Sometimes, following the use of the tcknormalise command, an unusual effect may be observed where although the bulk of the streamlines may be aligned correctly with the target volume / space, a subset of streamlines appear to converge very ‘sharply’ toward a particular point in space.
This is caused by the presence of zero-filling in the non-linear warp field image. In some softwares, voxels for which a proper non-linear transformation between the two images cannot be determined will be filled with zero values to indicate an invalid warp at that location. However, tcknormalise will interpret these values as representing a valid warp toward the point [0, 0, 0], such that streamline vertices within those voxels will be erroneously spatially transformed to the point [0, 0, 0] in space - this results in the convergence of many streamlines toward a sort of “singularity point”.
The solution is to use the warpcorrect command, which identifies voxels that contain the warp [0, 0, 0] and replaces them with [NaN, NaN, NaN] (“NaN” = “Not a Number”). This causes tcknormalise to discard those streamline points; this is then consistent with the results of registration, where appropriate non-linear transformation of these points could not be determined.