Thank you Donald. Great - now it's clear to me about the motion regressors. 

On Tue, Jun 9, 2015 at 3:07 PM, MCLAREN, Donald <[log in to unmask]> wrote:

See below.

On Tue, Jun 9, 2015 at 4:45 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Donald,

Thanks for your response. I actually already have motion correction done as well in FSL, and have the .par motion parameter output, which I will input as a regressor into SPM following the advice of some others on the mailing list.

Yes. The par file values would be fine. 

SPM uses a non-linear normalization approach. The non-linear approach is actually an affine transform+non-linear deformations. This is similar to FNIRT in FSL.

So more specifically, does SPM first put functional to T1 space, and then from there to standard space? That's the procedure I'm using with FSL.

No. Normalization takes an image and normalizes it to MNI space. This is the same in FSL. If you are saying you coregistered the EPI to the T1 and then normalized the T1 in FSL, you could do that in two steps in SPM as well. The programs are very similar.

 

As to whether to include the motion parameters in the analysis. The inclusion of the motion parameters usually reduces the noise in the model and could lead to more accurate estimates; however, if you move with each stimulus, then the motion would be correlated with the predicted HRF and this would reduce the accuracy of the estimates. As long as you don't have stimulus correlated motion, I would include them as a covariate.

That's an interesting point. Aren't the motion parameters included as "Multiple Regressor"? I am fairly new to this, so correct me if I am wrong. I mean that I would expect that the motion is REGRESSED OUT from the HRF. So any patterns in the motion are being excluded from the brain activity patterns....

Yes. They are regressed out of the HRF task regressors. Thus, the patterns of motion are being excluded. The issue is that if the "timecourse" of motion is correlated with the HRF task regressors, then you will remove the brain activity effect. For example, say I rotate my head forward at the beginning of each block and back at the end of each block. The "timecourse" of motion will be correlated with my motion. If I include the motion "timecourse" as a multiple regressor, I might remove the block effects in my data - particularly if the rotation is delayed by a few seconds.

 

Thanks,

Joelle

On Mon, Jun 8, 2015 at 8:57 PM, MCLAREN, Donald <[log in to unmask]> wrote:

If you use realignment in SPM, then you should use these images as the input into normalization (FSL or SPM). I would not repeat the realignment in FSL and then do normalization in FSL. You also need to smooth the data after normalization.

I don't see why you would use SPM for some pre-processing, then switch to FSL for a single step, then switch back for smoothing. I would do all the pre-processing in FSL if you prefer the FSL normalization.

SPM uses a non-linear normalization approach. The non-linear approach is actually an affine transform+non-linear deformations. This is similar to FNIRT in FSL.

As to whether to include the motion parameters in the analysis. The inclusion of the motion parameters usually reduces the noise in the model and could lead to more accurate estimates; however, if you move with each stimulus, then the motion would be correlated with the predicted HRF and this would reduce the accuracy of the estimates. As long as you don't have stimulus correlated motion, I would include them as a covariate.

Best Regards, Donald McLaren
=================
D.G. McLaren, Ph.D.
Research Fellow, Department of Neurology, Massachusetts General Hospital and
Harvard Medical School
Postdoctoral Research Fellow, GRECC, Bedford VA
Website: http://www.martinos.org/~mclaren
Office: (773) 406-2464
=====================
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at (773)
406-2464 or email.

On Mon, Jun 8, 2015 at 6:59 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi Luis,

Thanks very much for your response. So you would suggest:

1) doing realignment in SPM on the non-normalized images, to get the rp.txt motion parameter file

2) doing normalization in FSL. (question: would this be done on the realigned images from SPM or the original?)

3) doing a first-level analysis in SPM on the normalized (from FSL) images, with the rp.txt file that I made in step #1 as an additional regressor

Is that right?

A few more questions:

How necessary/standard procedure is it to include the realignment parameters (the rp.txt) as additional regressors?

In the GUI, does SPM do normalization in a linear way (ie linear registration?). 

Thanks,

Joelle

On Mon, Jun 8, 2015 at 12:32 PM, Luis Morís <[log in to unmask]> wrote:

Hi Joelle,

Why not take the non normalized images and do the realignment on them? Then you can normalize the images after realigning them using FSL.

I wouldn't expect the realignment parameters calculated this way to be very accurate. In fact I would expect them to be artificially close as when normalizing you are bringing images into a template space, and therefore realigning them indirectly.

Cheers,

Luis

On Mon, Jun 8, 2015 at 11:39 AM, Joelle Zimmermann <[log in to unmask]> wrote:

Hi - I'm doing a first-level analysis of fMRI data in SPM, and want to use realignment parameters (rp text file in the preprocessing folder) as additional regressors (ie select the rp file as "Multiple regressor").

However, I've already done normalization in FSL (using nonlinear registration).

Can I input the already normalized 4D functional to SPM, and run realignment on this in SPM, to get that rp text file that I can later input as a regressor?

I understand that in SPM, the procedure is usually to do realignment (followed by coregistration, segmentation), and only then normalization.

Thanks,
Joelle