An Improvement Method of Brain Extraction Tools for Magnetic Resonance Images

The brain segmentation of MRI images is the basic step for most studies of brain, such as quantification analysis of brain tissues, and function and fiber analysis of central nervous system. The Brain Extraction Tools (BET), which has been integrated into some application software such as MRIcro and FMRIB Software Library (FSL), plays an important role in the scientific study of brain. But there are still some problems in terms of accuracy and the degree of automatic processing. An improved method of BET is proposed in this paper, which gives a refined initial deformable model and an improved deformation process. It can provide higher accuracy and robustness than the original BET for automatic brain segmentation. The performance evaluation of this improvement has been done by using some clinical MRI data with glioma that are 3D T1-weighted sequence. Comparing the segmentation results with the "ground truth" of brain, the average overlay rate is 95.11%, greater than the 94.32% of the original BET, and the average extra rate, which indicates the ratio of the residual non-brain tissues after segmentation to total non-brain tissues, is 1.09%, much lower than 10.54% provided by the original BET in FSL.