Hi Chris,
the valid options to mri_compute_bias are:
-s <sigma> - size of smoothing kernel (default = 4) -n - normalize bias field to have mean=1
it will automatically use the ratio of the T1.mgz/orig.mgz to compute the bias field.
mri_apply_bias doesn't take any options. You probably want to apply it to the orig.mgz, but we really don't have much experience with it.
let us know if it helps cheers, Bruce
On Mon, 23 Nov 2009, Chris Hyatt wrote:
Hi,
In a recent Freesurfer mailing list question, someone asked how they could correct temporal lobe intensity normalization problems for a large number of subjects without having to add control points for every subject (very time consuming).
The recommendation from Bruce Fischl was:
" One thing you can try if the patterns of loss are stereotyped is to take 10 or 20 subjects or so and correct them by adding control points, then use mri_compute_bias to compute the common component of the bias field. You can then apply that to the other subjects in your study with mri_apply_bias. Nick can probably suggest a good way to slot this into the recon-all process"
From, the command line, the usage of mri_compute_bias is: mri_compute_bias [options] <subject 1> <subject 2> ... <output volume> But what are the [options], if any?
Next, the bias field output volume <bias_vol>, in MGH format I presume, is then applied to each subject as follows mri_apply_bias [options] <input_vol> <bias_vol> <output_volume> Again, are there any options here?
Also, what should be the input_vol (e.g., brainmask.mgz, T1.mgz) and output volume name?
Lastly, has Nick suggested a good way to 'slot this into the recon-all process', as yet?
I'm having many of the same temporal lobe intensity normalization problems as other folks and want to avoid adding controls points to hundreds of T1 images.
Thanks for your help.
Chris