Brain Extraction Algorithm using 3D Level set and Refinement

Introduction Brain region extraction is to segment brain region from nonbrain regions in magnetic resonance (MR) images. The segmented brain is usually used for measuring individual cortex thickness and brain volume, monitoring the development of the brain, and surface-based analysis. Brain region extraction is, therefore, an important step for various applications above mentioned. We present a brain region extraction method which consists of two parts 3D level set method and refinement process. The 3D level set method is used for estimating coarse boundary surfaces, where its result shows artifacts in some regions such as sudden changes between slices and within a slice. In order to improve an accuracy of the segmentation results, a refinement process is applied to correct the artifacts of the 3D level set method. The proposed method is applied to images in the BrainWeb [1] and IBSR (Internet Brain Segmentation Repository) [2], and images obtained from 1.5T MRI system. In addition, three well-known brain extraction algorithms of brain extraction tool (BET) [3], BrainVisa [4], and FreeSurfer [5], were compared to our method. Methods 1) Coarse segmentation step: We first perform the Otsu thresholding to separate foreground (brain region and skull) and background (residual regions). A normalization process of image intensity is then applied to overall slices. Then, we implement the 3D level set function proposed by Li et al. [6] as follows,