Brain extraction: A region based histogram analysis strategy

Abstract

Brain extraction is the task of removing non-brain tissues from brain magnetic resonance images. Brain extraction is a preprocessing step in many applications related to the brain image analysis. Accurate extraction of brain tissue is a laborious task. So, automatic extraction of it is a need in many applications. In this study we propose an automatic region based brain extraction method. In this method histogram of each region is independently analyzed and parameters relating to each tissue type is estimated by employing expectation-maximization algorithm. The estimated parameters of each tissue type including its mean and variance are used to determine tissues of interests. In this study tissues of interest are gray matter and white mater. Eventually a connected component analysis leads to select largest connected components of tissues of interest as brain mask. The proposed method is tested on BrainWeb dataset. Jaccard similarity index (J), Dice similarity coefficient (DSC), Sensitivity (Sen), and Specificity (Spec) are used to measure performance of the proposed method. The results are compared to three popular brain extraction methods namely hybrid watershed algorithm (HWA), brain extraction tools (BET), and brain surface extractor (BSE). The proposed method outperforms mentioned popular methods.