You can try bet with the -R option and see if you get any better results. If the -R is not gonna perform well then you could try some registration based brain extraction. So by using fsl-fast you can affine register the MNI T1 template to your T1 whole head image and then apply this transformation to the MNI brain mask. In that sense you extract your T1 brains using a registration process. This might perform better but you can only try and see if it will work better or not.
I hope this might help