Enhancement of MR Images

摘要: We propose a modification of Wells' et. al. technique for bias field estimation and segmentation MR images. Replacement the class other that includes all tissue not modeled explicitly by Gaussians with small variance uniform probability density, amending EM algorithm appropriately, gives significantly better results. The performance any is affected substantially number selection classes are explicitly, corresponding defining parameters, and, critically, spatial distribution tissues in image. present an initial exploration application minimum entropy to choose automatically associated parameters give best output.