I don't know of any difference that would have caused this, though it has been a long time since 5.1. The 0.3 shift is suspicious because that is the difference between a signed test and an unsigned test in log10(p). Can you capture the screen output from the two runs and send them to me?

On 05/26/2016 09:09 AM, Victor Montal Blancafort wrote:

Dear FS experts

We have been playing around with different mri_glmfit versions. I processed all my data with FS v5.1, and then I run the mri_glmfit with both the 5.1 and 5.3 versions, using exactly the same X, same contrasts matrix and obviously the same images. Our concern is that the results differ significantly between both versions. Concretely, the results from the 5.3 mri_glmfit version are shifted 0.3 points with respect from the 5.1 one. More precisely, the results using the 5.1 version at 1.0 look the same as the 5.3 at 1.3 significance level.

Is there any difference in parameter estimation or anything else between both versions? Should the results be the same given that we use the same data, parameters, etc?

Thanks in advance for your replies,

Best regards,

V.

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer