Hey i used


to plot the 100th volume of a 4D fMRI image, as shown on the image. The signal seems no so good since there are a lot background outside the brain. I obtained this image using fmriprep. just wondering if this would be a problem?
Any help would be greatly appreciated!

下载 (1)

just btw, i also tried denoising by high-pass and low-pass, applying ica-aroma, the result looks weird

Probably it would be a good idea to use a brain mask to mask the brain and get rid of the background noise outside the brain. You can use the NiftiMasker for this. Typically fmriprep outputs a brain mask for every subject. If this is an average image you can create a group average mask using intersect_masks.

Hey thank you LeSasse! i have used the mask and successfully blocked the signal outside the brain.

one thing that bothers me is, before the mask, the signal after performing confounds removal is spread out evenly even outside the brain… IMHO this implies something wrong with my confound regression, but i dont know how true that is

Not totally sure, but I know there can be various scanner artifacts and noise that can appear. I imagine those will go away if you do a first level glm for whatever task contrast you want. I think if you look at any one time point in your scan you would likely see that noise.

What confounds are you using?

hey brito, thanks for the reply! i tried ‘simple’, ‘compcr’ and icaaroma ; basically everything from this website:
nilearn.interfaces.fmriprep.load_confounds_strategy - Nilearn

It might be worth explicitly checking out all the confounds generated by fmriprep here: Outputs of fMRIPrep — fmriprep version documentation

Each of these can have a big effect on your data and there isn’t generally a consensus on the best solution (and it can be dataset dependent). It’s also worth noting that you won’t ever truly remove all noise, so its not surprising that you will see artifacts in your data even after regressing out confounds. Its useful to keep looking at the individual brains for your contrast of interest to see if you see consistent artifacts across subjects or just in some runs, in which case those runs might need to be removed.