In tcksift2, different fixels can contribute more of less to the derivation of the cost function, depending on the "confidence" with which the comparison between streamlines density and fibre density can be made. At the time, I came up with a heuristic based on the 5TT tissue segmentation. If you provide that image to tcksift2 using the -act option, it will do the requisite calculations.
The main thing this does is that in voxels with partial volume between WM and GM (as determined by the 5TT image), streamlines may terminate within the voxel and therefore the streamlines density is not entirely trustworthy, and the WM-like fibre density may be decreased with corresponding non-zero GM-like density but depending on the registration to the anatomical image that ratio is not guaranteed to be precisely equivalent to the fraction of the voxel classified as WM by the anatomical image segmentation. This approach kind of "dulls" the influence of these voxels on the optimisation.
In
tcksift2
, different fixels can contribute more of less to the derivation of the cost function, depending on the "confidence" with which the comparison between streamlines density and fibre density can be made. At the time, I came up with a heuristic based on the 5TT tissue segmentation. If you provide that image totcksift2
using the-act
option, it will do the requisite calculations.The main thing this does is that in voxels with partial volume between WM and GM (as determined by the 5TT image), streamlines may terminate within the voxel and therefore the streamlines density is not entirely trustworthy, and the WM-like fibre density may be decreased with corresponding non-zero GM-like density but depending on the registration to the anatomical image that ratio is not guaranteed to be precisely equivalent to the fraction of the voxel classified as WM by the anatomical image segmentation. This approach kind of "dulls" the influence of these voxels on the optimisation.