At the moment we have separate implementations for tensor tractography, machine learning tractography, ...
This could be implemented more generic, since only mechanism to decide in which direction to go is different.