Hi all,
I’m using FSL for T1 images, and I noticed that with BET (brain extraction), some tissue from the frontal lobe is also being removed. I’ve tried multiple times, and the same issue occurs with most of the data. How can I fix this, please?
thanks !
Hi @Oumayma, can you post screenshots of one of your problematic input images, and the brain extraction results?
Thanks @paulmccarthy ! yeah sure; attached catastrophic examples too much tissue deleted
Example 1
this image just after doing orientation and robust
then after bet i get this one
Example 2
orient and robustfov:
I also noticed that too much tissue is deleted in the patients, while only a little tissue is deleted in the controls.
Hi @Oumayma,
Did you preprocess the T1 images at all before BET? I imagine steps such as image intensity normalization would help the low intensity parts of the image being interpreted as non-brain.
Also, if you have your data in BIDS, as suggested by the comments in the code, you might just want to use something like sMRIPrep to preprocess your anatomical images.
Best,
Steven
hi @Steven
I have done orientation and robust FOV correction before applying BET,I did normalization after BET using the MNI 152 template
You recommend normalization before BET?. With which template should I perform this? can u clarify please
I used robust FOV for more precision, Am I wrong in my approach?
yes my Data in BIDS
Hi @Oumayma,
I was referring to intensity normalization, not spatial normalization which you are describing. Something like mri_nu_correct from FreeSurfer. SMRIprep would automate all of this.
Best,
Steven
Oh, sorry, I was confused. Okay, I got it!
thank you
I agree you should consider intensity normalization for homogeneity. It would also be nice to know the exact command you are running. In other words, is the robust FOV option you refer to -Z or is is -R. I generally have pretty good results with the latter.
You could also drag and drop them on brainchop and see if the Extract the brain model works for your data (though note that CSF in the ventricles is also considered non-brain). The Subcortical+GWM model will also segment the CSF, and can provide a good mask to be used with fslmaths.
MANAGED BY INCF
