Unsupervised statistical adaptive segmentation of brain MR images using the MDL principle
Tae-Woo Kim, Chul Hwa Paik · 2002
We present a novel statistical adaptive method using the minimum description length (MDL) principle for unsupervised segmentation of magnetic resonance (MR) images. In the method, random noise is accounted for by modeling tissue regions by a Markov random field (MRF). In order to account for magnetic field inhomogeneities and biological variations of tissues, intensity measurements of local regions defined by windows are modeled by a finite Gaussian mixture. The segmentation algorithm is based on iterative conditional modes (ICM) algorithm. The algorithm approximately finds the maximum a posteriori (MAP) estimation of the segmentation and estimates the model parameters from the local region. The window size for parameter estimation and segmentation is estimated from the image using the MDL principle. The technique shows better results than conventional methods in segmentation of MR images with inhomogeneities.