Robust Brain Extraction Across Datasets and Comparison With Publicly Available Methods
The skull stripping method is an important area of study in brain image processing applications. It acts as preliminary step in numerous medical applications as it increases speed and accuracy of diagnosis in manifold. It removes noncerebral tissues like skull, scalp, and dura from brain images. In this regard, a simple skull stripping algorithm, termed as S3, is proposed in this paper, which is based on brain anatomy and image intensity characteristics. The proposed S3 method is unsupervised and knowledge based. It uses adaptive intensity thresholding followed by morphological operations, for increased robustness, on brain magnetic resonance (MR) images. The threshold value is adaptively calculated based on the knowledge of intensity distribution in brain MR images. Experimental results, both qualitative and quantitative, are reported on a set of synthetic and real brain MR T1-weighted images. The performance of the proposed S3 algorithm is compared with that of three popular methods, namely, brain extraction tool (BET), brain surface extractor (BSE), and robust brain extraction (ROBEX) using standard validity indices. Keywords—Magnetic resonance imaging, skull stripping, thresholding, mathematical morphology.