Hippocampal segmentation by Random Forest classification

This paper presents an innovative approach for the hippocampal segmentation in magnetic resonance images (MRI). The core of the method consists of a Random Forest classifier, it is able to recognize hippocampal and not-hippocampal voxels on the basis of a very large number of discriminating features. Among them, in particular, the so called Haar-like features play a central role. This method is fully automated and it uses, as base of knowledge, a set of hippocampal images segmented by expert neuroradiologists. Such a system can be of paramount importance to assist the neuroradiologist in clinical diagnosis of probable Alzheimer's Disease.