BEaST: brain extraction based on nonlocal segmentation technique.

TLDR: A new robust method dedicated to produce consistent and accurate brain extraction based on nonlocal segmentation embedded in a multi-resolution framework, which provides results comparable to a recent label fusion approach, while being 40 times faster and requiring a much smaller library of priors.

About: This article is published in NeuroImage.The article was published on 2012-02-01 and is currently open access. It has received 429 citations till now. The article focuses on the topics: Image segmentation & Segmentation.

Figures

Fig. 3. The multiresolution segmentation process (row 1-3) compared to a single resolution approach (row 4). Column 1: Initialization mask. Column 2: Nonlocal means (NLM) estimator map. Column 3: Segmentation by thresholding the NLM estimator and adding the intersection mask. Processing times are accumulated time from initialization. Notice the inclusion of dura in the single resolution approach.

Fig. 1. Construction of library priors using multiple modalities. A) Intensities from T1w, T2w, and PDw images are added. B) BET (Smith, 2002) is used to produce an ICC mask. C) FACE (Eskildsen and Østergaard, 2006) is used to delineate the cortical boundary and produce a cerebrum mask. D) Cerebellum and brain stem are added by stereotaxic masks, and the mask is manually corrected.

Fig. 2. Adaptation of library priors using deformable surface. A) Semi-automatic GM/WM mask as used by MAPS (Leung et al., 2011). B) Adapted mask generated by deforming a surface mesh to the boundary of the GM/WM mask and manually corrected.

Fig. 7. Typical results using BET, VBM8 and BEaST on the five test groups. The figure shows sagittal slices and 3D renderings of the segmentations. Column 1-2: BET segmentation. Column 3-4: VBM8 segmentation. Column 5-6: BEaST segmentation. Blue voxels are overlapping voxels in the segmentation compared to the gold standard. Green voxels are false positives and red voxels are false negatives.

Fig. 6. False positive and false negative maps for BET, VBM8, and BEaST on NIHPD, ICBM and ADNI data. All the error maps are displayed with the same scale. BET provided errors mainly located in the cerebral falx and medial temporal lobe structures. On the ADNI data, BET had a few catastrophic failures, which is visible in the false positive image. VBM8 tended to produce a systematic over-segmention compared to the used manual gold standard. The errors obtained by BEaST were more uniformly distributed indicating non-systematic segmentation errors.

Fig. 8. Examples of ADNI brain masks produced by BEaST not passing the quality control in the first pass (first row) and passing the quality control after second pass (second row). Left segmentation is discarded due to cortex clipping, while right segmentation is discarded due to inclusion of dura as indicated by the arrows. After second pass these errors are removed.

Frequently asked questions

Q1. What have the authors contributed in "Beast: brain extraction based on nonlocal segmentation technique" ?

To address this issue, the authors 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. 

Q2. Who is the grantee organization for this study?

The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Disease Cooperative Study at the University of California, San Diego. 

Q3. What is the name of the research centre?

The Dementia Research Centre is an Alzheimer's Research UK Co-ordinating Centre and has also received equipment funded by the Alzheimer's Research UK. 

Q4. What are the sources of funding for this work?

This work has been supported by funding from the Canadian Institutes of Health Research MOP84360 & MOP-111169 as well as CDA (CECR)-Gevas-OE016. 

Q5. What is the funding source for this project?

Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904). 

Q6. Who is the author of this article?

The authors would like to thank Professor Nick Fox, Dementia Research Centre, Institute of Neurology, London, for contributing with the ADNI semi-automatic brain segmentations.