I published something about this a bit ago:
[
www.ncbi.nlm.nih.gov]
Essentially what we do is:
1) Trim to 0-100 HU.
2) Smooth the image a bit (1mm^3 gaussian) - not always necessary, but good idea depending on the convolution filter used on the CT (aka bone scans)
3) Run Brain extraction tool (BET) with fractional intensity threshold to 0.01 - 3dSkullStrip implements a modified version of this (I believe this is shrink_fac in 3dSkullStrip)
4) Fill some holes
Here's example FSL bash code:
[
github.com]