Hi Derek,
You are right, a pdf is computed for each parameter of the model at every voxel. Notice that we use MCMC, so we draw samples from the posterior distribution and these samples for each parameter describe the corresponding marginal posterior.
For the orientations, dyads1, dyads2 etc, describe the modes of the posterior. The individual samples describing the distribution for the spherical angles (theta,phi) are in the merged_th1samples, merged_ph1samples, etc. You will see that these files are 4D, containing 50 volumes, each volume corresponding to a different sample from the posterior.
You cannot plot the orientation samples in fslview for a voxel, but you can easily do that e.g. in matlab, using the merged_thsamples and merged_phsamples.
Finally, if you want to fully understand the model, the inference and the way tractography is ran you need to go through the relevant papers and the FDT manual.
http://www.ncbi.nlm.nih.gov/pubmed/14587019
http://www.ncbi.nlm.nih.gov/pubmed/17070705
http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/fdt
Cheers
Stam
On 26 Jun 2013, at 20:37, Derek Archer wrote:
> Hi All,
>
> I'm a little confused how the output of bedpostx is used as the input for probtrackx. From what I understand, each voxel's probability distribution function is calculated independently. How then, for instance, would I be able to view a PDF for one voxel? I ultimately would like to view a probability vs. angle plot for each voxel. I've tried overlaying the dyads1/dyads2 files (as lines) over my FA map, but I didn't find this to be very informative to what probtrackx would do to choose.
>
> From this, shouldn't the PDF in the corpus callosum voxel have high kurtosis whereas voxels within more complex fiber geometry would have lower kurtosis within the PDF?
>
> Also, how does the sampling in probtrackx work exactly when there are crossing fibers within the voxel? Are the PDFs for each direction summed and the sampling is done from both of these?
>
>
>
> Thanks,
> Derek
>
|