A case analysis of the impact of prior center of gravity estimation over skull-stripping algorithms in MR images

In this work, we propose a novel approach for improving center of gravity (COG) estimation of the brain in magnetic resonance (MR) images, that uses 3D Haar-like features. We hypothesize that better pose estimation will advance the posterior skull-stripping results of the popular Brain Extraction Tool (BET). The proposed methodology is quantitatively validated in 20 T1- and T2-weighted images of the brain. As compared to the native BET COG algorithm, our method produced COGs 87.3% closer to the expected coordinates for the T1-weighted dataset, and importantly this resulted in an average enhancement of 15.4% to the accuracy of skull-stripping masks. As far the authors know, we are first in analyzing the impact of COG estimation over skull-stripping of MR images.