Hi Colin,
Have you tried reducing the bias field smoothing (the -l option).
Or increasing the number of bias-field iterations (the -I option).
Or have you tried getting the restored output from FAST and running FAST on it again?
If your segmentation is good then I would think that one or more of these
things should help.
All the best,
Mark
yes, I've tried that.
I'm afraid I've given you the wrong end of the stick. sorry.
the WM/GM does not segment in some regions because the MR gives intensity values that locally are very close. as I see it, any algorthim would be challenged.
so, although I haven't actually checked, I'd be pretty surprised if FAST has segmented GM/WM. quite possibly in principle it cannot. nor fix intensities, estimate bias. It does latter on intensity basis. how else? (well, spatial priors. that's my point).
I have a segmentation as I say. This isn't my data. although it helps me too.
the MR is a PD or MT-sat of macaque at 150um. that's why this b0 thing or whatever is such an issue: it was in there a long time.
the segmentation is manual. by hand. almost perfect. better than any algorithm. even in the worst plane it's 1mm slices.
my question stands:
If I know, for sure, but in a yes or no, binary way where is GM in the volume and where is WM, already
can I use that info to estimate the bias field and fix intensity issues?
the answer is yes, but yes with FAST?
yes I'd imagine. but, with existing software in a simple, not time consuming way?
i.e.: I'd like it done, happy to do it. no time though. Can I do it using something that exists?
Colin
