FSL Smoothing: SUSAN vs. fslmaths -s

Hi all,

I ran FMRIPREP on my data, and so was expecting to do smoothing on my preproc bold files before moving forward with FSL first level modelling. I’m aware that a smoothing option is run during FMRIPREP if ICA-AROMA is enabled, but the smoothing parameters aren’t clear to me, and I’d like to have some flexibility at this step anyhow.

I wrote some batch scripts (high-performance computing) that run smoothing on my huge dataset in about 10 minutes, using fslmaths.

It’s something like this:
fslmaths <input> <output> -s 7,
the goal of which is to smooth with a 7mm kernel.

I got worried this wasn’t the correct solution, and recently heard about the Susan tool, which I don’t really understand how it’s different. Apparently the smoothing is designed to retain the underlying structure of the fMRI data?

My question is, what would a similar implementation of Susan be in command line that I can still run on my data with a 7mm kernel? And, do I need to do anything special because I’m using a pediatric MNI template for registration etc.?



You can read more about it at this article: S.M. Smith and J.M. Brady. SUSAN - a new approach to low level image processing. International Journal of Computer Vision, 23(1):45–78, May 1997.

Basically, if you are smoothing at a particular area, and there is a voxel with a much different intensity, it is ignored. This is expected to happen at different tissue classes (e.g. WM vs GM vs CSF), so smoothing should only happen within tissue classes.

You can use the susan command in FSL, specifying the spatial extent with the dt input (see susan -h for the documentation).

You should choose an fMRIPrep output space that is appropriate for your age range, such as the MNIPediatric template. If you already did that, you should be ready to go!