Research A rticle
Objective Ventricle Segmentation in Brain CT with Ischemic
Stroke Based on Anatomical Knowledge
Xiaohua Qian,
1
Yuan Lin,
2
Yue Zhao,
1
Xinyan Yue,
3
Bingheng Lu,
4
and Jing Wang
4
1
College of Electronic Science and Engineering, Jilin Un iversity, Changchun 130012, China
2
Division of Research and Innovations, Carestream Health, Inc., Rochester, NY 14615, USA
3
Aliated Hospital of the Changchun University of Chinese Medicine, Changchun 130021, China
4
Collaborative Innovation Center of High-End Manufacturing Equipment, Xi’an Jiaotong University , Xi’an 710054, China
Correspondence should be addressed to Bingheng Lu; bhlu@xjtu.edu.cn and Jing Wang; wjwjggg@gmail.com
Received June ; Revised August ; Accepted December ; Published February
Academic Editor: Dariusz Mrozek
Copyright © Xiaohua Qian et al. is is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Ventricle segmentation is a challenging technique for the development of detection system of ischemic stroke in computed
tomography (CT), as ischemic stroke regions are adjacent to the brain ventricle with similar intensity. To address this problem, we
developed an objective segmentation system of brain ventricle in CT. e intensity distribution of the ventricle was estimated based
on clustering technique, connectivity, and domain knowledge, and the initial ventricle segmentation results were then obtained. To
exclude the stroke regions from initial segmentation, a combined segmentation strategy was proposed, which is composed of three
dierent schemes: () the largest three-dimensional (D) connected component was considered as t he ventricu lar region; () the
big stroke areas were removed by the image dierence methods based on searching optimal threshold values; () the small stroke
regions were excluded by the adaptive template algorithm. e proposed method was evaluated on cases of patients with ischemic
stroke. e mean Dice, sensitivity, specicity, and root mean squared error were ., ., ., and . mm, respectively.
is system can oer a desirable performance. erefore, the proposed system is expected to bring insights into clinic research and
the development of detection system of ischemic stroke in CT.
1. Introduction
Computed tomography (CT) is generally used to assess
patientswithacuteischemicstrokeinAmerica,becauseofits
faster speed, the better contrast of bones and blood, and the
lower cost than magnetic resonance images (MRI). e ische-
mic stroke and cerebrospinal uid (CSF) regions have a
similar appearance in CT images; thus, accurate ventricle seg-
mentation can signicantly facilitate ischemic stroke region
localization and is an indispensable step for the develop-
ment of computer-aided detection (CAD) for acute ischemic
stroke.
Several state-of-the-art methods have been proposed to
segment ventricles in MRI [], including active contour-based
methods [–], fuzzy schemes [, ], and probability methods
[, ]. However, these methods may be inappropriate to work
on CT images, since there are lower contrast, higher noise
level, and larger slice thickness in brain CT images.
Only little literature on the seg mentation of brain CT
images has been published. For example, Wei et al. proposed
a segmentation scheme based on D Otsu thresholding
approach []. Lee et al. applied the -means and expectation
maximization clustering to segment CT images []. Another
method by Chen et al. was based on a Gaussian mixture
model []. Gupta et al. integrated the adaptive threshold,
connectivity, and domain knowledge to classify t he cere-
brospinal uid, white matter , and gray matter on CT images
[]. ese methods mentioned above were not designed
specically for ventricle segmentation and were not validated
on the images with severe abnormalities. Chen et al. devel-
oped a ventricular segmentation system by combining low-
level segmentation and high-level template matching [].
Similarity, Liu et al. propos ed a model-guided segmentation
for ventricle region []. e two methods are both based
on the template or model scheme for ventricle extraction in
CT. Since these templates were yielded from the MRI brain
Hindawi
BioMed Research International
Volume 2017, Article ID 8690892, 11 pages
https://doi.org/10.1155/2017/8690892
BioMed Research International
image and registration was linear, the templates only provided
a rough mask for the ventricle segmentation. erefore, it
is still challenging for these two methods to exclude stroke
regions from segmentation results. Qian et al. proposed a
level set model to segment CSF, but the result includes the
strokeregions[].isstudywillimprovethemethodsand
extensively validate our previous work [].
e signicant diculty of the accurate ventricle seg-
mentationistodealwithCTimagesofpatientswith
ischemic stroke. Some of the stroke regions and ventricles
are connected and have similar intensities. To address this
challenge, we developed an objective segmentation strategy
of brain ventricles in unenhanced CT with ischemic stroke.
We applied the following three schemes to exclude the stroke
regions from segmentation results:
() We took the largest three-dimensional (D) con-
nected component in a preliminary segmentation as
the ventricular region, removing the lesion or other
regions without the D connectivity relationship with
the ventricle, since the initial segmentation result
contained not only the ventricle but also some non-
ventricular regions, such as lesion or CSF.
() e large stroke regions were removed by the image
dierencemethod.elargestrokeareastendtoclose
the brain edge, and their intensities were generally
lower than that of the main parts of ventricles. us,
the stroke region can be extracted by the dier-
ence between segmentation results from two optimal
threshold values.
()esmallstrokeregionswereremovedbytheadap-
tive template algorithm. e adaptive template was
directly generated from the corresponding image
itself based on the big intensity dierence between the
main part of the ventricle and the brain parenchyma.
is template did not contain the whole ventricle but
did cover the main part of the ventricular region.
us, we applied this template to remove the small
lesions around the main part of the ventricle, which
was not subjected to the registration. Another eect
wasthattheexclusionofthesesmalllesionsmight
break t he connectivity relationship between the lesion
regions and the ventricular region in D space.
2. Materials and Methods
AsshowninFigure,theautomatedventriclesegmentation
method is comprised of two phases, that is, alignment phase
(Sect ion .) and segmentation phase (Section .). In the
alignment phase, the light curves/segment of the brain was
detected to determine the midsagittal line for each slice. We
then aligned the midsagittal line (MSL) with the vertical
line of each slice to achieve brain alignment. In the segmen-
tation phase, we rst estimated the intensity range of the
ventricle region based on clustering technique, connectivity,
and domain knowledge. An image dierence algorithm was
developed to identify and remove the large stroke regions in
the initial segmentation. e remaining small stroke region
was fur ther excluded by an adaptive template of the ventricle.
Finally, the largest D connectivity of the segmented ventricle
was employed to rene the segmentation result.
2.1. Dataset. WetestedtheproposedmethodonCTscans
of patients with ischemic stroke in this study. is dataset
was collected f rom Jilin University Medical Center using CT
scanners (Light Speed , GE Medical System) with an X-ray
tube voltage of kVp. Each patient has slices with the
thickness of mm in this study. e matrix size of each slice
is × pixels, and the pixel size is . mm with a -
bit gray level. e patients were composed of males and
females, and their average age is years with the range
between years and years. We established a reference
standard of ventricle for evaluation of segmentation result. A
medical physicist (XQ, eight years of experience) manually
delineated the ventricle boundaries for all the slices on an
LCD screen as the reference standard to assess the accuracy
of segmentation results.
2.2. Alignment of the Brain Image. Prior to the alignment of
brain image, the skull was stripped by a threshold method
since CT number of bone tissues are consistently higher than
brain tissues. Generally, the CT number of so tissue is less
than Hounseld units (HU) (such as – HU of ventricle,
– HU of white matter, and – HU of gray matter),
whileaverageCTintensityisHUforbones.us,we
extracted the skull using a xed threshold of HU. e
region inside the skull was considered as brain region and the
region outside the skull served as background.
Aer the extraction of the brain, the inclination angle and
position were corrected by aligning MSL with the vertical
centerline of each slice. e determination of MSL is a key
step in this alignment. Since the falx cerebri (i.e., narrow light
curve/segment) presents on about % images, we applied the
falx cerebri as a reference to identify the MSL. erefore, we
utilized two steps to achieve alignment of the brain, including
(1) detectionofalightcurveinthebrainand(2) ane
transformation based on MSL.
2.2.1. Detection of Light Curves in Brain. Figure shows the
schematic diagram of light curve detection. To accelerate
the detection, we dened a rectangle region of interest
(ROI), whose size was chosen to include the light curves
to be detected. We selected a smallest minimum bounding
rectangle of the brain area in the whole scan and then dened
the half width of this rectangle as the width of the ROI.
e height of the ROI was taken the default value of .
Figure (b) shows the rectangle ROI of the brain.
CT brain image has a high level of noise. e common
ltering may blur the weak edge, making detection of the
light curve dicult. e light curve has a slight angle with
the vertical direction; however, it is still regarded as vertical.
us, we designed a one-dimensional (7×1)Gaussianlter
with the variance of to smooth the image along the vertical
direction, which can preserve the edge information of the
light curve in the horizontal direction as shown in Figure (c).
We then design a horizontal Laplacian detection mask,
that is, [0.5,0,1,0,0.5], to detect the light curve, since the
verticalstripincludedmoreedgepointsofthelightcurve
BioMed Research International
Light curve detection Ane transformation
Alignment
Estimation for intensity range of ventricle
Preliminary
segmentation
Large stroke region
removed by image
dierence method
Small stroke region excluded by
the adaptive template matching
Maximum 3D
connected region as
the ventricle
Segmentation
F : Schematic framework for segmentation of the brain ventricle in CT of patients with ischemic stroke.
than other places. With the Laplacian image (Figure (c)),
we employed an adaptive threshold to yield an edge map, as
shown in Figure (d). We empirically set the threshold as the
average value with . multiple of the standard deviation of
the Laplacian image.
Aer that, we erased the small unconnected noise point
clouds in the edge map based on D connectivity. e noise
points in edge map may negatively aect the subsequent
D tting of the middle sagittal plane. However, the D
connectedvolumeofthesenoisepointsissmall;thus,we
canremovethemwithathresholdinDconnectedvolume.
In our experiment, we applied thirty pixels as the threshold
to obtain the clean edge map of light curve (Figure (f)).
Figure (g) shows the D edge map of light curves.
2.2.2. Ane Transformation Based on MSL. To obtain the
precise MSL, we rst tted a middle sagittal plane in D
Euclidean space through a set of edge segments of light
curves using least-squares tting approach. Let (
𝑖
,
𝑖
,
𝑖
)be
a point of edge segments, which has totally points and
=1,2,...,.So,theoptimumttingplanecanbeachieved
by the following formulation as
∗
,
∗
,
∗
=arg min
(𝑎,𝑏,𝑐)
𝑀
𝑖=1
𝑖
−
𝑖
−
𝑖
+
2
.
()
e MSL of each slice was dened as the intersection line
between the image and middle sagittal plane. Let
𝑖
denote
the thsliceofDimage,andwecanobtaintheMSLofthis
slice as
−=
𝑖
−.
()
e determined MSL was shown in Figure (a). Finally,
we aligned the MSL of the brain with the vertical center line
of a slice using the ane transformation dened by
=−
0
cos +−
0
sin +
0
=−
0
sin +−
0
cos +
0
,
()
where (
0
,
0
)is the center point of the vertical center line
of a slice and is the inclination angle between the MSL
BioMed Research International
ROI
extraction
Vert ica l
ltering
3D display of
light curve
80
40
100
150
200
250
300
350
400
450
4
8
12
Denoising
(a) (b) (c) (d) (e) (f)
Horizontal
Laplacian
detection
Light curve
detection
(g)
150
200
250
300
350
y
z
x
F : Diagram of light curve/segment detection: (a) original image without skull; (b) the ROI of the lig ht curve; (c) the vertical ltered
ROI; (d) the Laplacian image; (e) the detected light curve; (f) denoising light curve; (g) D display of light curves: -axis represents the slice
number; -and-axes denote the pixel number.
(a) (b) (c)
F : Alignment of the brain image: (a) original image with the midsagittal line (MSL, dashed line); (b) the vertical center line of a slice
with white color and the MSL; (c) aligned brain image.
and vertical center line. Figures (b) and (c) show that the
inclination angle and position of the brain were corrected.
2.3. Segmentation of the Ventricle. In the phase of ventricle
segmentation, we focused on excluding the stroke area in the
ventricle segmentation result. e owchart was shown in
Figure .
2.3.1. P arameter Estimation for the Ventricle. Prior to the
segmentation of ventricle, we estimated parameters of the
intensity distribution of the ventricle. We rst applied the -
means algorithm (=2) on the D images for stratication
of the brain image and took the largest D connected
component of low-intensity category as the ventricle. en,
an estimation method based on connectivity and domain
knowledge from the literature [] was utilized to compute
the intensity distribution of dierent tissues. Specically, we
tracked the slop of the histogram corresponding to the D
largest connected component in rough intensity range of
ventricle to determine a critical intensity, which serves as
an initial classier of cerebral spinal uid and white matter.
resholds of cerebral spinal uid, white matter, and gray
matter are optimally derived to minimize spatial overlap
errors in dierent tissue types. In this study, ventricular
intensity range of
min
max
will b e adopted to extract the
ventricular region.
2.3.2. Preliminary Segmentation for the Ventricle Based on
Estimated Parameters.
max
,theestimatedmaximumofven-
tricular intensity range, was applied as a threshold value for
BioMed Research International
(a)
(b)
(c)
(g)
(d)
(e)
(f)
(h)
Ventricle region without
big stroke regions
Extraction of the big
stroke region by image
dierence approach
Determination of the 2nd
critical segmentation
without “stroke region”
Preliminary
segmentation
Checking of
stroke region
Small stroke region
excluded by the adaptive
template matching
Maximum 3D connected
region as the ventricle
Ye s
No
F:Flowchartoftheexclusionofstrokeareaintheventricularsegmentationresult.
preliminary segmentation of the ventricle. If the intensity
rangeofthestrokeisgreaterthan
max
, the preliminary
segmentation is a good result. Whereas, if the intensity range
of the stroke is less than
max
,thesegmentationresultmaybe
unacceptable, since it may also contain some stroke regions.
en, we utilized the D connectivity of the preliminary
segmentation result to obtain the largest volume as the initial
segmentation of the ventricle. e stroke regions or noise
areas without the D connectivity to the ventricle could be
excluded by this step. Figure (b) shows that the large stroke
regions are connected to the ventricle in the segmentation.
2.3.3. Detection of the Big Stroke Regions. Since big stroke
regions are mainly related to the anterior cerebral artery or
middle cerebral artery, these stroke regions are mostly closed
to the brain edge. us, we proposed a brain edge checking
algorithm to determine whether the big stroke regions exist in
the segmentation result. An annular region of the brain edge
was dened to detect the objects. Assumed that the minimum
side length of the minimum bounding rectangle of the brain
was
min
, the width of the annular region could be calculated
by 0.15 ×
min
to avoid some parts of the ventricle falling
withintheannularregion.emaskofthebrainedgeannular
region was shown in Figure (c). us, if the objective area
was greater than the threshold, we labeled it as the stroke
region. e threshold was empirically selected as pixels to
allow the presence of noise.
2.3.4. Determination of the Big Stroke Regions. We proposed
an image dierence technique based on the heuristic search-
ing algorithm to extract the big stroke regions, which were
successfully detected in the preliminary segmentation by
the edge checking method. is image dierence technique
essentially applied the dierence between two segmentation