It’s not immediately clear to me the nature of the concern. The results between the three images are basically what I’d expect for your experiment.
One way to think about it is that there’s an infinite set of all possible streamlines trajectories, and we only ever reconstruct & visualise a subset of them. If your subset is finite but nevertheless still very large, what you see will be a reasonable depiction of that infinite set. Whereas if your subset is very small, there’s a good chance that there will be areas that are traversed in the hypothetical infinite set that are not traversed in your small subsample, simply because the size of that subsample is not adequate to faithfully represent the total extent of the infinite set.
I think perhaps some of the confusion might arise from saturation of the visualisation. When the streamlines are few and thin, you can see the gaps between the streamlines, and it is these gaps that give a perception of density. However once you have enough streamlines in an area such that there are no longer any pixels that aren’t traversed by at least one streamline, it becomes impossible to judge how many streamlines are actually there. So you may be looking at the 10M image and thinking “this has finished reconstructing those pathways already reconstructed with 200k, and is now traversing other areas; why doesn’t the 200k also traverse those areas?”. The better way to think about it is “the 200k image is just a less dense version of the 10M image; but in some areas, even the 200k image is fully saturated, and therefore I can’t see any difference in those areas even if I increase all the way to 10M”.
One thing you could play with in
tckedit is the
-skip option. So instead of just selecting the first 200k streamlines, you grab streamlines 200,001 - 400,000, 400,001 - 600,000, and so on; and then flick between them in the viewer and see the differences. Those differences aren’t biological; they instead communicate the imprecision in the reconstruction for that particular streamline count.
I am not sure which one is reliable for further analysis.
“Reliable” is unfortunately a bit open to interpretation, but I would pose it this way: If more streamlines = more precision (/ less intrinsic variance), and you have a large tractogram generated, for what reason would you use any fewer streamlines than the full set? The only reasons that come to me are: 1) Visualisation; 2) Actually interrogating the imprecision of the experiment by using multiple independent subsets of the tractogram.