Fully Automatic White Matter Hyperintensity Segmentation using U-net and Skip Connection

A Deep Learning Framework for Skull Stripping in Brain MRI

Skull-stripping, an important pre-processing step in neuroimage computing, involves the automated removal of non-brain anatomy (such as the skull, eyes, and ears) from brain images to facilitate brain segmentation and analysis. Manual segmentation is still practiced, but it is time-consuming and highly dependent on the expertise of clinicians or image analysts. Prior studies have developed various skull-stripping algorithms that perform well on brains with mild or no structural abnormalities. Nonetheless, they were not able to address the issue for brains with significant morphological changes, such as those caused by brain tumors, particularly when the tumors are located near the skull's border. In such cases, a portion of the normal brain may be stripped, or the reverse may occur during skull stripping. To address this limitation, we propose to use a novel deep learning framework based on nnUNet for skull stripping in brain MRI. Two publicly available datasets were used to evaluate the proposed method, including a normal brain MRI dataset - The Neurofeedback Skull-stripped Repository (NFBS), and a brain tumor MRI dataset - The Cancer Genome Atlas (TCGA). The method proposed in the study performed better than six other current methods, namely BSE, ROBEX, UNet, SC-UNet, MV-UNet, and 3D U-Net. The proposed method achieved an average Dice coefficient of 0.9960, a sensitivity of 0.9999, and a specificity of 0.9996 on the NFBS dataset, and an average Dice coefficient of 0.9296, a sensitivity of 0.9288, a specificity of 0.9866 and an accuracy of 0.9762 on the TCGA brain tumor dataset.

Automated 2D Slice-Based Skull Stripping Multi-View Ensemble Model on NFBS and IBSR Datasets

This study proposed and evaluated a two-dimensional (2D) slice-based multi-view U-Net (MVU-Net) architecture for skull stripping. The proposed model fused all three TI-weighted brain magnetic resonance imaging (MRI) views, i.e., axial, coronal, and sagittal. This 2D method performed equally well as a three-dimensional (3D) model of skull stripping. while using fewer computational resources. The predictions of all three views were fused linearly, producing a final brain mask with better accuracy and efficiency. Meanwhile, two publicly available datasets-the Internet Brain Segmentation Repository (IBSR) and Neurofeedback Skull-stripped (NFBS) repository-were trained and tested. The MVU-Net, U-Net, and skip connection U-Net (SCU-Net) architectures were then compared. For the IBSR dataset, compared to U-Net and SC-UNet, the MVU-Net architecture attained better mean dice score coefficient (DSC), sensitivity, and specificity, at 0.9184, 0.9397, and 0.9908, respectively. Similarly, the MVU-Net architecture achieved better mean DSC, sensitivity, and specificity, at 0.9681, 0.9763, and 0.9954, respectively, than the U-Net and SC-UNet for the NFBS dataset.