Dear Eelke,

thank you very much for your answer! It was very helpful.

Best,

Theodor.


Dear Theodor, 
 
> We have a bunch of DTI-datasets from patients who have undergone functional hemispherectomy. Since we want to run voxelwise analyses, we need to normalize these highly deformed brains. It seems that the normalization of T1 is affected by a field bias (i.e. occipital hyperintensities and frontal hypointensities). We, therefore, want to correct the field bias. There are four quesions we have: 
>  
> question1. When we generate a bias field by means of FAST, it turns out to be affected by the low signal in the lesioned hemisphere. Now we wonder whether we can run FAST with the lesion masked out? 
 
Yes, this should work fine. FAST doesn’t have an explicit option for masking, but it only considers voxels with values greater than zero. To apply a mask you can simply use fslmaths’ -mas option to zero out all the voxels that are inside the lesion mask (you will probably need to invert the mask first). You will not get a bias field for the masked region, but if you are going to use the same lesion mask for registration then that should not be a problem. 
 
 
> question2. Is it possible to create some kind of average field-map out of several unlesioned-brains (who have been scanned with the same sequence and the same scanner) bias-fields (generated by means of FAST) and apply this on the lesioned ones? 
 
Creating a straight average of the estimated bias fields would make sense under the assumption that the bias field doesn't change depending on what is in the scanner. I /think/ this may be a reasonable approximation at lower field strengths but I am really not an expert; it is probably worth checking this with an MR physicist. Alternatively, you could just look at how much the bias fields differ between subjects. 
 
If you decide to do this you will want to extrapolate the bias fields and/or take missing values into account when averaging to make sure that you end up with a map that will cover all your brains. It is probably also a good idea to use subjects that you will not use in your final analysis to create the average. 
 
 
> question3. Should we use segmentation with FAST to remove the CSF before creating the bias field? 
 
Bias field correction and WM/GM/CSF segmentation are an integrated process, so there should be no need to worry about this. 
 
 
> question4. Is there any advantage in using mfast (e.g. with a T2) while creating this bias-field (if the bias field is only noticeable on the T1)? 
 
In theory there could be a small advatage as adding the T2 may improve tissue type segmentation, which in turn would allow for better estimation of the bias field. In practice though, I suspect the advantage will be small unless the T1 is of very low quality and the T2 is of very high quality. 
 
Hope that helps, 
Eelke