the recon-all.log indicates that the build stamp is:

freesurfer-Linux-centos4_x86_64-stable-v5.0.0-20100808

which appears to be an internal release candidate version of stable5, so its very likely that mri_ca_normalize was rebuilt with an different utils lib prior to the stable-pub-v5.0.0 release. i'm currently re-running bert using the freesurfer-Linux-centos4_x86_64-stable-pub-v5.0.0 build and will repost bert.recon.tar.gz for you to compare against your own run (look for a new bert.recon.tar.gz on our site late tomorrow).

although ultimately the knowledge of having an exact match in results is not useful, relative to the fact that subjects in your study will each need to be inspected for correctness. note also that there will be variability in platform results (eg. Linux vs Mac) and versioning (v4.5 v5.0.0 and the upcoming v5.1.0), so all study subject should be run with the same version.

n.

On Tue, 2011-01-25 at 17:39 -0500, Tossi Ikuta wrote:

Hi everyone,

I'm wondering whether there are multiple versions of recon-all v 1.313.2.6 (2010/08/04 15:50:56), and/or mri_ca_normalize stable5.

I have been testing recon-all and compared our output against the sample processed-data in FS/TUtorial. I found that the output obtained was different from the sample, despite the fact that I ran recon-all with the same options as the sample outputs, on the same input files.

recon-all -s bert -i /analysis/Tests/FSTest_Tossi/Samples2/buckner_data/tutorial_subjs/bert/mri/orig/001.mgz -i /analysis/Tests/FSTest_Tossi/Samples2/buckner_data/tutorial_subjs/bert/mri/orig/002.mgz -i /analysis/Tests/FSTest_Tossi/Samples2/buckner_data/tutorial_subjs/bert/mri/orig/003.mgz -all -debug -clean -norandomness -allowcoredump -time -subjid BertByTossi20110119A

I performed a line-by-line comparison of the recon-all log between the sample and one from my run and found that our log differs from the sample (buckner_data/tutorial_subjs/bert/scripts/recon-all.log) at line 1034, where mri_ca_normalize runs the first path of smoothing (with sigma=0.200). Whereas the sample has 38 recursions, my run had 18 recursions for this section. The outputs hereafter differed from each other.

I used the same version as the sample was processed, recon-all: v 1.313.2.6 (2010/08/04 15:50:56), and mri_ca_normalize: stable5. Not to mention, the input files are identical.

Also, I ran the exact same line of the command more than three times, and obtained the exact same result. Therefore, I believe this is not due to any randomization introduced (if any) in normalization/smoothing. I used an x86_64 processor, the same as the sample. Thus, I do not believe this is due to any floating point resolution difference between architectures, either.

The only possible reason I can think of that might account for the discrepancy is that there are multiple versions or compiles "stable5" mri_ca_normalize.

I obtained the sample input and result from; ftp://surfer.nmr.mgh.harvard.edu/pub/data/bert.recon.tgz Our download md5sum matches the one posted.

Here's a list of input files; $ls -l /analysis/Tests/FSTest_Tossi/Samples2/buckner_data/tutorial_subjs/bert/mri/orig/ total 17776 -rw-rw-r-- 1 tossi mrilab 5996899 2010-08-09 13:01 001.mgz -rw-rw-r-- 1 tossi mrilab 6077242 2010-08-09 13:01 002.mgz -rw-rw-r-- 1 tossi mrilab 6083164 2010-08-09 13:01 003.mgz

Thank you very much in advance.

Tossi

/********************************************* Tossi Ikuta, Ph.D. Research Scientist The Feinstein Institute for Medical Research The Zucker Hillside Hospital (718) 470-4486 tikuta@nshs.edu *********************************************/ _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer