Partial Volume Tissue Segmentation using Grey-Level Gradient.
David C. Williamson, N. A. Thacker, S. R. Williams, Maja Pokrić · 2002
A Bayesian probability based tissue segmentation method is presented, which makes use of the grey level information in the images and also the local grey level slope. The grey level distributions are modelled as a combination of Gaussian distributions and triangle-Gaussian convolutions. The local grey level slope distribution is modelled as a linear combination of Rician distributions. The parameters are fitted and used to provide the information required to constructed a Bayesian tissue classifier. Results presented for a synthetic data set illustrate that the model distributions describe well the distribution of grey levels and local grey level slope in a 2D image. Application of the method to an MR image of a human brain demonstrate how the segmentation method removes commonly occuring artifacts in partial volume probability maps. Summary This work addresses the problem of partial volume estimation, where a mixture of two or more tissues combine to produce the image intensity value for a particular voxel. Bayes theory is used to generate probability maps for each segmented tissue which estimates the most likely tissue volume fraction within each voxel as opposed to previous approaches which attempt to compute how likely it is that a certain grey level would be generated by a particular