Magnetic resonance imaging (MRI) finds wide application in various studies and clinical practice related to the quantitative and intuitive assessment of cerebral nerve system because of its good contrast and high resolution for brain structures. Most of them require an image processing step called brain extraction by which only the brain part is segmented from cranial magnetic resonance (MR) images.
Though a number of brain extraction algorithms have been presented, brain extraction tool (BET) is still regarded as a favorable tool in the neuroimaging community, and most of the brain extraction algorithms proposed up to date have used the BET as an important competing method to compare their performance.
BET based on the deformable surface initializes the surface as a spherical mesh, and then evolves the surface toward the brain border with small movements applying iteratively a set of forces depending on local parameters to the vertices on the surface. Adopting the local parameters, in general, doesn’t guarantee the balanced evolution all over the surface to generate self-intersections because some vertices may move more quickly while others move slowly depending on the local conditions. This is why BET should have small movements for evolution. Because BET adopts local parameters and small movements, the evolution of deformable surface may not only require more iteration but also tend to easily fall into local optimum resulting in falsely negative regions. Though the computing efficiency of BET is acceptable for clinical applications at present, the computation time is still an important issue when taking the increasing resolution of MRI or large-scale studies into account.
Son Chang Il, a researcher at the Faculty of Biology and Medicine Engineering, has proposed a modified BET (BETWP) consisting of two steps of surface evolution for fast and accurate brain extraction.
He introduced a new fast model using a global parameter, the global mean inter-vertex distance of evolution surface. This fast model is adopted in the preprocessing step and then the original BET model completes the evolution of surface in the second step.
The experiments for evaluating the computation efficiency and segmentation quality have shown that the proposed scheme has a couple of advantages over BET. First, it can improve the evolution speed at least twice for brain extraction without any failure due to self-intersection. Second, it can significantly improve the segmentation quality on JC, TE and NE including the false negative ratio for both MRI modalities (T1-weighted image and DW image).
For more information, please refer to his paper “Fast BET Based on Pre-Processing Evolution Using Global Mean Inter-Vertex Distance of Deformable Surface for MRI Brain Extraction” in “International Journal of Image and Graphics” (SCI).
© 2021 Kim Chaek University of Technology