Hi Peggy - The reinit option is meant to use when you don't have a
valid-appearing tract the first time (you have a tract that looks like a
single line, which means that the first initialization of the path was
poor, i.e., outside the white matter or something like that). If you have
a valid-appearing tract with the first initialization, but not with
another initialization, then the first one is the one to keep. If you have
a case where you have different results but you are unsure which one is
valid, I can look at it if you upload here:
https://gate.nmr.mgh.harvard.edu/filedrop2/
The nsample is the number of path samples that are saved and added up to
give you the final path distribution. In principle, if you found that you
were not converging reliably to the same solution with this number of
samples (with a valid initialization), you would increase the number.
The number of burn-in samples is a number of samples that are run in the
very beginning and discarded, before the nsample samples that are actually
saved. As long as the algorithm starts from a resonable initial guess,
this shouldn't have much of an effect. The nkeep parameter means that 1
sample path is saved for every nkeep sample paths that are drawn. Not
keeping every sample increases independence between samples. These are
standard aspects of an MCMC algorithm.
a.y
On Mon, 15 Dec 2014, Peggy Skelly wrote:
I've been interested in these 2 questions, since I have similar
issues:https://www.mail-archive.com/[email protected]/msg38213.html
http://www.mail-archive.com/[email protected]/msg39282.html
What does it mean if I find a valid-appearing tract with reinit=0, but then
lose it with reinit=1?
How much variability should I expect in the path stats between runs using the
same parameters?
I haven't gotten as far as the previous question to see if statistical
significance changes, but since I'm looking at longitudinal data, I
expect the actual white matter changes over time to be rather small.
How do the dmrirc parameters nburnin, nsample, and nkeep affect the MCMC
process and results?
Since the built in randomness should decrease the number of samples is
increase, should I just increase nsample?
Thanks,
Peggy
_______________________________________________
Freesurfer mailing list
[email protected]
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
