Effects of Anatomical Asymmetry in Spatial Priors on Model-Based Segmentation of the Brain MRI: A Validation Study

This paper examines the effect of bilateral anatomical asymmetry of spatial priors on the final tissue classification based on maximum-likelihood (ML) estimates of model parameters, in a model-based intensity driven brain tissue segmentation algorithm from (possibly multispectral) MR images. The asymmetry inherent in the spatial priors is enforced on the segmentation routine by laterally flipping the priors during the initialization stage. The influence of asymmetry on the final classification is examined by making the priors subject-specific using non-rigid warping, by reducing the strength of the prior information, and by a combination of both. Our results, both qualitative and quantitative, indicate that reducing the prior strength alone does not have any significant impact on the segmentation performance, but when used in conjunction with the subject-specific priors, helps to remove the misclassifications due to the influence of the asymmetric priors.