@phcerdan Update: I just learned that it appears better not to call a Factorial function, when calculating the binomial coefficient, “n over m”, as I did before to calculate the number of offsets N needed for an N-connected neigborhood (TopologicalConnectivityImageNeighborhoodShape). My colleague Baldur van Lew suggested me to calculate the binomial coefficient as described by Walter (June 23 2017), at:
When using this clever binom function, the value N of an N-connected neighborhood can be calculated for images that have far more than 20 dimensions!!! Without integer overflow! 
So although it might be nice to make the existing itk::GeometryUtilities::Factorial function constexpr, it’s certainly not necessary for the implementation this shape class.
By the way, the hyperrectangular shape class is still “under review”, it has its first “+1” from @dzenanz: http://review.source.kitware.com/#/c/23389/ See also: A shape class for ShapedImageNeighborhoodRange and ShapedNeighborhoodIterator As you know, I would like the hyperrectangular shape class to be accepted before proposing an N-connected shape class.