Is it possible to estimate (unknown) smoothing kernel applied on an MRI image?

I have a bunch of 3D MRI images that were smoothed, but I can’t find information regarding the gaussian kernel that was used to smooth these images. Looking at them, I would guess 4 or 5, but I would like to be sure. Is there a program that can precisely estimate the smoothing that has been applied to an image?
Thank you.

Hi @AntoineB! :wave:

In short: no, there is no way to precisely estimate how much additional smoothing has been applied to MRI data. Given that the data can already be quite smooth (especially if you are dealing with fMRI data), estimating smoothness introduced in post-processing is poorly-defined.

However, that is not to say you cannot try to generate approximate estimates—you just have to be very, very careful about what you are going to use these estimates for! If they’re just for your own knowledge about the data it’s probably fine, but if you’re going to use these for some sort of statistical analysis you need to use a lot of caution. The program I would recommend for generating these approximate estimates is AFNI’s 3dFWHMx, but I would strongly encourage you to read the full documentation linked there with all its caveats and warnings!

Hope that helps!


As far as I know 3dFWHMx will provide an estimate of the effective smoothness of the data. Raw fMRI data has substantial intrinsic smoothness and the application of additional smoothing works in a multiplicative way. So the estimate of 3dFWHMx will not give you an estimate of your smoothing kernel, unless you compare it to the smoothness of the raw data. It seems feasible, but probably much simpler to regenerate the preprocessing and keep track of your fwhm …

1 Like