Problems with FSL BET on 7T data due to B1+ inhomogeneity

[jaystock]

Are there still plans to use BET as part of our processing pipeline?

I have had a lot of problems running FSL BET on scout images acquired at 7T. BET usually "misses" some parts of the brain. If I make the 'f' variable too small, then the mask can extend outside the brain. This problem occurs due to the severe B1+ inhomogeneity over the brain at 7T. I have never had the problem at 3T. For the Scout image, I typically use the PD-weighted magnitude image that accompanies the phase image from the Siemens B0 field mapping sequence. These images show significant contrast variation over the FOV, which confuses BET.

Maybe it is worth comparing several brain masking tools to see which is most robust to these effects. Or alternatively, using a scout image with more contrast rather than the PD-weighted magnitude image that comes with the B0 field map sequence.

I also have problems with BET running on axial oblique slices. It seems to prefer pure-axial slices.