Topology Correction for Brain Atlas Segmentation Using a Multiscale Algorithm

In medicine and neuroscience, the reconstruction of anatomical structures from brain MRI images is an important goal, especially for regions in the human cerebral cortex. Topological correctness is important because it is an essential prerequisite for brain atlas deformation and surface flattening. We propose a new approach to repair a binary volumetric brain segmentation so that it becomes topologically equivalent to a sphere. A morphological multiscale approach which acts on foreground and background simultaneously divides the segmentation into several connected components, and subsequent region growing guarantees convergence to the correct spherical topology and changes as few voxels as possible. In addition to existing graph-based procedures, this provides an alternate approach which has several advantages, including high speed, ease of operation without graph analysis, and measuring the size of a handle, cutting a handle or filling the corresponding tunnel based on their sizes.