Imaging techniques involving transillumination require detailed knowledge of radiation path(s) between source and detector. When imaging with near infra-red in tissue this is particularly problematic due to the high scattering cross section. The population of `direct path' photons is so small that information must be gathered from the much larger (but still small) population of `snaked path' photons. Path Integral (PI) models set out to find the most likely of these paths, not by random sampling as in Monte Carlo based techniques, but directly: a cost function (Lagrangian) is constructed based on the physics of the scattering processes/absorption and integrated along the photon path to generate a total cost (Action). This is minimized using variational calculus to extract the most likely path. Whilst the PI approach is not new, the work presented here is novel in constructing the Lagrangian using local path descriptors. This allows explicit inclusion of an absorption term and also lends itself to arbitrary numbers of constraints on intermediate `visit' points, path directions, and overall path length. Scaling symmetries are used to further reduce the computational expense of the method.