OK, there’s a lot to discuss here, I’ll try to keep it concise…
Why not just acquire 64 unique directions? 32 directions limits you to lmax=6, when you could go to lmax=8 with 64 - might be a good idea, especially given you’re operating at b=3000, this might start to make a difference. I’m guessing this is a Philips system (given your use of the term Halfscan)? There’s nothing to stop you from inputting your own directions - although I’m told it’s a pretty tedious process…
Also, beware the overplus option on Phillips scanners: do not ever turn that on. It doesn’t look like you have thankfully, otherwise I’d expect your response to look terrible - but it looks fine.
And in general, the tracrography looks sensible, so I don’t think changing the acquisition will solve any of the issues you’re asking about. It’s just that if you have the opportunity to change things, I would recommend going to a true 64 direction acquisition.
OK, from this point on, what you’re asking is dependent on exactly what it is you want to do. I can see you want to delineate the Aslant tract, but what do you want to do with the result? Do you need to extract quantitative information about it, in particular some measure of its ‘connectivity’? This makes quite a big difference to the advice you might get…
So first off: why focus on the Aslant tract specifically? As I understand it (and I may very well be wrong), there’s actually quite a bit of controversy as to whether it even exists… So it seems to me to be very much the wrong tract to practice on and optimise your analysis pipeline. Also, I remember Maxime Descoteaux showing results from their recent ‘Tractometer’ analysis: the FiberFox phantom they used for this was constructed with a limited subset of known in vivo tracts, not including the Aslant tract, yet just about every tracking algorithm tested produced a very convincing depiction of it…
All I’m saying is the evidence for its existence is in my opinion pretty weak, if you consider the diffusion MRI literature only. There may well be convincing independent evidence for it, I’m just not aware of it - but then I haven’t looked for it either, so please don’t take my word for it…
But assuming this isn’t an issue, you also need to be aware of the glaring differences between deterministic DTI and probabilitic CSD tractography: with deterministic approaches, you get the same result for the same seed. In fact, it’s even more restrictive than that: you get the same result for any seed along that streamline. So if you do whole-brain tracking, there are a finite number of possible unique streamlines you can get.
When you go multi-fibre (with CSD or other higher-order models), you now have the potential to choose from multiple directions at each step along the streamline. So the number of possibilities increases substantially. If on top of that you introduce probabilistic propagation, then no two streamlines will ever come out the same. You can keep generating streamlines, and observe new unique trajectories being produced. @rsmith may want to comment here, but if I recall correctly, using a simple Haussdorf-based clustering approach, new clusters (i.e. with at least 5 members) not yet seen in the output were still being produced at the 300,000,000 streamlines mark (or maybe he made it to a billion, I can’t remember). The point being, there’s a lot of possibilities in the data if you allow for the uncertainty you know is there.
So to characterise that uncertainty requires a lot more samples. Note this isn’t unique to tractography: any probabilistic sampling approach needs to produce a big enough sample for inferences to be stable. And the number of samples needed will grow with the complexity of the probability space you’re trying to characterise. It’s just that in tractography, the probability space is ridiculously/wonderfully complex…
So to come back to your original question:
If all you’re after is a clean delineation of the tract of interest, there is no substitute for using well-informed and anatomically justifiable regions of interest. This might consist of a seed ROI in the middle of the tract, an initial direction of tracking to avoid wasting time on unrelated structures, one or two inclusion ROIs near the expected end-points, and maybe some exclusion ROI to remove clearly spurious streamlines if they become a problem. This is basically the classical tract-editing technique, and it requires imposing a lot of prior anatomical knowledge to help constrain the algorithm.
If you want to perform quantification on these results, it depends what kind. If you’re after the mean FA within the tract or some other scalar measure, then the above will probably be fine. If on the other hand you want to quantify some measure of connectivity, then you can’t just do the above and expect the streamline count, or the proportion of streamlines to seeds, or any other metric related to streamline count to be meaningful (see this article on the topic) - unless you process your data differently. In your case, the most appropriate would probably be the SIFT2 technique - @rsmith mentioned this in another post recently. Essentially this would allow you to combine a whole-brain tractogram with a more focused, dense, tract-specific tractogram, and analyse them both jointly to derive a measure of connectivity that also matches the dMRI data over the whole brain - see the SIFT2 paper for details.
Finally, onto the specific questions that came up:
The problem here is that very few of the streamlines generated will end up meeting the criteria for inclusion in your Aslant tract. Increasing the size of the ROI doesn’t necessarily help, if the extra volume isn’t part of the Aslant tract to begin with - you’re just wasting time tracking unrelated structures. As I mentioned above, you can help the algorithm out by providing an initial direction of tracking, which would help increase the proportion of ‘successful’ streamlines. But really, looking at numbers like these is simply not all that informative, for the reason mentioned above (as per Derek Jones’ aforementioned article on streamline counts).
There is no such thing as ‘the correct number of streamlines’. This is a bit like asking how long is a piece of string… If you generate more streamlines, you will get more streamlines. If you use a smaller seed ROI that is located deeper in the core of the tract you’re interested in, you will most likely get a higher proportion of successful streamlines. It all depends on so many factors that it’s essentially meaningless. The only thing that really matters is that the streamlines sample the tract(s) you’re interested in with sufficient density.
If you use something like SIFT/SIFT2 on the other hand, your numbers will be meaningful, but only in relation to the total number generated over the whole brain. But I can’t comment on what the right ‘percentage’ of streamlines might be in any given tract.
I have no idea about this, but I would be surprised if the TrackConverter was somehow messing with the data without any warning. There may be issues if converting the full 500,000 (although I doubt it), but if you’re converting just those streamlines you’ve selected, I can’t see that being an issue. But then I’ve no experience with either TrackConverter or TrackViz, so I can’t really say much more than that. Other users may have something sensible to say here.