On Fri, Apr 6, 2012 at 4:05 AM, Knut J Bjuland knutjbj@hotmail.com wrote:

I used recon-all on two different system one with Ubuntu 11.10 and a cluster with Centos 5.7. When I compare the recon-all of these two recon-all run I see that there is a small difference. Is this difference caused by difference in math libraries, and difference in the way floating point is handle? Is there any way to get around this problem and get Freesurfer to behave is similar way on Centos 5.7 system as it did on the Ubuntu 11.10.

If it's exactly the same version of Freesurfer on both platforms, then a difference in the maths libraries would be a highly likely explanation. And it's possible that the library on one platform has a bug which the other doesn't (or even that each version has a different set of bugs).

However, may I make a suggestion: rather than look for a way to make the two runs identical, why not use the difference as an estimate of the experimental error in a Freesurfer run? If any conclusions you draw are robust against the change in platform, you can have extra confidence in them (and even note that in any paper you write up).

Regards,

Richard