Validation of Alternating Kernel Mixture Method Based Segmentation of the Human Brain
Nayoung A. Lee, Carey E. Priebe, J. Tilak Ratnanather, Michael I. Miller · 2007
This paper describes the application of a novel segmentation method in high resolution MRI subvolumes containing hippocampus in five subjects and occipital lobe in five subjects. The alternating kernel mixture (AKM) algorithm is used to segment the MRI subvolumes into cerebrospinal fluid, gray matter, and white matter. The segmentation is validated by comparison with manual segmentation. The misclassification errors are 0.10-0.17 (n=10). When compared with Bayesian segmentation method, AKM yields smaller errors. By generating multiple mixtures for each tissue compartment, AKM mimics the increasing variance in the manual segmentation in partial volumes between the highly folded tissues. AKM's superior performance makes it useful for automated segmentation of sub-cortical and cortical structures in neuro-imaging studies.