External Email - Use Caution

Dear Freesurfer Experts

I am trying to analyse tbss data using the FS lme module in Matlab. I converge the data from the tbss skeleton into matlab format and estimate an lme model with FA as the dependent variable. The problem has been that the lme algorithm does not converge very often (25-30%). I have now tried switching from using estimation by the Fisher scoring algorithm to estimation by expectation maximization, and the convergence increased to over 90%. Do you agree with the decision to change to expectation maximization? Are there any guidelines on what algorithm is best suited for DTI- data ? Also: Are there any other tweaks that can be applied to achieve better convergence?

Best regards Knut J Bjuland PhD

On 25 Nov 2018, at 09:21, Knut J Bjuland knutjorgen@outlook.com wrote:

   External Email - Use Caution        

Dear Freesurfer Experts

I used the FreeSurfer univariate lme_fit_FS.m on longitudinal DTI data (FA) from tbss, converted from the skeleton to a matlab matrix so as to be able to use the FreeSurfer lme code. However, the model fit converges in only 20 % of the cases, which is too low for me to be able to use these data.

I would appreciate any suggestion as to how I can rectify this situation. How do I troubleshoot this in order to increase the convergence? Can I adjust the convergence epsilon (gradient's norm) in order to increase the number of iterations, is that possible and would that help?

Best regards Knut J Bjuland PhD

---

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