© 2017, IJCSE All Rights Reserved 45
Research Paper Volume-5, Issue-3 E-ISSN: 2347-2693
An Automated Skull-Stripping Method by Windowing the Histogram
S. Sarkar
1*
, A. Mandal
2
, K. Sarkar
3
1*
Department of Computer Science & Application, University of North Bengal, Darjeeling, India
2
Department of Computer Science & Application, University of North Bengal, Darjeeling, India
3
Ananda Chandra College, Department of Computer Science, University of North Bengal, Darjeeling, India
*
Corresponding Author: suvro.nbu@gmail.com, Tel.: +918942921147
Available online at: www.ijcseonline.org
Received:26/Feb/2017 Revised: 03/Mar/2017 Accepted: 20/Mar/2017 Published: 31/Mar/2017
Abstract— An automated method for segmentation of Magnetic Resonance (MR) head images into brain and non-brain has
been proposed. It combines the strategy used in intensity and morphological skull-stripping methods. The method is very fast
and requires no preprocessing of MR images. It is testified on T1-Weighted MR image modality and produces accurate output.
Keywords—MRI, Skull-Stripping , ROI, T1-Weighted, Windowing
I. INTRODUCTION
The human brain is very complex structure composed of
Neurons, Glial cells, Neural Stem cells and blood vessels. A
key issue in neuro-image e.g. Magnetic Resonance Image,
(MRI) analysis is so called skull-stripping, which plays an
important role in neuro-image processing, is the task of
separating predominant tissues in head, e.g. Gray Matter
(GM), White Matter (WM) and Cerebrospinal Fluid(CSF),
from the other tissues in head such as Bone, Skin, Muscle,
Fat Dura etc. The accuracy of skull-stripping method affects
the later stages of neuro-image analysis to a great extent. The
measured signal intensity of brain tissue and non-brain tissue
can overlap. This overlapping produces ambiguity in
separation procedure and thus resulting false positive and
false negative identification. This problem is so called non-
separability in the context of digital image processing.
II. TECHNICAL REVIEW
In general, skull-stripping, is a non-trivial task as the
acquired MR images are imperfect and are often noisy. The
diversity of MR images of brain led to the development of
various skull-stripping techniques (BET [1], BSE [2], Bridge
Burner [3] and GCUT [4]). Different methods of skull-
stripping available in the literature are broadly classified into
five categories [5]:
i. Mathematical morphology based method.
ii. Intensity based method.
iii. Deformable surface based method.
iv. Atlas based method
v. Hybrid method
The proposed method is based on the combination of
Mathematical morphology and Intensity based methods. So
brief explanations of these two categories only are presented
here:
A. Morphology-based method
In this method, two important mathematical and
morphological operators are used namely: Erosion and
Dilation. In general, the image is first transformed into a
binary image then the following operators are applied:
i. Erosion: The erosion of a binary image ‘A’ by a
structuring element ‘B’ in an Euclidean space ‘E’ is
defined as-
} | { A B E z B A
z
⊂ ∈ = Θ (1)
Where B
z
is the translation of B by the vector z, i.e.
E B b z b B
z z
∈ ∀ ∈ + = }, | { (2)
ii. Dilation: The dilation a binary image ‘A’ , in the
Euclidean Space ‘E’, by a structuring element ‘B’ is
defined as-
} ) ( | { φ ≠ ∩ ∈ = ⊕ A B E z B A
z
s
(3)
Where B
s
denotes the symmetric of B, that is,
} | { B x E x B
s
∈ - ∈ = (4)
An appropriate threshold is applied to find the initial Region
of Interest (ROI), and then the morphological operators are
applied in order to justify the ROI more appropriately to the
desired outcome. One of the commonly used methods in this
category is discussed by Brummer et al [6]. The method is
based on the histogram threshold and morphological
operations. Main drawback of such method, in skull-
stripping, is that the final output is directly influenced by the
parameters such as size and shape of the structural element,