So I’ve been running our subjects through the basic preprocessing steps - denoise, degibbs, dwipreproc, and bias field correction - but at the end of this pipeline I’m still getting some bad geometric distortions in the orbitofrontal cortex. This seems to be occuring with all of our subjects, but I can’t seem to figure out what I’m doing wrong in my preprocessing pipeline.
Has anyone else come across this issue, or knows a potential cause? Maybe I’m putting in a parameter wrong for dwipreproc? This is the command I use:
Many working in this field believe that using only b=0 information to estimate the inhomogeneity field is far from ideal: it’s possible in some instances to to estimate a field from such an image pair that results in an almost perfect overlap between the corrected input images, but is entirely unfaithful to the underlying anatomy. Using the full DWI signal profiles would be preferable; but not only are implementations of such methods not widely available, they require b>0 images with variations in phase encoding, which are not as widely acquired as the solitary “reversed phase-encode b=0 pair”.
I can’t personally speak on that particular region, but a couple of things you could at least consider & look for in your data would be: a non-diffeomorphic field (i.e. the signal doesn’t just “compress”, but actually “wraps around” behind itself); and >2pi phase difference across an image voxel, in which case the requirement of equal total spin density in the image pair breaks down (i.e. each image needs to have the same amount of “stuff”, just redistributed). Either of these will lead to inaccurate inhomogeneity field estimates.