Nonparametric density gradient estimation for segmentation of cerebral MRI

The segmentation of cerebral MRI is approached as a classification problem where the density function is unknown. For MR images, the modes of the intensity distribution are related to cluster centers obtained from anatomical structures of brain images. The modes of the unknown density function can be calculated by applying the mean-shift method. This nonparametric technique allows an analysis which only depends on a specific bandwidth. The mean-shift approach was applied to brain MR images to obtain clusters of white and gray matters and cerebrospinal fluid. Segmented images with this, robust scheme show a higher index of similarity when compared against manually traced structures and its performance is superior to other segmentation procedures.