I haven’t made the comparison. VMTK is built around meshes, the goal of SGEXT is to take a spatial graph representation of the object (from an image) and work with it as a quantitative analytical tool.
It uses a thinning algorithm, and this one is LGPL from DGtal ( I contributed it from recent research of Digital Topology). Thinning is at the core of SGEXT, but it’s the only part that uses DGtal.
SGEXT includes modules doing reconstructions of graphs, graph comparisons, analysis of branching points (generations), even simulations with pair forces between nodes.
Up to you to decide if it is well suited for Slicer, but its integration won’t take long at all, the only extra dependency would be DGtal.
There are functions to convert the spatial graph to a vtu representation already, and to voxelize the graph into a label image. The library is 95% wrapped in python.
EDIT: The other dependency, which I guess is not that light is Boost, with components: graph, serialization and filesystem.