Brain Extraction from Magnetic Resonance Imaging Using Deep Neural Network with Edge Prior

Brain extraction from Magnetic resonance imaging (MRI) is the separation of brain tissue from non-brain tissue in MRI brain images and removal of extra brain tissue. Automatic extraction of brain tissue is an important step in neuroimage processing for clinical applications and scientific research. Accurate delineation of the region can be challenged due to the weak contrast between the brain and non-brain tissue. To overcome this issue, this paper proposed a novel brain extraction approach integrating the boundary information under the deep neural network framework. The boundary of the brain tissues was first detected by an edge detection neural network. The inception module was then utilized in the convolutional stage to capture multi-scale features. A dense decoder shortcut connection structure was finally introduced to leverage the multi-level features in the upsampling stage. The proposed method can achieve Dice overlap coefficients of 98.15% and 98.09%, for the OASIS and LPBA40 datasets respectively. The proposed method outperforms the commonly used brain extraction tools by qualitative and quantitative assessments on two publicly available benchmark datasets.