| T O P I C R E V I E W |
| morey |
Posted - Sep 20 2010 : 09:37:25 AM
We have performed subcortical segmentation on the Cluster and also on our linux box (Cohen). Both systems are apparently running FSL 4.1.4. However we are getting different volumes for the same region. For example for subject 1 for L-hippocampus, we got 3276 on the cluster but 3229 on Cohen. For subject 2 we got 3248 on the cluster and 4577 on Cohen. We tested 4 subjects with the L-Hippocampus and L-Amygdala and all the measurements were off by at least on voxel and sometimes off by hundreds of voxels !
Just to make sure we reran the same subjects on Cohen and the Cluster and got the exact same inconsistencies, i.e. same volumes within machines but the same inconsistency between machines.
Also a possibly unrelated issue was on a separate analysis, when we tried to concat bvars from Cluster and Cohen we got an error saying that the models were different.
Any thoughts ? |
| 4 L A T E S T R E P L I E S (Newest First) |
| syam.gadde |
Posted - Sep 20 2010 : 11:22:43 AM
In that case, are there log files that you can compare between the two? Are they running essentially the same commands? Or, maybe, despite the fact you are running the 4.1.4 versions of the top-level script, it is somehow finding other versions? |
| morey |
Posted - Sep 20 2010 : 10:27:54 AM
There are no design.fsf files because we have not used FEAT. This is simply the volumes of regions that are different from running FIRST.
The run_first_all on the cluster is bing run by call it from the directory from 4.1.4. $FSLDIR/bin/run_first_all where $FSLFIR is set to /usr/local/packages/fsl4.1.4/bin |
| petty |
Posted - Sep 20 2010 : 09:42:38 AM
The Cluster's default version of fsl is 4.1.5 ... so unless you specifically selected the older version and called the tools directly you were not running the same version of FSL.
Your last comment specifically makes me think you ran different versions.
|
| syam.gadde |
Posted - Sep 20 2010 : 09:42:18 AM
The last issue points to a potential red flag -- could the pre-processing have been done differently on both machines? You can find the design.fsf files for both the cohen and cluster .feat directories and diff them to see what the differences are:
diff -u design-cluster.fsf design-cohen.fsf
You might expect the filenames stored in the .fsf file to be different, but differences in other fields might be worth further investigation. |
