A digital image may contain objects that can be made up of multiple regions concerning different material properties, physical or chemical attributes. Thus, segmented simplicial meshes with non-manifold boundaries are generated to represent the partitioned regions. We focus on repairing non-manifold boundaries. Current methods modify the topology, geometry or both, using their own data structures. The problem of modifying the topology is that if the mesh has to be post-processed, for instance with the Delaunay refinement, the mesh becomes unsuitable. In this paper, we propose alternatives to repair non-manifold boundaries of segmented simplicial meshes, among them is the Delaunay based one, we use common data structures and only consider 2 and 3 dimensions. We developed algorithms for this purpose, composed of the following tools: relabeling, point insertion and simulated annealing. These algorithms are applied depending on the targeted contexts, if we want to speed the process, keep as possible the original segmented mesh or keep the number of elements in the mesh.