Anatomy Dependent Multi-context Fuzzy Clustering for Separation of Brain Tissues in MR Images

In a previous work, a local tissue distribution model and multicontext fuzzy clustering (MCFC) method had been proposed to successfully classify 3D T1-weighted MR images into tissues of white matter, gray matter, and cerebral spinal fluid in the condition of intensity inhomogeneities. This paper presents a complementary and improved version of MCFC. Firstly, quantitative analyses are added to validate the soundness of basic assumptions for MCFC. Carefully studies on the experiment results of MCFC on a set of simulated MR data disclose a fact that misclassification rate in a context of MCFC is spatially dependent on the anatomical position of the context; moreover, most of the misclassifications concentrate in regions of brain stem and cerebellum. Such unique distribution pattern of misclassification inspires us to choose different context size at such special anatomical regions. This anatomy-dependent MCFC (adMCFC) has been tested on both simulated and 10 clinical T1weighted images and the experiments results show that adMCFC outperforms MCFC and other related methods.