BEaST: Brain extraction based on nonlocal segmentation technique

Abstract

-Brain extraction is an important step in the analysis of brain images. The variability in brain morphology and the difference in intensity characteristics due to imaging sequences make the development of a general purpose brain extraction algorithm challenging. To address this issue, we propose a new robust method (BEaST) dedicated to produce consistent and accurate brain extraction. This method is based on nonlocal segmentation embedded in a multi-resolution framework. A library of 80 priors is semi-automatically constructed from the NIH-sponsored MRI study of normal brain development, the International Consortium for Brain Mapping, and the Alzheimer's Disease Neuroimaging Initiative databases.In testing, a mean Dice similarity coefficient of 0.9834±0.0053 was obtained when performing leave-one-out cross validation selecting only 20 priors from the library. Validation using the online Segmentation Validation Engine resulted in a top ranking position with a mean Dice coefficient of 0.9781±0.0047. Robustness of BEaST is demonstrated on all baseline ADNI data, resulting in a very low failure rate. The segmentation accuracy of the method is better than two widely used publicly available methods and recent state-of-the-art hybrid approaches. BEaST provides results comparable to a recent label fusion approach, while being 40 times faster and requiring a much smaller library of priors.Keywords: Brain extraction, skull stripping, patch-based segmentation, multi-resolution, MRI, BET ** Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (www.loni.ucla.edu/ADNI). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. IntroductionBrain extraction (or skull stripping) is an important step in many neuroimaging analyses, such as registration, tissue classification, and segmentation. While methods such as the estimation of intensity normalization fields and registration do not require perfect brain masks, other methods such as measuring cortical thickness rely on very accurate brain extraction to work properly. For instance, failure to remove the dura may lead to an overestimation of cortical thickness (van der Kouwe et al., 2008), while removing part of the brain would lead to an underestimation. In cases of incorrect brain extraction, subjects may be excluded from further processing, a potentially expensive consequence for many studies. The solution of manually correcting the brain masks is a labour intensive and time-consuming task that is highly sensitive to inter-and intra-rater variability (Warfield et al., 2004).An accurate brain extraction method should exclude all tissues external to the brain, such as skull, dura, and eyes, without removing any part of the brain. The number of methods proposed to address the brain segmentation problem reflects the importance of accurate and robust brain extraction. During th...

Methods on Skull Stripping of MRI Head Scan Images—a Review

“…BEaST is a brain extraction method based on nonlocal segmentation technique by Eskildsen et al [84]. In this, a nonlocal segmentation is embedded in a multi-resolution framework.…”

“…BEaST [84] Normalization, construction of brain atlas, patch-based segmentation and uses a library of 80 priors.…”

NABS: non-local automatic brain hemisphere segmentation

“…Previous studies [23,25] have shown that non-local patch-based label fusion approaches using a library of expert priors is a powerful approach for brain structure segmentation. In this paper, we have presented an accurate and fast patch-based multi-template brain segmentation method, termed NABS, for segmenting cerebral and cerebellar hemispheres and brainstem from T1-weighted MR brain images.…”

“…As originally proposed by Coupé et al [23], our method is based on a nonlocal means label fusion where labels from multiple templates are weighted according to the Euclidean distance between patch intensities. Using this approach we avoid the one-to-one matching assumption of nonlinear registration label fusion methods by enabling a one-to-many matching, therefore reducing the segmentation errors [23,25,26] by better managing the inherent inter-subject variability of human brain anatomy.…”

Automated fetal brain segmentation from 2D MRI slices for motion correction

“…Instead of a template, the new image can also be registered to a set of already segmented images, called an atlas (Van Leemput et al, 1999). Tissue classification can then be performed, for instance by seeking similar patches from the atlas in a local neighborhood and transferring their labels to the new image, such as in BEaST (Eskildsen et al, 2012). Such a patch-based segmentation has been applied to fetal data after motion correction (Wright et al, 2012).…”