I think perhaps your question would be better expressed as segmenting tracts from a whole-brain tractogram, as opposed to a TDI? A TDI is an image with a particular contrast, and so no longer contains information about the underlying streamlines trajectories; so while it could conceivably be used to derive ROIs that could then be used to perform tractogram segmentation, segmentation of tracts is not really accurately described as being “segmented from the TDI”.
It also sounds as though you are strictly looking for a binary mask corresponding to the spatial extent of a particular bundle, as opposed to the streamlines themselves?
If this is the case, then the “simplest” approach is to extract streamlines of interest from the whole-brain tractogram (application of SIFT is not hugely influential in this context), generate a TDI from the extracted streamlines, and then apply a threshold to produce a binary mask.
If this is not the case, then you may need to re-word the question in order to better communicate precisely what it is that you are trying to achieve.