Regularizing fieldmap derived from difference between 2 phase images

Following the FUGUE guide, I am trying to apply susceptibility (field inhomogeneity) distortion correction using 2 phase images taken with different echo times.

However, all my fieldmaps end up with small, sudden blobs at various places (mostly ventral) that cause the unwarping algorithm to tear out the brain at those places. Most common regularization methods don’t work on them completely. Example after median filter and smoothing:


What method or combination of methods might work on them? Or are they symptoms of problems with phase/phasediff images?

Or perhaps I shouldn’t worry too much about them? I will be using the images for functional connectivity analyses.

The solution turned out to be unwrapping (FSL PRELUDE) after phase image subtraction instead of before.

Eliminating phase jumps like these are exactly what unwrapping is supposed to do.