Automatic segmentation of hippocampal substructures
G Gerardo Santiago Flores · 2012
Segmentation of brain structures is an important component in the field of medical imaging because it provides support for diagnosis and treatment as well as therapy evaluation guidance. Modern technology allows acquisition of MR images with very high resolution up to 0.3 mm of spacing between voxels. In this context, we propose a fully automatic segmentation method focused on substructures of the Hippocampal Formation. For this purpose, we present the development and implementation of three relevant contributions. First, we introduce a fast registration method based on moments. Second, we present a multi-level initialization scheme for obtaining initial contours required for segmentation; this procedure is fully automatic and it requires a training set of images with manual annotations of the substructures of interest. Third, we introduce a new segmentation algorithm, which is based on active contours driven by moments prior. We minimize an energy cost function in order to get optimal segmentation employing signatures based on moments. We compared our results to state-of-the-art tools and show significantly improvement in time performance. In addition, we tested our method with patients with Alzheimer Disease.