Page 1 of 1


Posted: Mon May 18, 2009 5:15 pm
by PeterK
I wonder if there are any rules of thumb to optimize the bias correction parameters for a particular pattern of image contrast/artifacts?
Specifically I'm interested on how Entropy, Regularization and Temperature should be optimized for volumetric (e.g. a central bias) versus phasearray (e.g. peripheral bias) MR coils.

So far I resorted to a crude, brut-force permutation measurements where the optimal parameters are identified by permuting along the three axis using the histogram separations and the amount of between-the-humps voxels are the cost function. This obviously requires hundreds of runs but showed good results. I wonder if there are more intelligent ways of doing it?

Re: VipBiasCorrection

Posted: Mon May 18, 2009 8:34 pm
by Jean-Francois Mangin
I did not do any systematic study like this, and there are so many options to this command,
I am afraid I do not have any systematic idea of what is the best. Like for a lot of other commands
I did, one day they become alive...

The only thing I did carefully is for T1 weighted images. Unfortunately, I noticed during last years
that parallel coils can lead to some problems. The bias model is not sufficient because
large contrast variations sometime occur.

Re: VipBiasCorrection

Posted: Tue May 19, 2009 7:30 pm
by PeterK

I have to say that even after doing two or three times already I still can't meaningfully generalize on how these parameters affect the bias correction. However, permuting across the range of them and selecting the best setting does produce much nicer results than the default parameters, especially for the images from the phase array coils.