A Level Set Method Using Fuzzy Logic and Region Information for Infant Brain MRI

摘要

This paper proposes an modified region-based active contour model within a variational level set function for segmentation and bias field estimation of infant brain magnetic resonance (MR) images. The proposed model utilizes the property that MR images can be divided into two multiplicative parts, namely, the true image constituted by physical tissues and the bias field which accounts for intensity inhomogeneity. The tissue segmentation and intensity inhomogeneity correction are simultaneously accomplished by the process of energy function minimization. In this paper, we define an energy function in a combination of global and local intensity fitting terms. In addition, the energy function in our model is convex by introducing fuzzy logic, which contributes to the robustness of the energy function minimization. Experimental results on both synthetic images and real infant brain MR images show that the proposed method is advantageous in segmentation accuracy bias field correction and robustness to initialization and parameter.