FSL’s FILM-Prewhitening on preprocessed data

Software Support
, ,
1

Hi all,

I am running an fMRI analysis using fMRIPrep for the preprocessing and using FSL for the actual GLM analysis. I’m aware of the differences in MNI Standard Spaces used and I’m using this workaround, which works pretty fine for me.

Now, after looking into FSL’s FILM-Prewhitening algorithm to see if I could run into any issues with my data, I found a short description of it in this paper, stating that as a part of this process, spatial smoothing of autocorrelation estimates is used to reduce bias and the autocorrelation coefficients are smoothed only within matter type since they were found to vary considerably between matter types.

Now here’s my problem/question:
How does the FILM-Prewhitening algorithm get information about precise matter type locations in my data? All I’m giving FSL to deal with is a 4D bold image. Is it just looking at single voxel’s average value to determine if it’s grey or white matter? Or do I need to provide this information about my data somewhere? Or is it using an atlas? Because if so, I guess I would need to manipulate the exact taste of MNI space of that atlas…

2

Hi @t-debor, I think that statement is referring to the use of SUSAN for spatial smoothing. SUSAN implements “structure-preserving” noise reduction - it effectively smooths the image whilst preserving high-contrast edges. The algorithm is described in detail in the original reference: