- centerline splitting and grouping is not bullet-proof. It can fail in very weird situations. It would be good to review the code (and maybe lifting the tree assumption)
- branch clipping should use a connectivity criterion (region growing?) to ensure avoiding islands in strange geometries
- write script for computing eikonal solution on the Voronoi diagram starting from centerlines (it's useful for incrementally adding more detail to the tube description). The code is there in vtkvmtkNonManifoldFastMarching but it initializes from interseceted edge endpoints, not from the actual intersection points. 
- write class with Voronoi saliency measurements (not using centerlines, like sphere ratios etc)
- write selective smoothing filter

- use concrete array types (vtkIntArray, vtkDoubleArray) and call DeepCopy when fetching the array from PointData or CellData: DeepCopy handles eventual type casting.

- turn vmtkenv.py into vmtkenv.py.in, configurable by CMake. 

