Dear Michelle,
You should really include a regressor for those that you have catagorised as 'bad', or you have altered your effective degrees of freedom in you data without properly accounting for it. However this extra regressor makes the interpolation pointless because the effect of the regressor is to remove the variance associated with that volume.
You will find the approach of adding regressors to account for large motion events in a GLM described in the following paper.
Modelling large motion events in fMRI studies of patients with epilepsy.
Lemieux L, Salek-Haddadi A, Lund TE, Laufs H, Carmichael D.
Magn Reson Imaging. 2007 Jul;25(6):894-901.
http://www.ncbi.nlm.nih.gov/pubmed/17490845
You can also find a very effective alternative/addition to art repair
FIACH: A biophysical model for automatic retrospective noise control in fMRI.
Tierney TM, Weiss-Croft LJ, Centeno M, Shamshiri EA, Perani S, Baldeweg T, Clark CA, Carmichael DW.
Neuroimage. 2015 Sep 28;
http://www.ncbi.nlm.nih.gov/pubmed/26416652
You can download the tools here http://www.homepages.ucl.ac.uk/~ucjttie/index.html
Best regards
David
David Carmichael
Reader in Neuroimaging and Biophysics
Honorary Reader, Radiology, Great Ormond Street Hospital NHS Foundation Trust
Developmental Imaging and Biophysics Section
Developmental Neurosciences Program
UCL Institute of Child Health
30 Guilford Street
London, UK
WC1N 1EH
Tel +44 (0) 2079052298
Fax +44 (0) 2079052358
http://www.action.org.uk/our_research/epilepsy_improving_brain_scanning_surgery
https://iris.ucl.ac.uk/research/personal/index?upi=DWCAR66
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Achterberg, M.
Sent: 04 November 2015 09:02
To: [log in to unmask]
Subject: [SPM] FW: Artrepair and Motion Regressors
-----Original Message-----
From: Achterberg, M.
Sent: woensdag 4 november 2015 10:02
To: 'H. Nebl'
Subject: RE: Artrepair and Motion Regressors
Dear Helmut,
Thank you for your advice!
What we've now done is the second option, so replacing bad volumes by interpolated good ones.
Would you then recommend including one regressor with all repaired volumes in it, or a separate regressor for each repaired volume?
Thanks in advance!
Best, Michelle
-----Original Message-----
From: H. Nebl [mailto:[log in to unmask]]
Sent: dinsdag 3 november 2015 14:27
To: [log in to unmask]; Achterberg, M.
Subject: Re: Artrepair and Motion Regressors
Dear Michelle,
I'd say you have to provide some more details on the Artrepair features you want to use for your preprocessing, as it provides several options: A very simple strategy (which I personally prefer due to its simplicity) would be to just calculate the scan-to-scan motion, then go on with the default preprocessing and add dummy regressors to the design matrices for each of the bad volumes (possibly also for the preceding and the successive volumes) when setting up the models, in addition to the rp file. This way you would model each of the bad volumes separately. As normalisation and smoothing might reduce global signal artefacts it might be better to rely on the realigned files in that case.
However, you can also replace bad volumes by interpolated good ones. You would usually still want to deweight them from analysis though, as it's artificial data, and you would still add the rp files (as you "controlled" for the volumes with fast head motion but not for the slow changes over time). This might be somewhat problematic though, as the rp files might still include jerks that are no longer present in the data (because the volumes were replaced by other data, resulting in a smoother transition).
Then, there's also the "Motion Adjustment" feature (see http://cibsr.stanford.edu/content/dam/sm/cibsr/documents/tools/methods/artrepair-software/MotionandDespike.pdf ), which is meant as a replacement for adding rp files. The proposed preprocessing pipeline is described in the manual.
Best
Helmut
|