A SEGMENTATION ALGORITHM FOR BRAIN MR IMAGES USING FUZZY MODEL AND LEVEL SETS

This paper presents a novel algorithm based on level set techniques for tissue segmentation of brain magnetic resonance (MR) images. The method initially proposed by Suri is improved by using a new regional term based on the investigation and analysis of its stability. The improved algorithm solves the stability problem associated with the original algorithm resulting in a greatly improved quality in MR image segmentation. The multi-seed initialization is used to minimize the sensitivity of the proposed algorithm to the initial condition, as well as speeds up overall convergence. Both simulated and real MR images experiments demonstrate the feasibility and the effectiveness of the improved algorithm, as evidenced by the successful segmentation for various cerebral tissues (white matter, gray matter, and cerebrospinal fluid) of a variety of modal images (T1-, T2- and PD-weighted MR images). Quantitative evaluations of the segmentation results indicate the good performance of the proposed method.