=Paper=
{{Paper
|id=Vol-2485/paper34
|storemode=property
|title=Segmentation Algorithm Based on Square Blocks Propagation
|pdfUrl=https://ceur-ws.org/Vol-2485/paper34.pdf
|volume=Vol-2485
|authors=Viacheslav Danilov,Igor Skirnevskiy,Roman Manakov,Dmitrii Kolpashchikov,Olga Gerget
}}
==Segmentation Algorithm Based on Square Blocks Propagation==
https://ceur-ws.org/Vol-2485/paper34.pdf
Segmentation Algorithm Based on Square Blocks Propagation
V.V. Danilov1, I.P. Skirnevskiy1, R.A. Manakov1, D.Yu. Kolpashchikov1, O.M. Gerget1
viacheslav.v.danilov@gmail.com|skirnevskiy@tpu.ru|ram290495@gmail.com|dyk1@tpu.ru|gerget@tpu.ru
1
Medical Devices Design Laboratory, Tomsk Polytechnic University, Tomsk, Russia
This research is devoted to the segmentation of heart and brain anatomical structures. In the study, we present a segmentation
algorithm based on the square blocks (superpixels) propagation. The square blocks propagation algorithm checks two criteria. For the
first criteria, the current intensity of the pixel is compared to the average intensity of the segmented region. For the second criterion,
the intensity difference of the pixels lying on the superpixel sides is compared to the threshold. Once these criteria are successfully
checked, the algorithm merges homogeneous superpixels into one region. Then the following superpixels are attached to the final
superpixel set. The last step of the proposed method is the spline generation. The spline delineates the borders of the region of interest.
The main parameter of the algorithm is the size of a square block. The cardiac MRI dataset of the University of York and the brain tumor
dataset of Southern Medical University were used to estimate the segmentation accuracy and processing time. The highest Dice similarity
coefficients obtained by the presented algorithm for the left ventricle and the brain tumor are 0.93±0.03 and 0.89±0.07 respectively.
One of the most important features of the border detection step is its scalability. It allows implementing different one-dimensional
methods for border detection.
Keywords: square blocks propagation, superpixels, region growing, left ventricle segmentation, brain tumor segmentation.
algorithms. The watershed segmentation is used as an initial step to
1. Introduction find a seed region.
Joung Park and Chulhee Lee used the seeded region growing
Medical image segmentation is one of the most challenging
algorithm for the skull stripping [18]. In that algorithm, a
tasks in the field of medical image processing. The segmentation
morphological mask was used for the automatic identification of
and the subsequent analysis of medical images allow clinicians to
the initial seed points of background and foreground. Other region-
predict disease, plan surgery procedures or assess the condition of
based methods such as watershed segmentation and morphological
internal organs. At the moment, many robust two- and three-
segmentation are used in tasks of the skull stripping [8, 23].
dimensional segmentation techniques have been proposed [19, 22, However, many of these approaches have drawbacks, such as
27]. The recent and the most popular articles on medical image
oversegmentation and noise sensitivity.
segmentation are inextricably connected with machine learning and
Nor Isa in paper [13] proposed a modified seed-based region
neural networks. Ozan Oktay used neural networks for cardiac
growing algorithm. Several important blocks of the algorithm such
image enhancement and segmentation in paper [17]. In paper [20] as setting the threshold value, determining the initial seed point,
machine learning algorithms are used for brain tumor
and the growing process were modified and automated. For
segmentation. Similar approaches have been used in many tasks of
instance, the automatic determination of the seed point is based on
medical image analysis [14, 16]. However, algorithms based on
the k-means clustering algorithm. However, the approaches
machine learning often solve a narrow problem and require large described in the paper have high computational complexity. This is
training datasets. Despite the popularity of machine learning
explained by a number of preliminary calculations. For instance,
algorithms, common segmentation techniques remain relevant and
the k-means clustering algorithm is working with an entire image.
keep improving [15, 25]. Semi-automatic image segmentation
Jamshid Dehmeshki combined the region growing and the fuzzy
techniques are still popular because of their simplicity, a small
connectivity region growing approaches in paper [11].
number of parameters, and scalability.
Paper [12] shows a modified region growing based on the
Today classical image segmentation techniques (active
merging superpixels. A superpixel is a group of pixels combined
contours, region growing, watershed segmentation) are used in
by a certain feature. The superpixel term was introduced by
many semi-automatic image processing algorithms. It is worth
Xiaofeng Ren and Jitendra Malik in [21]. The superpixel concept
noticing that analysis and processing of three-dimensional images
is used in the presented study. However, the major difference of the
are still difficult especially in the field of cardiology or brain
proposed algorithm is that it does not check every pixel of the
imaging. Therefore, there are cases when clinicians use two-
superpixel. In general, a number of segmentation algorithms based
dimensional planes for the analysis and segmentation. Moreover,
on superpixels were proposed before [9, 10, 24]. It should be noted
the more popular machine learning algorithms become, the more
that the clustering methods are used for a superpixel generation in
data they require for training machine learning models. Thus, there
most studies. In this approach, each pixel included in a superpixel
is a need in an easy-to-use environment for data labeling. Two-
should be processed separately, thus causing a relatively high level
dimensional segmentation techniques are often used as such
of complexity. The complexity of such algorithms is O(n2). A
environments.
simple linear iterative clustering (SLIC) algorithm is used in studies
In this study, we present a two-dimensional segmentation
[12], [20], and [9]. Initially, this method was presented in paper [1],
algorithm based on square blocks propagation (SBP). Dana Ballard
where Radhakrishna Achanta developed a superpixel-based
and Rofl Adams’ algorithms [2, 4] inspired us to develop this
segmentation method using k-means clustering for a five-
method. We also used the approaches described in work [5]. The
dimensional feature space. The first three dimensions are the color
proposed algorithm is slightly similar to a classical region growing
space and the last two are the pixel coordinates. The SLIC
and based on the merging of the samples with similar properties of
algorithm is a modified k-means clustering algorithm which does
the region.
not compare each pixel with other pixels in the image. Ovidiu
Csillik in paper [10] demonstrated a method based on SLIC
2. Related works superpixels for a high-resolution segmentation. The paper presents
Most of the articles devoted to the region growing (RG) a processing time of superpixels at different resolutions. This
algorithms were presented more than a decade ago. Jun Tang method processes 1347 × 1042 and 3701 × 3301 images for 2 and
proposed a method for color image segmentation based on a 26 seconds respectively.
combination of the seeded region growing and watershed algorithm As shown above, segmentation algorithms based on
[26]. However, there were no modifications or improvements in the superpixels are quite popular. However, most of the considered
Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
approaches have a high level of algorithmic complexity. In this 3.3 Square blocks propagation
regard, we propose a 2D semi-automatic segmentation algorithm
based on a superpixel growing where superpixels have a floating There is no conceptual difference between the classical seeded
size. The latter allowed us to achieve better algorithmic region growing and the proposed algorithm. However, the SPB
performance. algorithm assumes a translation to the domain of superpixels. This
allows reducing the complexity of the algorithm and increasing the
3. Data and methods processing speed. In the proposed algorithm a superpixel represents
a square block comprising of the pixels. All pixels inside the block
3.1 Data description have a 4-connected neighborhood by default. The SPB algorithm
checks two criteria, described in Section 3.6 in more detail, and
In order to develop and validate the proposed algorithm, we used merges superpixels into one region. Then points lying on
open-access datasets. The first dataset was provided by the superpixels borders are used for spline generation. The workflow
University of York (York, United Kingdom) and contains 33 of the proposed algorithm is shown below in Fig. 3.
subjects [3]. Each subject's sequence consists of 20 frames and 8-
15 slices (256x256 pixels) along the long axis, for a total of 7980 Initialization of Placing of the seed Square
the parameters square propagation
images. Two clinicians manually segmented all the images of the
dataset. The ground truth of the left ventricles' endocardial and
epicardial was acquired.
The second source of data is the brain tumor dataset. This Repeating steps 3
Searching for the Square size
and 4 until stop
dataset includes 3064 T1-weighted contrast-enhanced images. The nodal points reduction
criterion is met
dataset was acquired at Southern Medical University and contains
data from 233 patients with three kinds of brain tumors:
meningioma, glioma, and pituitary tumor (Guangzhou, China) [6,
Getting the final
7]. The size of MRI images is 512*512 pixels. Examples of the Spline generation mask
heart and the brain tumor images are shown in Fig. 1.
Both datasets were processed offline on the computer equipped Fig. 3. The workflow of the square blocks propagation algorithm
with Intel Core i7-4820K 3.7GHz CPU and NVIDIA GeForce 960
GT using MATLAB (MathWorks, Natick MA). The main difference between the proposed algorithm and the
methods reflected in papers [12, 28] is that the proposed algorithm
does not analyze single pixels belonging to superpixels. For
instance, the standard region growing algorithm processes all 100
pixels of a 10x10 superpixel. In turn, the presented algorithm
processes only 50 out of 100 pixels. Thus, the larger the initial
square size is, the higher the algorithm speed is. The concept of the
square propagation and the size reducing procedure is shown in Fig.
4.
(a) Heart MRI sample (b) Brain tumor MRI sample
Fig. 1. Examples of source data
3.2 Region growing algorithm
Starting from a seed of the region of interest (ROI) the region
growing algorithm performs a segmentation. The region is growing
due to the connection of the neighboring pixels, which satisfy the
(a) Propagation using initial (b) Propagation using squares
criterion of homogeneity. There are two versions of the algorithm:
size squares with the reduced size
a seeded version with a manual selection of the seed point and an
unseeded version with a random seed point. A classical region Fig. 4. Square propagation and square size reduction. Blue blocks
growing algorithm is conceptually shown in Fig. 2. show the first step of the square propagation with an initial
superpixel size. Purple blocks demonstrate the second step of the
Combining a seed pixel
Select a Stop the square propagation and the square size reduction
with the neighboring pixels
seed algorithm
using similarity criterion
The proposed algorithm applies superpixel growing instead of
Fig. 2. The workflow of the region growing algorithm pixels merging. This approach accelerates the segmentation
process. Similar to the region growing algorithm, a merging
A criterion of merging neighboring pixels is presented below process occurs as long as there is at least one group of pixels that
and also shown in [2]. could be included in the final set. The superpixel is included in the
𝑃(𝑥, 𝑟) = |𝑓(𝑥) − 𝜇𝑟 | < 𝑇, (1) final superpixel set when the superpixel sides do not cross the
where f(x) is the intensity of the current pixel, 𝜇𝑟 is the arithmetic border of the ROI (see Section 3.6). If there is at least one border
mean intensity of the region, T is the threshold level. The approach crossing on the checked segments, then the square block is not
described above is a standard implementation of a region growing included in the final superpixel set.
algorithm. The first square center is the starting point chosen manually. A
square of a given size is placed around the first point. The diagonals
and the sides of the square are checked for the border crossing. If
the square does not cross the border of the ROI, it is placed in an
image. Otherwise, the algorithm can continue only if the size of the a mask. However, it is possible to switch from spline to mask using
initial square block is reduced. existing methods.
3.4 Search for outer squares
Superpixels are validated and attached to the final superpixel
set as long as possible. If there are no squares with the given size
which can be added to the studied region, the size of the square is
halved and the propagation process continues. An example of the
outer squares found by the algorithm is shown in Fig. 5. The green
points indicate an intersection of the squares with the ROI.
Fig. 6. Hermite cubic spline (blue line) passing through the nodal
points (green points)
Fig. 5. Example of outer squares that cannot be added to the ROI
(yellow squares) because of the intersection with the segmented
area border. Green points demonstrate this intersection
The square size reducing procedure can be repeated several
times until the minimum size of the square is reached. The
minimum square size is one of the parameters of the algorithm. At
the final step of the algorithm, the superpixels with the smallest size
are located close enough to the border of the segmented area but
never cross it. An obligatory condition for completing this stage is
the impossibility of further attaching the square blocks.
3.5 Delineation and masking
Fig. 7. Obtaining a segmentation mask. Cubic spline (blue) and
At this step, the algorithm bypasses the outer superpixels which segmentation mask (purple) represent the region of interest
have not been attached to the region. These squares form a contour
in accordance with the mandatory condition described above in 3.6 Border detection
Section 3.4. The contour consisted of outer superpixels is bypassed
clockwise. The intersection points are saved for each superpixel To detect the border of the ROI, the proposed algorithm applies
crossing the contour of the ROI, creating a list of nodal border two conditions. For the first condition, the intensity of the
points. The same procedure is performed to bypass the inner side/diagonal pixel is compared with the mean intensity of center
borders of the region if such exist. pixels of already placed squares. The threshold parameter is a
Having a list of nodal points, it is possible to construct a configurable parameter of the algorithm.
Hermite cubic spline describing the inner and outer contours of the ∑𝑘𝑗=1 𝑝𝑐𝑗
segmented region (see Fig. 6). Thus, the result of the algorithm is a ∆𝐼 = |𝑝𝑖 − | ≥ 𝑇, (2)
𝑘
spline or a set of splines.
Fig. 6 shows a conceptual scheme of spline generation. As where 𝑝𝑖 is the intensity of the current pixel, 𝑝𝑐𝑗 is the intensity of
shown, the blue line has an unusual shape for a cubic spline. The the center pixel of a certain square, 𝑘 is the number of already
output segmentation mask obtained using spline generation is placed squares, 𝑇 is the threshold level.
shown in Fig. 7. For the second condition, the intensity difference of the pixels
Among the entire set of image pixels L, the algorithm finds the lying on the superpixel sides is compared with the threshold. The
set of intersection points P. Each intersection point represents a approximation is performed using the method of least squares at 5
point where a square block sides or diagonals cross the border of a points. The coefficient, representing a slope of the straight line is
region. One of the ways to obtain a contour is to construct a calculated as follows:
regression of these points. However, we did not use a linear spline,
𝑛 ∑𝑛𝑖=1 𝑥𝑖 𝑦𝑖 − ∑𝑛𝑖=1 𝑥𝑖 ∑𝑛𝑖=1 𝑦𝑖
as it gives a significant error in constructing the contour borders. 𝑡𝑎𝑛 (𝜑) = | | ≥ 𝑠𝑙𝑜𝑝𝑒, (3)
We also refused to use B-splines because they do not pass through 𝑛 ∑𝑛𝑖=1 𝑥𝑖2 − (∑𝑛𝑖=1 𝑥𝑖 )2
the extreme points. The latter is not acceptable since it significantly where x is the edge points varied in a certain range (in our case this
reduces the accuracy of segmentation. range is from 1 to 5), y are intensity values, n are positions of the
Thus, the result of the algorithm is the set of cubic splines edge points. When constructing the vector x, it should be
describing the contours. Such an analytical presentation may be considered that the distance between pixels is the Euclidean.
more preferable than representing a segmented area in the form of For the current pixel, the approximation is done using two
pixels on the right and two on the left. If the pixels are in the corner
of a superpixel, the additional pixels that slightly exceed the the first square where a number of iterations is equal to 6×N. In this
borders of the square block are taken. The slope module allows the regard, the asymptotic complexity of the SBP algorithm is O(n). As
algorithm to accurately detect the region borders and makes the shown, the proposed approach moves from the pixel level to the
algorithm resistant to noise. Border search is applied on the sides level of the pixel groups and fragments. The latter allows
and diagonals of the square blocks using conditions described remarkably reducing the execution time of the algorithm.
above. If there is at least one side/diagonal crossing of the region
border, the square block is not included in the final set. Thus, the 4. Results
reliability of the algorithm rises.
In this section, we studied how the accuracy and processing
As shown in Fig. 8, each square block has six-line segments
time changed with respect to different sizes of the squares. The
(AB, BC, CD, AD, AC, and BD). These lines consist of pixels. All
Dice Similarity Coefficient (DSC) was used as the main metric for
that we need to do is to perform a one-dimensional segmentation
for each line. This approach significantly reduces the execution the accuracy assessment.
time of the algorithm. Another advantage of the proposed solution 4.1 Left ventricle segmentation
is the possibility to apply any set of one-dimensional segmentation
The left ventricle segmentation of the presented algorithm,
methods to a line segment. Therefore, any segmentation method
region growing algorithm, and the ground truth (GT) manual
can be implemented in the proposed algorithm as a plug-in for the
segmentation is presented in Fig. 9. In the case of patient 2 and
additional verification of the border crossing. It is worth noticing
patient 3, region growing leaks out through the gaps in the borders
that if at least one segment has crossed the border of the region, the
of the ROI. This is because the region growing approach processes
algorithm does not process the rest of the line segments. The latter
an image at the pixel domain. In turn, the SPB algorithm avoids the
allows reducing the algorithm runtime and increasing the
problem of the border gaps.
performance of the algorithm by 5 times in the extreme case.
Patient 1
Patient 2
Fig. 8. Square block (red) crossing the ROI in three points (green)
3.7 Size of square blocks
The region growing algorithm often leaks through the holes in
borders. Some modifications of the region growing [21] can help
to avoid this effect, but these methods still have high computational
complexity. The proposed algorithm eliminates this disadvantage
due to the variability of the square blocks size. The minimum
square size also has an influence on the contour details and its Patient 3
smoothness. The larger the side of the square is, the smoother and
less detailed the contour is. As the square size decreases, the border
becomes more detailed while the probability of negative leakage
effect increases.
The initial square size defines the propagation speed. It should
also be noted that the initial square size depends on the segmented
area and the image resolution. For initial square size K we used the
method presented in paper [12], where parameter K is calculated as
follows:
𝑁 (4)
𝐾=
𝑆𝐹
where N is a number of image pixels, and SF is the size of the (a) Square blocks propagation (b) Region growing algorithm
smallest segmented object in the image. algorithm (blue) vs manual (red) vs manual segmentation
3.8 Algorithm complexity segmentation (cyan) (cyan)
Fig. 9. Segmentation of the left ventricle using the proposed SBP
In the case of the region growing, at least N2 steps are required and RG algorithms in comparison with the ground truth manual
to process a block of N×N pixels. Consequently, the complexity of segmentation
the region growing is O(n2). Each image pixel is processed
separately by the region growing algorithm. For the proposed To test the segmentation accuracy and processing time of the
algorithm, a number of iterations for the block size of N×N pixels left ventricle, we used a dataset comprised of 156 slices. The left
varies from 3×N to 5×N. However, there is an exceptional case for ventricle segmentation accuracy for different square sizes is shown
in Fig. 10 and Table 1. Additionally, the total number of low-
accuracy cases when DSC is less than 0.5 is shown in Fig. 11.
Patient 2
Fig. 10. Left ventricle segmentation accuracy of the proposed
algorithm
Table 1. DSC obtained on the left ventricle dataset for different
algorithms.
SBP 20-10-5 SBP 16-8-4 SBP 12-6-3 SBP 8-4-2 RG
Patient 3
0.93±0.03 0.91±0.07 0.89±0.10 0.86±0.11 0.88±0.09
As shown in Fig. 10 and Table 1, DSC values of SBP 8-4-2,
SBP 12-6-3 and RG do not differ significantly. However, the DSC
interquartile range of SPB with parameters 20-10-5 and 16-8-4 is
significantly better than RG’s one. The average accuracy of the
SPB algorithm has grown significantly due to the fact that
superpixels do not leak through the border gaps.
Better performance of the proposed algorithm is indirectly
confirmed by a number of low-accuracy segmentation cases (see
Fig. 11). A low-accuracy case is a leakage case or a case with the
value of DSC less than 0.5. In 32% of the studied cases, RG leaks
through the borders defects what confirms its unreliability. (a) Square blocks propagation (b) Region growing algorithm
algorithm (blue) vs manual (red) vs manual segmentation
60
50 segmentation (cyan) (cyan)
50 Fig. 12. Segmentation of the brain tumor using the proposed SBP
Leakage cases
and RG algorithms in comparison with the ground truth manual
40
segmentation
30
To test the segmentation accuracy and processing time of the
20 brain tumor, we used a dataset comprised of 300 slices. The brain
13
tumor segmentation accuracy for different square sizes is shown in
10 4 6 Fig. 13 and Table 2. Additionally, the total number of low-accuracy
1
cases when DSC is less than 0.5 is shown in Fig. 14.
0
SBP 20-10-5 SBP 16-8-4 SBP 12-6-3
SBP 8-4-2 RG
Fig. 11. Low-accuracy cases during heart segmentation
4.2 Brain tumor segmentation
The brain tumor segmentation of the presented algorithm,
region growing algorithm, and the ground truth manual
segmentation is presented in Fig. 12. As shown, the region growing
algorithm has a problem related to the leakage through the border
gaps. In this case, the bone tissue is mistakenly segmented for the
three presented patients. Such properties of the image lead to low
accuracy of the region growing. In turn, SPB allows configuring
the size of superpixels to avoid oversegmentation and then Fig. 13. Brain tumor segmentation accuracy of the proposed
segmenting the tumor successfully. algorithm
Patient 1
Table 2. DSC obtained on the brain tumor dataset for different applied to data labeling. The most significant factor of the
algorithms. algorithm speed is the maximum square size and the sequence of
SBP 20-10-5 SBP 16-8-4 SBP 12-6-3 SBP 8-4-2 RG sizes of the square blocks in common. Increasing the size of the
largest square from the chosen sequence makes the image
0.89±0.07 0.88±0.08 0.88±0.08 0.87±0.09 0.86±0.10 processing faster. On the other hand, the sizes should be chosen in
the way that at least one square block is placed in the ROI.
In the case of the brain tumor segmentation, pseudo
proportionality between the DSC and the size of the squares is
observed. The latter means that the smaller the square size is, the 400
less the DSC value is. It should be noted that the reason for these 350
Average execution time, sec
leaks is not the borders defects. In this case, the tumor has
approximately the same level of intensity as external bone tissue. 300
35 31 33 250
30 27 200
Leakage cases
25 21 150
20 16 100
15
50
10
0
5
0
SBP 20-10-5 SBP 16-8-4 SBP 12-6-3
Image size
SBP 8-4-2 RG
Region growing
Fig. 14. Low-accuracy cases during brain tumor segmentation
Fig. 16. The average execution time of the RG algorithm for
4.3 Execution time testing different image and square block sizes
To compare the propagation speed of the region growing The proposed algorithm has opportunities to improve execution
algorithm and the proposed algorithm, a synthetic test image with time, robustness, and final accuracy. All squares are processed
a white circle in the center and the black background was generated. independently to each other, so the algorithm can be paralleled on
This test image was created in different sizes. The dependence GPU for minimizing execution time.
between processing time and image sizes for both algorithms is An important feature of the algorithm is its scalability. It means
represented in Fig. 15 and Fig. 16. As seen, both algorithms have that several different algorithms can be used for border detection at
asymptotic complexity O(n2) but the region growing algorithm is the same time. We used two criteria: one-dimensional region
much slower, and cannot be adapted for optimal speed. growing and intensity gradient check. As an additional method,
3,5 machine learning or one-dimensional neural networks can be
applied to the border detection. It should also be noted that the
Average execution time, sec
3,0 algorithm can be extended for three-dimensional imaging.
2,5 6. Acknowledgments
2,0 This work was supported in part by the Russian Federation
Governmental Program “Nauka” № 12.8205.2017/БЧ (additional
1,5 number: 4.1769.ГЗБ.2017).
1,0
7. References
0,5 [1] Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P. and
Süsstrunk, S. 2012. SLIC superpixels compared to state-
0,0
of-the-art superpixel methods. IEEE Transactions on
Pattern Analysis and Machine Intelligence. 34, 11 (2012),
2274–2281.
DOI:https://doi.org/10.1109/TPAMI.2012.120.
Image size [2] Adams, R. and Bischof, L. 1994. Seeded Region Growing.
IEEE Transactions on Pattern Analysis and Machine
SBP 20-10-5 SBP 16-8-4 SBP 12-6-3 SBP 8-4-2 Intelligence. 16, 6 (1994), 641–647.
DOI:https://doi.org/10.1109/34.295913.
Fig. 15. The average execution time of the SBP algorithm for [3] Andreopoulos, A. and Tsotsos, J.K. 2008. Efficient and
different sizes of images and square blocks generalizable statistical models of shape and appearance
for analysis of cardiac MRI. Medical Image Analysis. 12,
3 (Jun. 2008), 335–357.
5. Conclusion
DOI:https://doi.org/10.1016/j.media.2007.12.003.
The proposed algorithm is devoted to the segmentation of [4] Ballard, D.H. and Brown, C.M. 1982. Computer Vision.
images with high resolution or medical images with ROI border [5] Bankman, I.N. 2000. Handbook of Medical Imaging.
defects and low contrast. Additionally, this algorithm can be [6] Cheng, J., Huang, W., Cao, S., Yang, R., Yang, W., Yun,
Z., Wang, Z. and Feng, Q. 2015. Enhanced performance DOI:https://doi.org/10.1146/annurev.bioeng.2.1.315.
of brain tumor classification via tumor region [20] Pinto, A., Alves, V. and Silva, C.A. 2016. Brain Tumor
augmentation and partition. PLoS ONE. 10, 10 (2015). Segmentation using Convolutional Neural Networks in
DOI:https://doi.org/10.1371/journal.pone.0140381. MRI Images. IEEE Transactions on Medical Imaging. 35,
[7] Cheng, J., Yang, W., Huang, M., Huang, W., Jiang, J., 5 (2016), 1240–1251.
Zhou, Y., Yang, R., Zhao, J., Feng, Y., Feng, Q. and Chen, DOI:https://doi.org/10.1109/TMI.2016.2538465.
W. 2016. Retrieval of Brain Tumors by Adaptive Spatial [21] Ren, X. and Malik, J. 2003. Learning a classification
Pooling and Fisher Vector Representation. PLoS ONE. 11, model for segmentation. Proceedings Ninth IEEE
6 (2016). International Conference on Computer Vision. 1, c
DOI:https://doi.org/10.1371/journal.pone.0157112. (2003), 10–17 vol.1.
[8] Chiverton, J., Wells, K., Lewis, E., Chen, C., Podda, B. DOI:https://doi.org/10.1109/ICCV.2003.1238308.
and Johnson, D. 2007. Statistical morphological skull [22] Rogowska, J. 2009. Overview and fundamentals of
stripping of adult and infant MRI data. Computers in medical image segmentation. Handbook of Medical Image
Biology and Medicine. 37, 3 (2007), 342–357. Processing and Analysis. 73–90.
DOI:https://doi.org/10.1016/j.compbiomed.2006.04.001. [23] Roy, S. and Maji, P. 2015. A simple skull stripping
[9] Crommelinck, S., Bennett, R., Gerke, M., Koeva, M.N., algorithm for brain MRI. ICAPR 2015 - 2015 8th
Yang, M.Y. and Vosselman, G. 2017. SLIC Superpixels International Conference on Advances in Pattern
for Object Delineation from UAV Data. ISPRS Annals of Recognition (2015).
the Photogrammetry, Remote Sensing and Spatial [24] Saxen, F. and Al-Hamadi, A. 2014. Superpixels for Skin
Information Sciences (2017), 9–16. Segmentation. Www-E.Uni-Magdeburg.De. (2014).
[10] Csillik, O. 2017. Fast segmentation and classification of [25] Soliman, A., Khalifa, F., Elnakib, A., El-Ghar, M.A.,
very high resolution remote sensing data using SLIC Dunlap, N., Wang, B., Gimel’farb, G., Keynton, R. and
superpixels. Remote Sensing. 9, 3 (2017). El-Baz, A. 2017. Accurate lungs segmentation on CT
DOI:https://doi.org/10.3390/rs9030243. chest images by adaptive appearance-guided shape
[11] Dehmeshki, J., Amin, H., Valdivieso, M. and Ye, X. 2008. modeling. IEEE Transactions on Medical Imaging. 36, 1
Segmentation of pulmonary nodules in thoracic CT scans: (2017), 263–276.
A region growing approach. IEEE Transactions on DOI:https://doi.org/10.1109/TMI.2016.2606370.
Medical Imaging. 27, 4 (2008), 467–480. [26] Tang, J.T.J. 2010. A color image segmentation algorithm
DOI:https://doi.org/10.1109/TMI.2007.907555. based on region growing. Computer Engineering and
[12] Duong, T.H. and Hoberock, L.L. 2018. DUHO image Technology (ICCET), 2010 2nd International Conference
segmentation based on unseeded region growing on on. 6, (2010), 634–637.
superpixels. 2018 IEEE 8th Annual Computing and DOI:https://doi.org/10.1109/ICCET.2010.5486012.
Communication Workshop and Conference (CCWC) (Jan. [27] Tsechpenakis, X.H.G. 2013. Medical Image
2018), 558–563. Segmentation. Advanced Materials Research. i (2013), 1–
[13] Isa, N.A.M., Sabarudin, S., Ngah, U.K. and Zamli, K.Z. 35. DOI:https://doi.org/10.1201/9781420090413-c10.
2005. Automatic detection of breast tumours from [28] Wang, L., Pei, M., Codella, N.C.F., Kochar, M., Weinsaft,
ultrasound images using the modified seed based region J.W., Li, J., Prince, M.R. and Wang, Y. 2015. Left
growing technique. ventricle: Fully automated segmentation based on
[14] Kamnitsas, K., Ledig, C., Newcombe, V.F.J., Simpson, spatiotemporal continuity and myocardium information in
J.P., Kane, A.D., Menon, D.K., Rueckert, D. and Glocker, cine cardiac magnetic resonance imaging (LV-FAST).
B. 2017. Efficient multi-scale 3D CNN with fully BioMed Research International. (2015).
connected CRF for accurate brain lesion segmentation. DOI:https://doi.org/10.1155/2015/367583.
Medical Image Analysis. 36, (2017), 61–78.
DOI:https://doi.org/10.1016/j.media.2016.10.004.
[15] Kang, H.C., Lee, J. and Shin, J. 2016. Automatic four-
chamber segmentation using level-set method and split
energy function. Healthcare Informatics Research. 22, 4
(2016), 285–292.
DOI:https://doi.org/10.4258/hir.2016.22.4.285.
[16] Litjens, G., Kooi, T., Bejnordi, B.E., Setio, A.A.A.,
Ciompi, F., Ghafoorian, M., van der Laak, J.A.W.M., van
Ginneken, B. and Sánchez, C.I. 2017. A survey on deep
learning in medical image analysis. Medical Image
Analysis.
[17] Oktay, O., Ferrante, E., Kamnitsas, K., Heinrich, M., Bai,
W., Caballero, J., Cook, S.A., De Marvao, A., Dawes, T.,
O’Regan, D.P., Kainz, B., Glocker, B. and Rueckert, D.
2018. Anatomically Constrained Neural Networks
(ACNNs): Application to Cardiac Image Enhancement
and Segmentation. IEEE Transactions on Medical
Imaging. 37, 2 (2018), 384–395.
DOI:https://doi.org/10.1109/TMI.2017.2743464.
[18] Park, J.G. and Lee, C. 2009. Skull stripping based on
region growing for magnetic resonance brain images.
NeuroImage. 47, 4 (2009), 1394–1407.
DOI:https://doi.org/10.1016/j.neuroimage.2009.04.047.
[19] Pham, D.L., Xu, C. and Prince, J.L. 2000. Current
Methods in Medical Image Segmentation. Annual Review
of Biomedical Engineering. 2, 1 (2000), 315–337.