Hi Sal,
1. I have no experience with the sun grid engine, perhaps someone else on the list can help with that.
2. The -bedp step needs the outputs from the -corr and -masks steps. The -path step needs the outputs from the -bedp step. The output files from each part are listed here: http://www.freesurfer.net/fswiki/trac-all#Outputdirectoriesandfiles
3. Checking the aparc+aseg should be enough for tracula purposes as well, since that's the freesurfer output that tracula uses. Keep in mind that DWI resolution is usually lower than T1 resolution, so tiny changes in the aparc+aseg might not affect tracula at all.
4. Yes, I do recommend using bbregister. It's not surprising that it would give better results since it uses the additional information of the surfaces. If I remember correctly, Doug didn't manage to replicate your bbregister error, right? I'd check if you get the same error with the 5.1 version of bbregister.
Hope this helps, a.y
On Wed, 27 Mar 2013, Salil Soman wrote:
Hi, I have been able to implement FS 5.2 on our Sun Grid Engine Cluster using the CentOS 6 distribution. After modifying the fsl_sub_mgh file to work with out cluster queue names and to modify the email notifications, I am able to run the entire processing pipeline using FLT. (Everytime I try bbr I get errors. I had email doug about this but have not heard any response).
I am submitting the jobs to the cluster as follows
recon-all -i ./*.nii -s WCA_0202_T1_FS -nuintensitycor-3T -nocanorm -openmp 50 -hippo-subfields -all trac-all -prep -no-isrunning -c /mnt/glusterfs/salsoman/SRC/WCA_0202_T1/DTI/dmrirc trac-all -bedp -no-isrunning -c /mnt/glusterfs/salsoman/SRC/WCA_0202_T1/DTI/dmrirc trac-all -path -no-isrunning -c /mnt/glusterfs/salsoman/SRC/WCA_0202_T1/DTI/dmrirc
There are a few issues I am trying to sort out:
- how should one conifigure the openmp flag for the Sun Grid Engine? the flag I used
did not change processing time, and each subject is still taking ~24 hrs?
- once the script starts bedpost, bedpost has controller code which paralelizes all
of the slices (which has been great going from 12 hrs of processing to 1.5 hrs). However, my original script goes on to execute trace-all -path ... before bepost is done, generating errors. I have gotten around this by running the trace-all -path command again after bedpost sends me an email confirming completion. Is there a way to more cleanly do this (some kind of hold command I can do based on output from bedpost)?
- In your experience, how important is quality control of freesurfer on tracula
results? We have come to the compromise for freesurfer output to manually check the segmentation of aparc+aseg.mgz using free view, only correcting and rerunning steps of freesurfer when there are specific structures we want the volumes of (eg hippocampus).
- have you found a significant benefit of bbr over flt? If so, any suggestions on
things I may need to check to get bbr to work (given flt works)?
Best wishes,
Sal