Since one possible limitation of topup is reducing the angular resolution of diffusion MRI data, It seems that I have two possibilities to acquire diffusion data with the scan time certain:
(1) one repetition of diffusion data with 128 directions and then acquire multiple b0s with opposite phase encoding directions for estimating the warping field
(2) two repetitions of diffusion data with with 64 diffusion directions and with opposite phase encoding directions,
Option 1 and 2 should both work and both have advantages and
disadvantages. If you do one it means you have better angular sampling, but also means that you cannot use the least-squares reconstruction option for resampling the individual dwis. The least-squares reconstruction is an alternative to traditional interpolation that potentially does a better job at restoring the resolution in areas where your signal has been squeezed together.
Additionally option one means that you have information in your data that might potentially help eddy to to a better job at correcting for eddy current distortions.
The trade off between 1 and 2 is not obvious and depends both on the number of directions you have and on the distortions you have to begin with. The worse your distortions to begin with, the more advantage for option 1. The fewer total directions you have the more advantage for option 2.
For example in the HCP project we had data with very bad
distortions and with several hundreds of directions so in that case option 1 was the best.
I hope that is clear.
Jesper