Hi,
I have quantitative T1-data saved as Nifti 16bit integer with scale=1 and intercept=0, intensity range 0-10000.
I convert them to hdr/img and can read them fine into MRIcro or MRIcron.
Now I want to apply brain extraction with BET.
 
It works perfectly as expected when I call BET via the menu entry in MRIcro "Etc" -> Skull strip image ... with fi=0.2
 
But I would like to use the standalone version (bet.exe) via a script. However using bet.exe as
 
bet

-f 0.2
 
produces only 8bit output images with a weird scale. The overall input intensity range 0-10000
is reduced to 256 (obviously 8bit) but this does not suffice since I lose the entire dynamic range
no matter what scaling is applied.
I have tried all sorts of settings using 32bit integer or float inputs etc. nothing works.
When I use BET via menu from MRIcro it preserves the same output data type as the input type
(16 bit integer, 32 bit float whatever).
 
I also tried the new newest version of MRIcro and MRIcron with the accompanying bet.exe
but all the same behaviour.
 
Is there any parameter that forces to maintain the input data type?
 
Any help would be appreciated.

Hello-

I assume you are using my software on a Windows computer. This uses a special version of BET that we compiled for the Windows operating system. At the time, there were different versions of BET for different image precisions, so my software essentially creates a temporary image with a specific precision to generate a mask. With MRIcron you can save the final masked image using the File/SaveAsNIfTI command. This does mean that the BET software will not be particularly useful from the command line: it works in concert with MRIcron to set an appropriate brightness and mask. For using BET from the command line I would strongly suggest using a modern implementation of BET using an OSX or Unix computer. One solution would be to set up a neurodebian virtual box. I think you can set up a shared folder so BET will be able to process an image in a Windows folder. This will give you a much more modern variant of BET (the version we used did not have the terrific "robust" -R option). The modern versions of BET do preserve data precision.