Hi Paul,
I've finally got the script that I talked about ready - the one that calculates a local linear approximation to your non-linear warps. If you give it the warp from structural to standard (not the invwarp result but the original warp) plus a mask in the standard space for the ROI (one hippocampus for you, as it will probably work better to do each one separately) then it will return a FLIRT matrix that is the approximation for the structural to standard space registration. You can check the results by applying this matrix to the structural image (and realise that it is only around the ROI that it will be good) and you can also use this matrix directly as the -a input to run_first_all. I would definitely recommend that you check it the first few times you run it (by applying the matrix to the structural image and looking at the results) to make sure it is working OK.
The script can be found at:
http://users.fmrib.ox.ac.uk/~mark/files/aff_warp_approx.sh
The usage is:
Usage: aff_warp_approx.sh <img2std warp> <ROI mask> <aff matrix output>
All the best,
Mark
P.S. Sorry for the delay - the script needed a bit of tidying/fixing before sending it.
On 18 Jun 2012, at 02:39, Paul wrote:
> Dear Mark
>
> Firstly, thanks for your explanation about the issues of non-linear normalization in FIRST procedure. It really help me to realize FIRST more deeply. Currently, i work with Alzheimer's disease dataset. In my experience, when i perform FIRST analysis in patients with serve hippocampus atrophy, the FIRST can't segment hippocampus very well and cover part of CSF in final segmentation results. I also tried the -3 option with run_first_all command, the results also seems not good enough for me to perform further vertex based analysis. Based on my limited knowledge about FIRST, i think the possible reasons of this finding may come from the registration procedure (linear registration seems not well enough for AD subjects) and shape model (which only include limited numbers of aging/patient population) in current FIRST analysis procedure. Am i right ?
>
> Could you email me this script, i have highly interest to try it in AD dataset and hope to get more reliable segmentation results! Thank you !
>
> Best !
>
|