i think cbsmat still gives its output by using formulas from the previous versions. Its physical coordinates are simply

(i+0.5) * xVoxelSize
(j+0.5) * yVoxelSize
(k+0.5) * zVoxelSize

I have this opinion from an experiment I conducted. using pynifti, I manually added "1" to the voxels of an empty image (header is same with my first message) and then run

tract -seedfile manuallyaddedfile ...  | cbsmat ...

the voxels i added 1 are
(67,54,33), (47,81,49)

then, cbsmat output has following physical coordinates
(1.26562E02, 1.02188E02, 6.70000E01),  (8.90625E01, 1.52812E02, 9.90000E01)

which are calculated using the mentioned formula, not Nifti header... what do you think?