I don’t think there is anything wrong with that pipeline, however I don’t think the non-rigid registration will be as accurate as you need for this. My feeling is that the 5TT won’t be well aligned with your diffusion data, and neither the labels from freesurfer. I would use an atlas similar to the AAL for this case and without ACT. But you could try your approach and see if the results are good enough.
I can’t change my atlas , but I think I can perform tckgen without ACT and to see if the result is better than with ACT.
for suggestions on EPI distortion correction you gave before, I think they non-rigid register the DWI to T1 which firstly rigid transforming to DWI, I wonder if this transformation would affect the bvec?
Yes, this is one way to do it. From my understanding, you have to constrain the registration to the phase encoding direction (see here), but I’ve seen some works when they don’t constrain this transformation (or at least, they don’t say it). A version of this approach is implemented in fmriprep and sdcflows.
If I’m correct, the non-rigid transformation from the B0 to the rigidly registered T1w doesn’t affect to the bvecs. I hope this helps.