In this paper, we present how 3D split and merge segmentation using topological and geometrical structuring with an Oriented Boundary Graph may be optimized by parallel algorithms. This structuring allows to implement efficiently split and merge operations, but since these treatments have often to be applied with large images, we have studied how to improve performances by parallelizing this process. After a short description of the structuring model and its construction, we describe algorithms for parallelizing the construction of the structuring and describe how this model can be maintained while using parallel processes. We explained the way of partitioning data for use with multiprocessor systems, and extension for use with NUMA architectures and graphics processing units (GPU) will be described. Examples on two medical images of different sizes will be presented and execution time will be given.