors. This approach has worked pretty well with our pediatric subjects, and your data set has similar motion.
If different methods produce the same activation maps, then your result is robust over methods. If they produce very different maps, then you need to check if the GLM assumptions of stationary noise on every voxel were satisfied. The gold standard is to examine the residuals as described in Luo and Nichols, 2003. A quick qualitative alternative is to view all the images after the motion and repair steps using the Contrast Movie. If the movie has no bright flashes, then all the large artifacts and motion effects have been suppressed during preprocessing before the GLM.
Good luck,
Paul
----- Original Message -----
From: "Chris Watson" <[log in to unmask]>
To: [log in to unmask]
Sent: Wednesday, December 8, 2010 8:12:58 AM
Subject: Re: [SPM] Use of ArtRepair-large differences
Yea, all those large movements around scan 450 or so can make a big
difference when repaired.
Nicole Van Hoeck wrote:
> Hi Christopher and Paul,
>
> Thank you for your help.
>
> Attached you can find the realignment and art_global plots after only the slices being repaired.
>
> Nicole
>
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Paul Mazaika
> Sent: dinsdag 7 december 2010 20:02
> To: [log in to unmask]
> Subject: Re: [SPM] Use of ArtRepair-large differences
>
> The figures show more (7 vs. 2) significant clusters using the repaired data. The difference is more than usual, but the presence of artifacts and large motions can strongly reduce estimated effects. Since there are 1081 degrees of freedom in the not-repaired option, and 708 dof in the repaired option, there were likely about 373 scans marked for deweighting in the GLM. Generally, I suggest limiting repairs to fewer than 15-20% of the scans by raising the threshold, even for difficult head motions. It may help to try an example without using clipping, and to raise the threshold by one increment to get rid of any excess deweighted scans. If the activations remain after this process, I would be fairly confident in them.
>
> Generally, I don't recommend using the "Clipping" option, which was built to support the rule of thumb of 3 mm maximum motion, and kept in the GUI as a legacy capability. Using art_motionregress is more effective. If art_motionregress is used, it is not necessary to add the six motion regressors to the GLM.
>
> As mentioned in the previous post, it will help to see the realignment and art_global plots.
>
> Best regards,
> Paul
>
> ----- Original Message -----
> From: "Christopher Watson" <[log in to unmask]>
> To: [log in to unmask]
> Sent: Tuesday, December 7, 2010 9:00:23 AM
> Subject: Re: [SPM] Use of ArtRepair-large differences
>
> The difference could be "normal". It depends on how much motion there was. Can you post the realignment plot and/or the art_global plot?
> ________________________________________
> From: SPM (Statistical Parametric Mapping) [[log in to unmask]] On Behalf Of Nicole Van Hoeck [[log in to unmask]]
> Sent: Tuesday, December 07, 2010 11:33 AM
> To: [log in to unmask]
> Subject: [SPM] Use of ArtRepair-large differences
>
> Dear SPM users,
>
> I’m using the toolbox ArtRepair for the first time. I’m surprised how much my results change. Therefore, I start to doubt myself in correctly applying this program.
>
> This is the procedure I followed:
>
> 1) art_slice before preprocessing (Number of slices repaired: 515 = 1.5%)
>
> 2) art_global: ‘repair’ volumes (+ use of ‘clip’)
>
> 3) 1st model with the v-files (repaired files) and motion parameters as regressors
>
> 4) art_summary to compare repaired model to non-repaired model
> Attached you can find the results (repaired and not repaired) of one of my subjects for one contrast.
>
> Is the difference in results ‘normal’?
>
> Thank you,
> Nicole
>
> --
> Nicole Van Hoeck
> Department of Clinical & Lifespan Psychology | Department of Experimental & Applied Psychology
> Faculty of Psychology & Educational Science
> Vrije Universiteit Brussel
> Pleinlaan 2 | 1050 Brussels | Belgium
>
> office: 3C235
> phone: +32-(0)-2 / 629 25 26
> email: [log in to unmask]<mailto:[log in to unmask]>
>
>
>
> ------------------------------------------------------------------------
>
--
Paul K. Mazaika, PhD.
Center for Interdisciplinary Brain Sciences Research
Stanford University School of Medicine
Office: (650)724-6646 Cell: (650)799-8319
CONFIDENTIALITY NOTICE: Information contained in this message and any
attachments is intended only for the addressee(s). If you believe
that you have received this message in error, please notify the
sender immediately by return electronic mail, and please delete it
without further review, disclosure, or copying.
--
Paul K. Mazaika, PhD.
Center for Interdisciplinary Brain Sciences Research
Stanford University School of Medicine
Office: (650)724-6646 Cell: (650)799-8319
CONFIDENTIALITY NOTICE: Information contained in this message and any
attachments is intended only for the addressee(s). If you believe
that you have received this message in error, please notify the
sender immediately by return electronic mail, and please delete it
without further review, disclosure, or copying.
|
