A method of deriving a directional structure from an object dataset is proposed. The object dataset assigns local directions to positions in a multidimensional geometrical space. For example the local directions concern local flow directions in a diffusion tensor magnetic resonance image at least one `region of interest' is selected on the basis of spatial functional information, such as an fMRI image, time correlation of an fMRI image series with a functional paradigm or an anatomical image. These `region of interest' are employed to initialize a fibre tracking to derive the directional structure that represents the nervous fibre system.