MRI Skull Stripping Review

Posted by Kaiyuan Chen on May 16, 2018

A review of Skull Stripping For Head Scan Images

what is MRI

Computer aided diagnosis based on medical images from MRI(magnetic resonance image) has gained ubiquitous usage for its “noninvasive, nondestructive, flexible” properties[2]. With the help of FLAIR(Fluid-attenuated inversion recovery), Diffusion-weighted(DW) MRI, people can get an anatomical structure of human soft tissues with high resolution. Especially to satisfy the demand for interior and exterior structure of brain structures, MRI can produce cross-sectional images from different angles, for example, top-down, side-to-side and front-to-back: however, having slices from different angles give a lot of challenges in stripping those tissues which people are interested in, from xtra-cranial or non-brain tissues that has nothing to do with brain diseases such as Alzheimer’s disease, aneurysm in the brain, arteriovenous malformation-cerebral and Cushings disease and etc[1].

Why Skull Stripping

As a preliminary step for further analysis, brain segmentation, i.e. skull stripping, needs both speed and accuracy in practice, which should be considered in any algorithms proposed. By Kalavathi et al. [2], they can be classified into five categories: mathematical morphology-based methods, intensity-based methods, deformable surface-based methods, atlas-based methods, and hybrid methods. However, as we further reviewed on state-of-arts that are vaguely described in hybrid methods, we believe machine learning-based methods should also have its own place in brain segmentation. Based on above six categories proposed in Figure 1, we will introduce major algorithms and properties under each category.

Categories

Mathematical morphology-based methods, for example,the one proposed by Brummer et al.[2], is parametric and depend heavily on presetted parameters such as the size and shape of the structure to perform morphological operation. Intensity-based methods relies on intensity of each image pixel. However, as equipments failures and heavy assumptions may result in high-biased imperfections. These imperfections will reduce the accuracy of this highly sensitive algorithm. Another class of method is Deformable Surface-Based Method. These methods use an active contour model that starts with a closed curve, and progress with two operations, shrink or expand, based on the influence functional. With these two operations, the method tries to change the shape of the curve towards the desired object boundaries. For example, Suri devised an active contour algorithm that uses the level set methods to evolve the active contour [3]. It uses a fuzzy membership function to classify brain images into four components: WM, GM, CSF, and background, then used a gradient detector and a deformable model to evolve an active contour to fit the surface between the CSF and GM.

Machine learning

Machine learning is a broad concept that include many interesting algorithms that we would like to implement and experiment on. For example, Butman introduced a robust machine learning method that detects the brain boundary by random forest[2]. As random forest has high expressive power on voxels of brain boundary, this method can reach an high accuracy robustly. Popular methods like deep learning can also applied. For example, Kleesiek et al.[8] used non-parametric 3D Convolutional Neural Network(CNN) to learn important features and reach the highest Dice score among all the methods we have reviewed. However, as a parametric algorithm, GMM also has its place in brain segmentation. For example, Yunjie et al. developed a skull stripping method with an adaptive gauss mixture model and a 3D mathematical morphology method. The GMM is used to classify brain tissues and to estimate the bias field in the brain tissues [5]. These methods, along with well-implemented libraries such as sklearn[6], Tensorflow[7], are readily available for our use.

Reference:

[1] NLM-National Library of Medicine, (Rockville Pike, Bethesda U.S., 2011) [2] Kalavathi, P., and V. B. Surya Prasath. “Methods on Skull Stripping of MRI Head Scan Images—a Review.” Advances in Pediatrics., U.S. National Library of Medicine, June 2016, www.ncbi.nlm.nih.gov/pmc/articles/PMC4879034. [3]Two-dimensional fast magnetic resonance brain segmentation. Suri JS IEEE Eng Med Biol Mag. 2001 Jul-Aug; 20(4):84-95. [4] Butman J, Roy S, Pham D. Robust skull stripping using multiple MR image contrasts insensitive to pathology. NeuroImage . 2017;146:132-147. [5] Yunjie C, Jianwei Z, Shunfeng W. A new fast brain skull stripping method, biomedical engineering and informatics. Tianjin: Proc. 2nd International Conference on Biomedical Engineering and Informatics, BMEI09; 2009. [6] Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011. [7] Martín, et al. “TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems.” [1402.1128] Long Short-Term Memory Based Recurrent Neural Network Architectures for Large Vocabulary Speech Recognition, 16 Mar. 2016, arxiv.org/abs/1603.04467. [8] Kleesiek J, Urban G, Hubert A, Schwarz D, Maier -Hein K, Bendszus M, Biller A. Deep MRI brain extraction: A 3D convolutional neural network for skull stripping. NeuroImage . 2016;129:460-469.