Hi Vincent - What type of population is your longitudinal data from? Do you expect a big change between time points?
Also, for the subjects that have issues, did you try the cross-sectional tractography on them? If you point me to the subject where no tracts were reconstructed, I can take a look.
I'd expect the 2-stick model to work reasonably well in your data.
a.y
On Fri, 26 Apr 2013, vbrunsch@nmr.mgh.harvard.edu wrote:
Thank you so much for your help! :)
- Field inhomogeneity: great, I'll stick with the default option then
- Number of sticks: thank you..I had a look at Figures 8 and 9 in the
mentioned paper..and this does not look good: we have 60 non b0 images with a b-value of 700 and the dwi_snr text file in the dmri folder states that our SNRs are around 5 or even lower..if I understand the figures correctly, this means that we are far from recovering 2 anisotropic compartments, not to mention more. :( Do you think that tracula is then even a good option for our data or do you suggest we just have a look at the output of dtifit?
- Yes, I saw that you increased the number of iterations in the
longitudinal example file from 5000 to 7500, but maybe it should be even more?
I ran through all of the three trac-all steps with our data (nsample = 7500, sticks = 2) and I think I have a similar problem as was already mentioned in the list today: almost for every subject I have at least one out of the 18 paths that was not reconstructed and it seems to be random which path that is. For one subject no path was reconstructed at all. Maybe this is all due to the low b-value and SNR? Or should I just rerun the analysis?
Best, Vincent
Hi Vincent:
- Field inhomogeneity correction: This is done using the FSL fugue
utility, which uses a field map to mitigate the geometric distortions in the DWI due to field inhomogeneity. However, as you point out, DWI-to-T1 registration will be performed anyway, and this may mitigate this type of geometric distortions.
- Number of sticks: How many you can fit to your data depends on the
angular resolution in your data. You can keep adding parameters to the model, but if there's not enough data to estimate those parameters reliably, there's no point. You can get an idea from Behrens et al 2007 on what's feasible.
- The one MCMC parameter that may be useful to play with is nsample, the
total number of path samples collected. Particularly for longitudinal, you may want to increase it if the tracts look narrow, since it may take longer to sample the full tract with data from more than one time point.
Hope this helps, a.y
On Wed, 17 Apr 2013, vbrunsch@nmr.mgh.harvard.edu wrote:
Hi all! I am almost ready to do my first longitudinal run with Tracula. I guess that for the remaining insecurities it is probably best to stick with the default options but I'd like to be more confident with it and just wanted to ask which option to choose and why or why not:
My config file contains information about two time points of 39 subjects.
- Perform "Registration-based B0-inhomogeneity compensation"
If I understand correctly, although I would have the fieldmap information this step (as well as BET) is not necessary as I can use the respective T1 weighted image (usemaskanat=1) for registration?
- Number of sticks in the ball-stick model
Since last month it is possible to change the default value of 2. Would you recommend a higher value? That would give me more information also about crossing fibers, right? But if so, which value would be reasonable and why?
- MCMC burn-in iterations (default:200), iterations (default:7500) and
frequency (default:5) (Why are we allowed to change these options at all?)
Thank you! Best, Vincent _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer