Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2018) 000–000

ScienceDirect www.elsevier.com/locate/procedia

Procedia Computer Science 120 (2017) 580–587

9th
9thInternational
InternationalConference
Conferenceon
onTheory
Theoryand
andApplication
Applicationof
ofSoft
SoftComputing,
Computing,Computing
Computingwith
with
Words and Perception, ICSCCW 2017, 24-25
22-23 August
August 2017,
2017, Budapest,
Budapest, Hungary
Hungary

Brain tumor segmentation based on a new threshold approach

Umit Ilhana*, Ahmet Ilhana
a
Department of Computer Engineering, Near East University, P.O.BOX:99138, Nicosia, North Cyprus, Mersin 10, Turkey

Abstract

Brain cancer is an abnormal cell population that occurs in the brain. Nowadays, medical imaging techniques play an important
role in cancer diagnosis. Magnetic resonance imaging (MRI) is one of the most used techniques to identify and locate the tumor
in the brain. Images obtained by medical imaging techniques may become a better quality image thru applying image processing
techniques. In this study, we aim to develop a method for clearly distinguishing the tissues affected by the cancer. The proposed
approach is used to obtain a segmented tumor region clear enough to be observed by the medical practitioner and give them more
detail about the tumor in their diagnosis. In the proposed approach, morphological operations, pixel subtraction, threshold based
segmentation and image filtering techniques are used. The proposed approach is based on obtaining clear images of the skull,
brain and the tumor. When compared, the proposed approach gave a better result than the other approach.

© 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

Keywords: Brain cancer; image processing; threshold; MRI; segmentation.

1. Introduction

Cancer, which is increasing in proportion in the aging population, has become a worldwide health problem. Based
on the latest statistics of the World Cancer Research Fund, cancer is the deadliest disease in the world.
Approximately 12.7 million people are diagnosed with cancer each year and 7.6 million people die from this disease
(Gao et al., 2009). Every year the number of people die from cancer is increasing.
There are many subtypes of each cancer and it is difficult to give detailed information about most of them. Brain
cancer is the fifth cancer type with the highest incidence and mortality, just behind stomach cancer, uterine cancer,

*Ümit İlhan. Tel: +90 392 64 64; Fax: +903922236622.

E-mail address: umit.ilhan@neu.edu.tr

1877-0509© 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of Soft
Computing, Computing with Words and Perception.
10.1016/j.procs.2017.11.282
 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587 581
2 Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000

breast cancer and esophageal cancer (Gao et al., 2009). Based on the statistics of World Health Organization
(WHO), brain tumors have more than 120 varieties. Brain tumors are categorized according to the region, tissue
type, non-cancerous or cancerous (benign or malignant) cells, site of originate (primary or secondary) and other
important factors (Lois et al., 2007).
For diagnosing of brain tumors, medical imaging techniques such as Computed Tomography (CT), Positron
Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) are used (Ozols et al., 2006). CT uses the
radioactive rays to penetrate the human body, and the imaging is based on the different characteristics reflecting the
rays of different tissues. PET, injects with radioactive drugs to the human body, and the drugs flow to all the cells,
tissues and organs with the blood in the whole body (Prince and Links, 2006).
In the MRI process, there are neither instruments inserted nor any medication injected into the human body.
There is no radiation damage to the human body, and the whole process is quite safe. In addition, MRI has high-
resolution and accurate positioning of soft tissues and is sensitive to the characteristics of diseases, thus it is
especially suitable for the diagnosis of brain diseases (Xuan and Liao, 2007).
In addition to medical imaging techniques, tools that have become popular in the medical field also play an
important role in the diagnosis of diseases. Image processing is one of the most used and most efficient tools in the
medical field. Medical image processing is aimed at developing systems to solve the medical diagnosis problems
using computerized systems (Mishra et al., 2014).
Medical image processing can be carried out in four stages. The first stage is to use a set of images collected from
the patients to be investigated. The second stage is to apply image enhancement techniques to obtained better quality
images. The third stage is to apply image segmentation methods which are the most important part of image
processing procedures. The last stage is to extract the necessary features from enhanced and segmented images
which give important information of normality or abnormality of images (Mishra et al., 2014).
Taheri et. al (2010) presented a threshold-based approach called TLS for 3D tumor segmentation in brain MRI
images. Ali et. al (2014) used Enhanced Thresholding Algorithm to area calculation of tumor. Vijayarangan (2014)
represented a method for the Brain Tumor Detection in MRI images using segmentation and histogram thresholding.
Anandgaonkar and Sable (2014) implemented an algorithm for detection and identification of brain tumor in brain
MRI, based on Fuzzy C-Means clustering to divide the MRI in number of clusters and adaptive thresholding to
extract the tumor then deciding the type of tumor on its area. Sujan et. al (2016) presented a method based on
thresholding along with morphological image analysis techniques to detect brain tumor from MRI image. They
convert the MRI image to grayscale and then remove noises with different filtering techniques then convert this new
image to binary with the Otsu's threshold method for segmentation and afterwards morphological operations were
performed to detect the tumor that contains the brightest part of the image.
The aim of this study is to develop a new threshold method that is more efficient and easier to implement than the
existing threshold methods. The proposed threshold method is compared with the OTSU which is one of the most
popular threshold methods. The rest of the study is organized as follows. Section 2 presents the proposed approach,
Section 3 describes the methodology, Section 4 presents the results and the conclusions addressed in the final
section.

2. Proposed approach

In this section, the proposed approach is described. The proposed approach can be summarized in three stages.
First stage is the pre-processing which enhances the brain MRI image and make it more suitable to analyze. Second
stage is the segmentation which separates the region of the tumor from the enhanced image. Last stage is the
filtering which removes noise from the segmented image.
582 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587
Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000 3

2.1. Pre-processing stage

In this stage, the techniques used to enhance the images are described. These techniques are morphological
operations and pixel subtraction operations.

2.1.1. Morphological operations

Morphological operations are used to identify the boundaries and skeleton of objects in an image. Morphological
operations create a window called a structure element on an image (Gonzalez and Woods, 2008). This window is
placed in all possible locations on the image and the corresponding pixel is compared with the neighborhood. The
most commonly morphological operations are dilation and erosion.
Dilation is a morphological operation that can be applied binary and grayscale images. Dilation operation grows
objects by expanding the boundaries (Radha and Lakshman, 2013). With the effect of this operation, the small holes
in the regions that begin to expand become even smaller. Dilation is defined as a set operation. A is dilated by B,
written as A ⊕ B, is defined as:

 � )   }
A  B {z | ( B (1)
z

� is for the reflection of collection B.

Among them, ∅ is for the empty set, B is for the structure element, and B
Erosion is a morphological operation that can be applied to both binary and grayscale images like dilation.
Erosion operation erodes the boundaries of objects (Radha and Lakshman, 2013). With the effect of this operation,
the size of the pixels shrinks and the size of the holes starts to grow. Erosion is defined as a set operation. A is
eroded by B, written as A ⊖ B, is defined as:

A
 ! B {z |(B)z  Ac   } (2)

Among them, ∅ is for the empty set, B is for the structure element, and Ac is for the supplement of collection A.
The combination of the two main operations, dilation and erosion, can produce more complex sequences. The
opening operation is one of the most popular of these sequences. An opening operation is defined as erosion and
then dilation using the same structuring element (Raid et al., 2014). Opening is defined as a set operation. Using the
structure element B to do the opening operation on the set A, written as A ∘ B, defined as:

A  B ( A ! B)  B
 (3)

In this study, the skull (bright region) of the image is extracted using an opening operation with the structure
element “disk” and radius “7”.
When the skull is removed, some of the pixels are also removed from the tumor region unintentionally.
Therefore, opened image cannot be used directly for segmentation of the tumor. The opening operation on its own is
not enough to remove the skull properly. Pixel subtraction is used to extract the skull from the brain image in a more
efficient manner.

2.1.2. Pixel subtraction

The pixel subtraction operator takes as input two images of the same size. The pixel values of the second image
are subtracted from the respective pixel values in the first image and a third image is obtained at the end of this
process (Qidwai and Chen, 2009). The pixel subtraction operation can be defined as:

R
(m, n) P(m, n)  Q(m, n) (4)
 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587 583
4 Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000

In this study, the following pixel subtraction operations are conducted. First, the image obtained from opening
operation is subtracted from the original image which seperates the skull from the original image. Then image
obtained from opening operation is subtracted from the skull image which removes the undesierd gray pixels from
the skull image. Lastly, the cleaned skull image is subtracted from the original image. The result of the pixel
subtraction operations, the image of the brain is obtained without the skull.

2.2. Segmentation stage

In this stage, working principle of the threshold segmentation technique and the proposed threshold method are
described.

2.2.1. Threshold based segmentation

Threshold technique is one of the important techniques in image segmentation. This technique can be described
as:

T = T[x, y, p(x, y), f(x, y)] (5)

T is the threshold value. x and y are the coordinates of the threshold value. p(x,y) and f(x,y) are the gray level
pixels of the image (Senthilkumaran and Vaithegi, 2016). The threshold image g(x,y) can be defined as:

1 if f(x, y) > 1 
g(x, y)    (6)
0 if f(x, y)  0

2.2.2. Proposed threshold method

In this method, sum of unique pixel values excluding zeros (black pixels) are divided by the count of unique pixel
values. By this operation, the average gray value (threshold value) is calculated to convert the grayscale image to
binary image. This method can be described as:

n
( (a,b)  S)(a  b  0 )  (a,b)
T 
S 1
n
(7)

T is the threshold value. a and b are pixel values. n is the count of unique pixel values.

2.3. Filtering stage

The purpose of image filtering is to remove the noise on the digital images. Image filtering is one of the most
common issues in digital image processing (Gonzalez and Woods, 2008). One of the most popular methods for
image filtering is the median filter.

2.3.1. Median filter

The median filter is the most commonly used non-linear filter. In this filter, the median pixel value in the
neighborhood is calculated and the middle pixel value in the neighborhood is replaced by the calculated median
pixel value (Singh and Neeru, 2014). While calculating the median pixel value, all the pixel values in the
neighborhood are first sorted in ascending order, and then the pixel values are evaluated together with the middle
pixel value. The median filter can be represented by the following equation:
584 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587
Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000 5

f(x, y) = median{g(s,t)} where (x, y)  S xy (8)

Sxy is the set of coordinates in a rectangular subimage window which has center at (x,y). The median filter
calculates the median of the corrupted image g(x,y) under the area Sxy. f(x,y) represents the restored image. In this
study, 5x5 neighborhood pixels around the evaluated pixel are used.

3. Methodology

3.1. Database

The MRI images obtained from TCIA are used to evaluate the accuracy of the proposed approach (The Cancer
Imaging Archive, 2017). There are 100 images in the database which contain 70 abnormal and 30 normal (with and
without tumor) brain MRI images. All images are 256 x 256 pixels, 8–bit grayscale and in JPEG format.

3.2. System performance

In this study, all images in the database have been tested. To evaluate the system accuracy, sensitivity and
specificity the following attributes are used in the calculations.

 TP (True Positive): Existing tumor and detected correctly.

 TN (True Negative): Non-existing tumor and not detected.
 FP (False Positive): Non-existing tumor and detected.
 FN (False Negative): Existing tumor and not detected.

Sensitivity is the measure of successful determination of the person having a tumor. The sensitivity is given by:

TP
Sensitivity (%)  *100 (9)
TP  FN

Specificity is the measure of successful determination of the person not having a tumor. The specificity is given
by:

TN
Specificity (%)  *100 (10)
TN  FP

Accuracy is the measure of successful classification. The accuracy is given by:

TP  TN
Accuracy(%)  *100 (11)
TP  TN  FP  FN

4. Experimental results

In this section, the simulation of the proposed approach is shown. Figure 1 shows the result of the image opening
operation. Figure 2,3 and 4 shows the implementation of the pixel subtraction operations respectively. Figure 5
shows tumor segmentation by the proposed threshold method. Figure 6 shows the filtering of the segmented image.
 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587 585
6 Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000

Fig. 1. (a) original image; (b) opened image.

Fig. 2. (a) original image; (b) opened image; (c) skull image.

Fig. 3. (a) skull image; (b) opened image; (c) cleaned skull image.

Fig. 4. (a) original image; (b) cleaned skull image; (c) skull removed image.
586 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587
Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000 7

Fig. 5. (a) skull removed image; (b) segmented image.

Fig. 6. (a) segmented image; (b) filtered image.

Table 1 shows the performance of the proposed system which is calculated in system performance stage.

Table 1. Performance of the proposed system.

Number of total images TP TN FP FN Sensitivity Specificity Accuracy
100 66 30 0 4 94.28% 100% 96%

Out of 70 images with the tumor 66 images evaulated correctly by the system. Other 4 images are identified as
without a tumor incorrectly. The reason of this mulfunctioning was due to the fact that there is no clear distinction
between the tumor area and the rest of the brain in terms of intensity of the pixels. 30 images without a tumor was
correctly identified by the system as images without a tumor with a 100% success. The proposed approach gave a
better result than the compared approach. Table 2 shows the result of the comparison.

Table 2. The comparison of the proposed system.

Approach Method Accuracy

Sujan et al. (2016) OTSU’s threshold 84.72%

Proposed Proposed threshold 96%

5. Conclusion

The system developed in this study is an aid for the medical people to diagnose the brain cancer using MRI
images. The images obtained from The Cancer Imaging Archive (TCIA) are used in this study. The proposed
approach has 94.28% recognition rate on the images with a tumor and 100% recognition rate on the ones without a
tumor. The overall success rate of our system is 96% which shows a better performance in comparison. For the
future work, there are two issues that are to be improved in our system. First one is the inaccurately classified
images that have a tumor. Here maybe different methods of classification like neuro-fuzzy or support vector
 Umit Ilhan et al. / Procedia Computer Science 120 (2017) 580–587 587
8 Umit Ilhan et al./ Procedia Computer Science 00 (2018) 000–000

machine should be tried. The Second is that the proposed method is not an automatic procedure. These technologies
may be brought together in a single processing environment.

References

Ali, A. H., Khalaph, K. A., Nema, I. S., 2014. IOSR Journal of Research & Method in Education (IOSR-JRME) 4 (1.-II). 58-62.
Anandgaonkar, G., Sable, G., 2014. Brain tumor detection and identification from T1 post contrast MR images using cluster based segmentation.
International Journal of Science and Research, 3(4), 814-7.
Gao, Q., Wang, X. Y., Qiu, S. J., Yamato, I., Sho, M., Nakajima, Y., Xu, Y., 2009. Overexpression of PD-L1 significantly associates with tumor
aggressiveness and post operative recurrence in human hepatocellular carcinoma. Clinical Cancer Research, 15(3), 971-979.
Gonzalez, R. C.,Woods, R. E., 2008. Digital image processing. Upper Saddle River, New Jersey.
Louis, D. N., Ohgaki, H., Wiestler, O. D., Cavenee, W. K., Burger, P. C., Jouvet, A., Kleihues, P., 2007. The 2007 WHO classification of tumours
of the central nervous system. Actaneuro pathologica, 114(2), 97-109.
Mishra, A., Rai, A., Yadav, A., 2014. Medical Image Processing: A Challenging Analysis. International Journal of Bio-Science and Bio-
Technology, 6(2), 187-194.
Ozols, R. F., Herbst, R. S., Colson, Y. L., Gralow, J., Bonner, J., Curran, W. J., Markman, M., 2006. Major research advances in cancer treatment,
prevention, and screening report from the American Society of Clinical Oncology. Journal of Clinical Oncology, 25(1), 146-162.
Prince, J. L., Links, J. M., 2006. Medical imaging signals and systems. Upper Saddle River, New Jersey.
Qidwai, U., Chen, C.H., 2009. Digital image processing: An algorithmic approach with MATLAB. Chapman & Hall/CRC, Boca Raton.
Radha, R., Lakshman, B., 2013. Retinal image analysis using morphological process and clustering technique. International Journal of Signal and
Image Processing, 4(6), 55-32.
Raid, A., Khedr, W., El-dosuky, M., Aoud, M., 2014. Image Restoration Based on Morphological Operations. International Journal of Computer
Science, Engineering and Information Technology, 4(3), 9-21.
Senthilkumaran, N.,Vaithegi, S., 2016. Image Segmentation By Using Thresholding Techniques For Medical Images. Computer Science &
Engineering: An International Journal, 6(1), 1-13.
Singh, I., Neeru, N., 2014. Performance Comparison of Various Image Denoising Filters Under Spatial Domain. International Journal of
Computer Applications, 96(19), 21-30.
Sujan, M., Alam, N., Abdullah, S., Jahirul, M., 2016. A Segmentation based Automated System for Brain Tumor Detection. International Journal
of Computer Applications, 153(10), 41-49.
Taheri, S., Ong, S. H., Chong, V.F.H., 2010. Level-set segmentation of brain tumors using a threshold-based speed function. Image and Vision
Computing, 28(1), 26–37.
The Cancer Imaging Archive, 2017. Rembrandt. https://wiki.cancerimagingarchivefrom.net/display/Public/REMBRANDT;jsessionid=D8DC3A
DCD 6110CF52B45F4E7B1020F0E (accessed 17.03.05).
Vijayarangan V., 2014. Multiscale Modeling For Image Analysis of Brain Tumor Detection and Segmentation Using Histogram Thresholding.
International Journal of Engineering and Computer Science, 3(8), 7525-7534.
Xuan, X., Liao, Q., 2007. Statistical structure analysis in MRI brain tumor segmentation. Fourth International Conference on Image and
Graphics. Sichuan, China. 421-426.