Hi Stefan,
as you said, you can increase the brain outline estimates by using smaller fractional intensity thresholds. Small errors in the BET outline are not in general a problem.
Using BET before dtifit is not theoretically necessary, you can fit a tensor outside the brain. However, the estimates you get there are totally meaningless and having more voxels to process just adds up on computation time. For dtifit that might not be a problem, but it will be crucial for time-consuming processes, such as bedpostx. Also the brain mask is used to stop curve propagation during probabilistic tractography, so you really need a reasonable brain outline there.
Another reason that you BET is that you need a skull-stripped image to perform registration (e.g. of the diffusion images to T1-w), which may be needed in different FSL pipelines.
Cheers,
Stam
On 28 Jul 2010, at 11:02, Stefan Kreisel wrote:
> Hi there.
> Just a small query: we've been getting inconsistently small masks after BET'ting our b0 images before plugging them into dtifit. Clearly, we don't want too much brain chopped away; but what about too little (including non-brain or non-head for that matter), i.e. when changing the fractional intensity threshold in BET to smaller numbers? Theoretically, why BET the b0s at all - one might get tensors calculated outside of the head, but is that really a problem?
> Cheers
> Stefan
>
|