Print

Print
FIX (ICA-based denoising) does not take care of truly global artifacts, but as you say removing the global mean is controversial (with some people saying it is essential to do and others saying it is essential not to do).

Peace,

Matt.

From: Jeanette Mumford <[log in to unmask]>
Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
Date: Thursday, July 10, 2014 at 4:02 PM
To: <[log in to unmask]>
Subject: Re: [FSL] preprocessing for residuals

Hi Meredith,

Prewhitening isn't recommended for resting state data, since it can remove variability that is important in resting state connectivity.

Intensity normalization is up to you, this is equivalent to modeling the global mean and is somewhat controversial.  I'll let you make that decision on your own, since preferences vary.  My understanding is if you do something like ICA-based denoising the global mean is not necessary.  If you do not use ICA denoising and don't do much else in terms of modeling out nuisance variability, if you omit the global mean regressor all of your ROI's will end up highly correlated with each other.  

Normalizing the residuals isn't necessary if you're correlating ROI averaged time series later.  Sounds like maybe they were running a seed based correlation analysis with the GLM and if you want the regression parameter estimates to reflect correlations normalization is necessary.

Hope that helps,
Jeanette




On Thu, Jul 10, 2014 at 11:34 AM, Meredith Braskie <[log in to unmask]> wrote:
Dear FSL experts,
I am running a resting state fMRI analysis in which we will regress out the time series from regions of no interest (white matter etc.), and then apply several ROIs of interest to the residuals to get time series information. I found a website (unrelated to you) describing how to do this in FSL, and they suggested turning on intensity normalization and turning off prewhitening to get the residuals. Looking at the descriptions you provide in the gui help, I don't see why this would be the case, but wanted to check with you experts before I proceeded. My TR is 2.1 s. Is there some reason specific to getting residuals for rs-fMRI that I'd want to turn on intensity normalization and turn off prewhitening, or is this likely just a preference of the lab that posted the instructions?

Also, it looks like the lab that posted the website instructions normalizes the residuals after they are obtained. Is this a necessary step before using fslmeants to get the time series information, or does this depend on whether prewhitening is used?

Thank you for your advice & help.
Meredith