see http://dx.plos.org/10.1371/journal.pone.0038234
On 22 Oct 2013, at 18:18, freesurfer-request@nmr.mgh.harvard.edu wrote:
Message: 2 Date: Tue, 22 Oct 2013 16:15:44 +0000 From: "West, John D." jdwest@iupui.edu Subject: [Freesurfer] Subject processing difference between computers To: "freesurfer " freesurfer@nmr.mgh.harvard.edu Message-ID: 793D43FE7976404CB5E172709AB1F4081123FECA@IU-MSSG-MBX102.ads.iu.edu Content-Type: text/plain; charset="us-ascii"
Hello.
We are moving our processing from an older super computer to a new one here at Indiana University. Preliminary testing is showing a significant difference between final summary stats results between subjects processed on the old system versus the new. The version of freesurfer on both systems is identical. Comparing the logs of the same subject run on both systems, I've found that the differences seem to start with the mri_watershed function specifically the main basin size. For the same subject we find:
Old computer: main basin size=22667431830 voxels, voxel volume =1.000 = 22667431830 mmm3 = 22667431.936 cm3
New computer: main basin size=22208239350 voxels, voxel volume =1.000 = 22208239350 mmm3 = 22208239.616 cm3
As you can see they are very different. My theory is this is propagating through the rest of the process and giving us different summary stats at the end. I should note that both aseg and aparc stats are affected.
I am also enclosing the logs for both runs as well as the differences found through the diff command (this is a lot parse through as even time stamp differences are picked up). Essentially what I am trying to find out is if these differences we are noticing are typical when changing computers and if not what could be causing these differences that I may not be picking up on.
Thanks for the help. -John
Thanks Ed. This is consistent with the differences we are seeing as well. The two super computers are very different platforms so it makes sense.
I think we'll have to reprocess all cases for ongoing studies on the newer super computer in this case. Not the answer I was hoping for, but what I was expecting. Thanks again for the information.
From: freesurfer-bounces@nmr.mgh.harvard.edu [mailto:freesurfer-bounces@nmr.mgh.harvard.edu] On Behalf Of Ed Gronenschild Sent: Wednesday, October 23, 2013 9:55 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] Subject processing difference between computers
see http://dx.plos.org/10.1371/journal.pone.0038234
On 22 Oct 2013, at 18:18, freesurfer-request@nmr.mgh.harvard.edumailto:freesurfer-request@nmr.mgh.harvard.edu wrote:
Message: 2
Date: Tue, 22 Oct 2013 16:15:44 +0000
From: "West, John D." <jdwest@iupui.edumailto:jdwest@iupui.edu>
Subject: [Freesurfer] Subject processing difference between computers
To: "freesurfer " <freesurfer@nmr.mgh.harvard.edumailto:freesurfer@nmr.mgh.harvard.edu>
Message-ID:
<793D43FE7976404CB5E172709AB1F4081123FECA@IU-MSSG-MBX102.ads.iu.edumailto:793D43FE7976404CB5E172709AB1F4081123FECA@IU-MSSG-MBX102.ads.iu.edu>
Content-Type: text/plain; charset="us-ascii"
Hello.
We are moving our processing from an older super computer to a new one here at Indiana University. Preliminary testing is showing a significant difference between final summary stats results between subjects processed on the old system versus the new. The version of freesurfer on both systems is identical. Comparing the logs of the same subject run on both systems, I've found that the differences seem to start with the mri_watershed function specifically the main basin size. For the same subject we find:
Old computer:
main basin size=22667431830 voxels, voxel volume =1.000
= 22667431830 mmm3 = 22667431.936 cm3
New computer:
main basin size=22208239350 voxels, voxel volume =1.000
= 22208239350 mmm3 = 22208239.616 cm3
As you can see they are very different. My theory is this is propagating through the rest of the process and giving us different summary stats at the end. I should note that both aseg and aparc stats are affected.
I am also enclosing the logs for both runs as well as the differences found through the diff command (this is a lot parse through as even time stamp differences are picked up). Essentially what I am trying to find out is if these differences we are noticing are typical when changing computers and if not what could be causing these differences that I may not be picking up on.
Thanks for the help.
-John
freesurfer@nmr.mgh.harvard.edu
-
Ed Gronenschild -
West, John D.