The trouble with this kind of situation is that being presented only with the endpoint of the analysis (i.e. fewer streamlines than expected) doesn’t provide nearly enough information to diagnose any potential issue: there are so many steps from DWI acquisition to streamline extraction that could conceivably go wrong to lead to such a result that only looking at the endpoint simply can’t distinguish.
What’s needed in such a circumstance is to go in and interrogate the data in further detail. In particular, you need to try to find those streamlines that you personally think should be connecting those two regions, but aren’t.
So, for instance:
Look at the whole-brain tractogram and make sure there isn’t some larger fundamental issue with your data.
Use connectome2tck to extract all streamlines connecting prefrontal cortex to anything, and all streamlines connecting from amygdala to anything, and see if there’s any information that can be drawn from that.
Do a targeted tracking experiment, seeding from each region in turn, but investigating the various mechanisms by which streamlines are terminated and accepted / rejected by using tckgen -info, and also seeing how many streamlines are generated successfully from the seed but don’t reach the corresponding target region.
Those are only basic suggestions, but it’s really up to you to interrogate the data very closely at each step of the pipeline that got you to this end result; and particularly, to try to discover how it is that streamlines that could have been reconstructed within that edge are instead led astray, rather than only looking at their absence. For instance, if using ACT, depending on the tissue segmentation sometimes artificial bottlenecks can be created that greatly reduce streamline count as streamlines instead terminate at the bottleneck location.
