Unfortunately such issues are very difficult to diagnose from just the streamlines data alone. What would help is more information about your data, and some images showing the data from the intermediate steps.
You data kind of look like the result of the “null distribution” algorithms, where no fibre orientation information is used and streamlines just select orientations entirely at random, which is what would happen if (for some reason) your ODFs were spherical with no information beyond l=0.
I also note that you’re using mtnormalise with only a single tissue as input. I don’t know from experience exactly what effect this will have, but it does intuitively defeat the purpose of a “multi-tissue”-based approach. If you have both b=0 and b>0 data you can do a two-tissue WM+CSF deconvolution; don’t know that that’s the source of your reported issue, but it’s worth considering its use anyway.