Hi, Maksym--

So, if you use dcm2nii, you should get one file of bvalues and one file of bvectors; if you put them into 1dDW_Grad_o_Mat++, then you can output the TORTOISE-style b-matrix file. The program is described more here:
[afni.nimh.nih.gov]

An example for your case might be:

1dDW_Grad_o_Mat++                         \
   -in_row_vec   ALL.bvec                 \
   -in_bvals     ALL.bval                 \
   -out_col_matT dwi_matT.dat

More generally, note that there are some specific new tools for going from DICOMs to NIFTI, and up to running TORTOISE (and also FreeSurfer, if so interested), and onward to tensor fitting. These wrap around a lot of functions (fat_proc_*), and try to have some convenient features (keep files ~centered around (0,0,0), keep same orientation, make snapshots of data at most steps, etc.). These are described *here*, in documentation that is still growing:
[afni.nimh.nih.gov]
and more specifically for converting dicoms:
[afni.nimh.nih.gov]

You should probably update your AFNI to have as recent a version as possible to make sure you have the most uptodate fat_proc_* functions:

@update.afni.binaries -d

Note that, for the TORTOISE parts there, I was using a beta version 3.1 of TORTOISE; the update should be released quite soon (Okan keeps promising...). But using v3.0 and v3.1 are pretty similar.

Feel free to ping back with any questoins on usage/description that isn't clear-- as I said, the docs are still being full built and ironed out. Feedbacks be useful.

--pt