Closed AndrewZalesky closed 2 years ago
Note that the total number of streamline contributing to the connectivity matrix will be different for each individual.
Is 100K enough, given that only ~20K make it to the matrix? Should we push up to 500K, if time permits?
Numbers of "seeds" vs. "streamlines" can be accessed currently via -info
; if requested I can add a command-line option to tckgen
to write that information to an e.g. JSON file for easier access than parsing stderr
contents.
I don't recall the extent to which we discussed deriving multiple matrices per connectivity metric. On top of exemplar deterministic vs. probabilistic tracking algorithms there's also different seeding strategies etc., and the more variants you include the lower the streamline count will be for each.
100K is commented as being a reduced count for the sake of testing.
Yes, we have the information on the number of seeds and streamlines generated from those as reports from -info.
It should be fairly easy to parse the logs afterward to get these numbers and put them into a CSV report.
And regarding the 100k seeds, it is just a small value to speed up testing and fine-tuning the codes. Once that's final, I'll make sure to increase it when finalizing the execution pipeline on spartan. Ideally, I would generate at least 5 million, but if runtime becomes too long, maybe 1 million?
Given the rough estimate of ~20% seeds turning to actual streamlines, it would result in approximately 1 million streamlines per individual.
I'll keep this issue open to report and document the final seeds used once the codes are final.
MRtrix3/mrtrix3#2413
https://github.com/MRtrix3/mrtrix3/pull/2414
That's the proposed change to the dev
branch; I've cherry-picked that commit onto the ukb
branch for easy access.
https://github.com/sina-mansour/UKB-connectomics/blob/8558c97d973ad6050a92ea79256bea980f8fa6ae/scripts/bash/probabilistic_tractography_native_space.sh#L198
Given that we are using -seeds without -select, it would be informative to report the total number of streamlines contributing to the connectivity matrix relative to the total number of seeding attempts. Ideally, we would want to specify the total number of streamlines to initiate, rather than the total number of seeds, but I think what we have seems to be a good compromise.
Also are we planning to provide a deterministic set of connectivity matrices?