Within the iterative loop of the pre-processing analysis, instead of running the legacy dwi2mask algorithm on bias-field-corrected DWI data, instead take the FOD data that has been corrected for both bias field and unbalanced response functions, apply a threshold equivalent to a 0.5 partial volume fraction, and use that as the brain mask.
The process only repeats twice, as for test data with a strong bias field, further iterations tends to expand the mask anterior & inferior beyond the brain to continually envelop more non-brain voxels amplified by the bias field (and inclusion of those voxels in subsequent iterations exaggerate the bias field estimate even further).
Within the iterative loop of the pre-processing analysis, instead of running the legacy
dwi2mask
algorithm on bias-field-corrected DWI data, instead take the FOD data that has been corrected for both bias field and unbalanced response functions, apply a threshold equivalent to a 0.5 partial volume fraction, and use that as the brain mask.The process only repeats twice, as for test data with a strong bias field, further iterations tends to expand the mask anterior & inferior beyond the brain to continually envelop more non-brain voxels amplified by the bias field (and inclusion of those voxels in subsequent iterations exaggerate the bias field estimate even further).