A skull stripping from brain MRI using adaptive iterative thresholding and mathematical morphology

Abstract

Skull striping is a crucial pre-processing step incorporated in several brain image processing applications. It deals with the removal of non-brain tissues from the brain Magnetic Resonance Imaging (MRI). The skull striping of brain MRI is not a trivial task due to the complex structure of the brain and presence of intensity in homogeneity artifact in MRI. In this paper a novel approach of skull stripping has been presented. The method is composed of adaptive iterative thresholding in addition to Otsu's global thresholding. The global thresholding is followed by analysis and removal of connected components. Finally morphological operations are carried out to obtain the brain mask. The method has been validated using 20 T1w normal coronal brain MRI images of Internet Brain Segmentation Repository (IBSR) dataset, 40 T1w MRI scans of LONI Probabilistic Brain Atlas project (LPBA40) dataset and 77 T1w MRI images from Open Access Series of Imaging Studies (OASIS) dataset. The comparative analysis using standard metrices (such as Dice Similarity Coefficient (DSC), Jacard Index (JI), sensitivity, and specificity) shows that the proposed method performs better than existing skull striping methods for brain MRI.