A SIMPLE IMAGE PROCESSING APPROACH TO ABNORMAL SLICES DETECTION FROM MRI TUMOR VOLUMES

The International Journal of Multimedia & Its Applications (IJMA) Vol.8, No.1, February 2016
DOI : 10.5121/ijma.2016.8105 55
A SIMPLE IMAGE PROCESSING APPROACH TO
ABNORMAL SLICES DETECTION FROM MRI
TUMOR VOLUMES
T.Kalaiselvi, P.Nagaraja and P.Sriramakrishnan
Department of Computer Science and Applications,
Gandhigram Rural Institute – Deemed University, Gandhigram, Tamil Nadu, India
ABSTRACT
This paper proposed a method for brain tumor detection from the magnetic resonance imaging (MRI) of
human head scans. The proposed work explained the tumor detection process by means of image
processing transformations and thresholding technique. The MRI images are preprocessed by
transformation techniques and thus enhance the tumor region. Then the images are checked for
abnormality using fuzzy symmetric measure (FSM). If abnormal, then Otsu’s thresholding is used to extract
the tumor region. Experiments with the proposed method were done on 17 datasets. Various evaluation
parameters were used to validate the proposed method. The predictive accuracy (PA) and dice coefficient
(DC) values of proposed method reached maximum.
KEYWORDS
Minima Transform, Thresholding, Transforms, Tumor.
1. INTRODUCTION
Brain tumor is cluster of abnormal and an uncontrolled growth of cells in the brain [1]. Brain
tumors can be classified according to their origin or degree of aggressiveness. Primary brain
tumors arise in the brain, while metastatic brain tumors frequently originate from other parts of
the body [2]. Magnetic resonance imaging (MRI) plays important role in many medical imaging
applications. MRI provides prosperous information about the human soft tissue anatomy as well
as helps to diagnosis of brain tumor [3]. In recent years, MRI has become an important modality
for neurological image diagnosis. The combination of different sequences of MRI techniques is
used to diagnose tumor [4]. The sequences include T1-weighted, T1-weighted with contrast
enhancement (T1c), T2-weighted and fluid attenuated inversion recovery (FLAIR). Nowadays,
brain tumor detection for MRI is difficult task for medical applications [5].
In recent years, many approaches have been developed for brain tumor detection. Jayachandran
and Dhanasekaran proposed a hybrid algorithm for detection brain tumor in MRI images using
statistical features and Fuzzy Support Vector Machine (FSVM) classifier. The proposed
technique consists of four stages namely, Noise reduction, Feature extraction, Feature reduction
and Classification. Stage I, anisotropic filter is applied for noise reduction and extracting the
features. Stage II, obtains the texture features related to MRI images. Stage III, the features of
magnetic resonance images have been reduced using principles component analysis to the most
essential features. Stage IV, the supervisor classifier based FSVM has been used to classify
subjects as normal and abnormal brain MR images [6]. Somasundaram and Kalaiselvi proposed
an automatic method to analyze the MRI head scans and detect abnormality in brain due to
tumors. This method consist four stages: brain extraction algorithm, transformation, fuzzy
segmentation and fuzzy symmetric analysis. This method used two measures: false alarm (FA)

56
and missed alarm (MA) to quantify the performance of the method. The mean FA no more
detected, however the MA was detecting minimum quantitative value [7]. Logeswari and karnan
proposed an enhanced implementation of brain tumor detection using segmentation based on soft
computing. The proposed method used two phased for detection tumor. In first phase MRI image
of brain is collected. After that using preprocessing technique image is converted into standard
form. Second phase for image segmentation using hierarchical self organizing map (HSOM)
method is applied on image [8].
Roy and Bandyopadhyay proposed an interactive segmentation method that enables users can
quickly and efficiently segment tumors in MRI of brain. In addition to area of the region and
edge information the proposed method uses a type of prior information also its detecting the
tumor region on exactly from MRI [9]. Padole and Chaudhari proposed an efficient method for
brain tumor detection. Combination of two standard algorithm, mean shift and normalized cut is
performed to detect the brain tumor surface area in MRI. Segmentation of brain, detects tumor
and also its physical dimension and its segmentation accuracy is discussed [10]. Anandgaonkar
and Sable proposed a survey on different segmentation techniques applied to MR images for
locating tumor. It also includes a proposed method for the same using Fuzzy C-Means algorithm
and an algorithm to find area of tumor which is useful to decide type of brain tumor whether it is
benign or malignant [11].
The proposed work is an automatic method for extraction of the complete tumor region which
overcomes the above said problems and works efficiently for FLAIR and T2-weighted images.
The proposed method introduces enhancing process using top and bottom hat transformations and
minima transform. Initially MRI head scans are enhanced by using top and bottom-hat process.
Extended minima transform is used to separate tumor region from the enhanced image. In
resultant segmentation, the fuzzy symmetric measure (FSM) is used to check the abnormality
detection. Finally applied Otsu’s thresholding in abnormal slice, it extracts the tumor region. The
proposed method gives better results for detecting the tumor regions from FLAIR and T2-
weighted images.
This paper is organized as follows. The image processing transformations are explained in section
2, the Otsu’s thresholding is explained in section 3, the proposed method is explained in section 4,
the results and discussion are given in section 5 and the conclusion is given in section 6.
2. IMAGE PROCESSING TRANSFORMATIONS
2.1. Top-hat and Bottom-hat Transforms
The top-hat transform is defined as the difference between the input image and its opening by
some structuring element [12]. Structuring element (SE) of proposed method is given by,
SE= (‘disk’, 15) (1)
The application of these transforms is in removing objects from an image by using an SE in the
opening and closing that does not fit the objects to be removed. The difference then yields an
image with only the removed objects.
Then, the top-hat transform of f is given by:
(2)

57
where ○ denotes the opening operation. The top-hat transform is an operation that extracts small
elements and details from given images. The top-hat transform performs light objects on a dark
background.
The bottom-hat transform is defined dually as the difference between the closing and the input
image. Then, the bottom-hat transform of f is given by:
(3)
where ● is the closing operation. Bottom-hat performs morphological bottom-hat filtering on the
grayscale or binary input image. The bottom-hat performs for dark objects on a light background.
2.2. Extended Minima Transform
The extended minima transform, which is the regional minima of the H-minima transform.
Extended minima used 8-connected neighborhoods for 2-D images and 26-connected
neighborhoods for 3-D image. Performing the H-minima transform on the inverse distance image
can effectively decrease over segmentation is reduced to some extend after applying
morphological filters. The H-minima transform [13] is performed by:
)()( hfRfH fh +=
(4)
where h represents the given depth. R represents the reconstruction and erosion operators,
respectively.
3. Otsu’s Method
This method is called as optimum threshold method and provides satisfied results in MRI
brain images [14]. Otsu’s thresholding involves all possible threshold values and
calculate the pixel levels in each side of the threshold. This threshold value separates the
foreground or background of pixels. This algorithm compute the image to be threshold
contains the two classes of pixels. We can use the within class variance, it is the weighted
sum of the variances of each foreground and background [15].
)()()()()( 222
TTWTTWT fgFbgBwithin σσσ += (5)
Where,
∑∑
−
=
=
=
==
11
0
)()()()(
N
TiF
T
iB ipTWipTW (6)
pi – is the probability of occurring of pixel value xi.
The mean of foreground and background pixels is,
∑=
=
T
iB
B iiP
TW
T
1
)(
)(
1
)(µ
(7)

58
∑=
=
T
iF
F iiP
TW
T
1
)(
)(
1
)(µ
(8)
The variance of foreground and background pixels is,
∑=
−=
T
i
B
B
bg iPTi
TW
T
1
22
)())((
)(
1
)( µσ
(9)
∑=
−=
T
i
F
F
fg iPTi
TW
T
1
22
)())((
)(
1
)( µσ
(10)
)(2
Tbgσ -variance of the pixels in the background.
)(2
Tfgσ -variance of the pixels in the foreground.
4. PROPOSED METHOD
This proposed method is a fully automated brain tumor detection method. The enhanced MRI
images are produced by top, bottom-hat, minima transformations and segmented by Otsu’s
thresholding to extract the tumor portion from both T2-weighted and FLAIR images. The flow
chart of tumor extraction methods are shown in Figure 1.
Figure 1. Flow chart for proposed method

59
4.1. Pre-processing
The proposed method initially has done a pre-processing technique on MRI images. The flow
diagram of preprocessing method is given in Figure 2.
Figure 2. Flow chart for preprocessing
The datasets are processed by top-hat and bottom-hat transformation for enhancing the original
image. The preprocessing images are shown in Figure 2. The original image is shown in column
1. The top-hat transformation as shown in column 2 is used to maximize the contrast between the
objects and the gaps that separate them from each other. The top-hat transform is defined dually
as the difference between the opening and the input image. Bottom-hat performs morphological
bottom-hat filtering on the grayscale or binary input image and performs dark objects on a light
background. The bottom-hat transform is defined dually as the difference between the closing and
the input image. The bottom-hat transform image is shown in column 3. Then add the top-hat
image to the original image, and then subtracts the bottom-hat image. This stage tumor area get
sharpened region and it is shown in column 4
Then complement the function is used to enhance the image based on the intensity valleys as
represented in column 5. The minima transformation is used to detect the intensity valleys deeper
than a particular threshold with that function. It removes local peaks which are lower than h
intensity values from the background. Based on the analysis done during our experiments, h is set
to 8. The enhanced tumor regions extracted by using extended minima transform is shown in
column 6. In column 6, the normal volume enhanced images are given in row 1.
Figure 3. Preprocessing Stage, original images for Normal, FLAIR and T2-weighted

60
Figure 3. Preprocessing Stage, original images for Normal, FLAIR and T2-weighted are in
column 1, top-hat images are in column 2, bottom-hat images are in column 3, enhanced images
are in column 4, complement images are in column 5 and minima transformation images are in
column 6.
4.2. Abnormality Detection
Abnormal tissues in the enhanced tumor region image can be detected by measuring the
symmetry of enhanced tumor image. Here the symmetry is computed by the fuzzy symmetric
measure (FSM) [16] given by,
(11)
where nL and nR are the number of foreground (white) pixels in the left and right half of the given
image present at either side of the central vertical line of slice. The symmetry values calculated
from normal enhanced image are generally much larger than 0.1, and the values for abnormal
enhanced image are much smaller than 0.1 [7] [17].
4.3. Brain Tumor Detection
Then the process is carried out by abnormal slice. The tumor extraction method is used by Otsu’s
thresholding technique. It separates the tumor region from background. In the resultant detection,
tumors are extract from the abnormal slice and as a largest connected region as shown in Figure
4.
Figure 4. (a) FLAIR tumor region, (b) T2-weighted tumor region.
5. RESULTS AND DISCUSSION
Our algorithm was implemented in MATLAB2009 on a PC with Intel Pentium Core Duo 1.6GHz
processor and 512MB RAM. Our proposed method used 17 datasets from Brain Tumor Image
Repository (BTIR) maintained by our research group [18]. The first two datasets (N01–N02)
consisting of 2 normal volumes and V01-V15 are abnormal volumes with glioma brain tumor. In
tumor portion, two types of outputs from FLAIR and T2-weighted were added and its represents
to the complete tumor region is as shown in Figure 5.The qualitative validation in the form of
visual inspection is done with some of the sample MRI brain images shown in Figure 6. In Figure
6, the original FLAIR images are given in column 1, the original T2-weighted MRI images are
given in column 2, the corresponding ground truth images are given in column 3, and the results
of proposed method are given in column 4. This proposed work gives good results for glioma
images.

61
Figure 5. Complete tumor region.
For quantitative validation, the performance is checked against three parameters. They are
predictive accuracy (PA), dice coefficient (DC). The segmented images could have an error rate
and an instance may either fail to identify an abnormality or identify an abnormality when there is
none. The parameter PA is used to describe the error rate by the terms true and false positive (TP,
FP) and true and false negative (TN, FN).
(11)
where,
TP = the test result is positive in the abnormal cases correctly classified.
TN= the test result is negative in the normal cases correctly classified.
FP = the test result is positive in the normal cases classified abnormal.
FN= the test result is negative in the abnormal cases classified normal.
The parameter DC is used to compare the similarity between manual segmentation and the result
of the proposed work. The value for DC ranges from 0 to 1 where 0 for no agreement and 1 for
exact agreement.
The DC is given by:
(12)
where A represents the ground truth image and B represents the proposed result image.
Table 1 is shown the abnormal detection by evaluated parameters FA and MA. The missed alarm
is 2.3 % from the proposed method and no more FA in entire datasets. The performance analysis
of proposed method is based on the parameters PA and DC values given in Table 2. The graph
shows the quantitative representation of PA and DC of proposed method with line representation.
In Figure 7, X axis represents volume id considered for experiment and Y axis represents the PA
and DC values proposed method. The PA and DC of proposed method are 98% and 75% in
glioma images. Our proposed method gives good results for glioma images.
Figure 6. FLAIR images are in column 1, T2-weighted images are in column 2, corresponding ground truth
images are in column 3 and the results of proposed method are in column 4.
Table 1. FA and MA values for proposed method

62
Table 2. PA and DC values of the proposed method
S.No Vol. id
PA (%)
(Complete
Tumor)
DC (%)
(Complete
Tumor)
1 V01 98 84
2 V02 97 73
3 V03 97 87
4 V04 98 85
5 V05 98 60
6 V06 98 77
7 V07 98 62
8 V08 98 81
9 V09 96 69
10 V10 99 61
11 V11 99 92
12 V12 97 75
13 V13 98 73
14 V14 98 74
15 V15 99 77
Mean 97.9 75.3
No
Vol.
id
Tota
l
Slice
Abnormal Slice FA MA
Actual Detected Slices % Slices %
1 N01 54 - - 0 0 0 0
- - 43 - - 0 0 0 0
3 V01 176 46-119 50 - 119 0 0 46-49 2
4 V02 176 44-110 47-106 0 0 44-46, 107-110 4
5 V03 176 62-146 67-141 0 0 62-66, 142-146 6
6 V04 176 81-150 81-145 0 0 146-150 3
7 V05 176 68-131 69-131 0 0 68 0.6
8 V06 176 87-161 87-160 0 0 161 0.6
9 V07 176 78-137 81-135 0 0 78-80, 136,137 3
10 V08 176 55-133 60-133 0 0 55-59 3
11 V09 176 75-147 77-145 0 0 75,76, 146,147 2
12 V10 176 77-117 80-115 0 0 77-79, 116,117 3
13 V11 230 118-167 118-166 0 0 167 0.6
14 V12 165 78-119 78 - 119 0 0 0 0.4
15 V13 220 81-132 85-127 0 0 81-84, 127-132 5
16 V14 230 94-120 96-114 0 0 95-96, 115-120 2
17 V15 163 61-101 61-101 0 0 0 0
Average Performance 0 0 2.3

63
Figure 7. PA and DC measurement Graph for the proposed method
6. CONCLUSIONS
The proposed work developed a brain tumor detection method using top-hat, bottom-hat, minima
transformations and Otsu’s thresholding. The experimental method is done on 17 datasets. This
work detected the abnormal portion of the brain which is present in MRI images. The advantage
of this proposed method is it takes minimum missed alarm while detecting the tumor in MRI
images. It gives 75% similarity index while compared with ground truth dataset.
REFERENCES
[1] Somasundaram, K. & Kalaiselvi, T., (2010) “Automatic detection of brain tumor from MRI scans
using maxima transform”, National Conference on Image Processing (NCIMP), pp136-141.
[2] Kalaiselvi, T. & Nagaraja, P., (2015) “Brain Tumor Segmentation of MRI Brain Images through FCM
clustering and Seeded Region Growing Technique”, International Journal of Applied Engineering
Research, vol.10, no.76, pp427-432.
[3] Kalaiselvi, T., (2011) Brain Portion Extraction and Brain Abnormality Detection from Magnetic
Resonance Imaging of Human Head Scans, Pallavi Publications.
[4] Drevelegas, A. & Papanikolaou, (2011) “Imaging modalities in brain tumors”, Imaging of Brain
Tumors with Histological Correlations, chapter 2, pp 13–34.
[5] Kalaiselvi, T. & Nagaraja, P., (2015) “A Rapid Automatic Brain Tumor Detection Method for MRI
Images using Modified Minimum Error Thresholding Technique” International Journal of Imaging
Systems and Technology, Vol.25, No.1, pp77–85.
[6] Jayachandran, A. & Dhanasekaran, R., (2013) “Brain Tumor Detection and Classification of MR
Images Using Texture Features and Fuzzy SVM Classifier”, Research Journal of Applied Sciences,
Engineering and Technology, Vol. 6, No.12, pp2264-2269.
[7] Somasundaram, K. & Kalaiselvi, T., (2010) “Fully Automatic Method to Identify Abnormal MRI
Head Scans using Fuzzy Segmentation and Fuzzy Symmetric Measure”, ICGST International
Journal on Graphics, Vision and Image Processing (GVIP), Vol.10, No.3, pp1-9.
[8] Logeswari, T. & Karnan, M., (2010)”An Enhanced Implementation of Brain Tumor Detection Using
Segmentation Based on Soft Computing”, International Conference on Signal Acquisition and
Processing
[9] Roy, S. & Bandyopadhyay S.K., (2012) “Detection and Quantification of Brain Tumor from MRI of
Brain and its Symmetric Analysis”, International Journal of Information and Communication
Technology Research, Vol.22.
[10] Padole, V.B & Chaudhari, D.S., (2012) “Detection of Brain Tumor in MRI Images Using Mean Shift
Algorithm and Normalized Cut Method”, International Journal of Engineering and Advanced
Technology.

64
[11] Anandgaonkar, G.P., and Sable, G.S., “Detection and Identification of Brain Tumor in Brain MR
Images Using Fuzzy C-Means Segmentation”, International Journal of Advanced Research in
Computer and Communication Engineering, vol.2, no.10, 2013, pp-3964-3967.
[12] Thapar, S. & Garg, S., (2012) “Study and Implementation of Various Morphology Based Image
Contrast Enhancement Techniques”, International Journal of Computing Business Research, pp2229-
6166.
[13] Jung, C. & Kim, C., (2010) “Segmenting Clustered Nuclei Using H-minima Transform-Based Marker
Extraction and Contour Parameterization”, IEEE Transactions on Biomedical Engineering, Vol.57,
No.10, pp2600-2604.
[14] Kalaiselvi, T. & Nagaraja. P., (2013), "A Robust Thresholding Technique for Image Segmentation
from Gray Images." in Proceedings of the International Conference on Applied Mathematics and
Theoretical Computer Science, p. 183-188.
[15] Otsu, N., (1979) “A Threshold Selection from Gray level Histograms,” IEEE Transactions of systems,
Man and Cybernetics (SMC), Vol. 9, No.1, pp62-66.
[16] Zimmermann, H. J., (1991) Fuzzy Set Theory and its Applications, Kluwer Academic Publishers,
second edition, Boston, MA.
[17] Clark, M.C., Hall, L.O., Goldgof, D.B., Velthuizen, R., Murtagh, F.R. & Silbiger, M.S., (1998)
“Automatic Tumor Segmentation using Knowledge-Based Technique”, IEEE Transactions on
Medical Imaging, Vol. 17, No. 2, pp187–201.
[18] Kalaiselvi, T., "Brain Tumor Image Repository (BTIR)", Department of Computer Science and
Applications, Gandhigram Rural Institute- Deemed University, Gandhigram.
Authors
Kalaiselvi T. currently working as an Assistant Professor in Department of
Computer Science and Applications, Gandhigram Rural Institute - Deemed
University, Dindigul, India. She received her Bachelor of Science (BSc) degree in
Mathematics and Physics in the year 1994 and Master of Computer Applications
(MCA) degree in the year 1997 from Avinashilingem University, Coimbatore. She
received her Ph.D (Full-time) degree from Gandhigram Rural Institute in the year
2010. In the year 2008, she received a project from Department of Science and
Technology (DST), Government of India under Scheme for Young Scientists and Professionals (SYSP) by
Science for Equity, Empowerment and Development (SEED) Division for three years (2008-2011). Her
research focuses on Brain Image Processing and brain tumor or lesion detection from MR Head Scans to
enrich the Computer Aided Diagnostic process, Telemedicine and Tele radiology services. She is Academic
Community Member (ACM) in International Congress for Global Science and Technology (ICGST), Life
Member (LM) in Indian Society for Technical Education (ISTE) and Lifetime Member (LM) in
Telemedicine Society of India (TSI).
Nagaraja P. is a Research Scholar (Full-time) in the Department of Computer
Science and Applications, Gandhigram Rural Institute - Deemed University,
Dindigul, India. He received his Bachelor of Science (B.Sc) degree in Physics in
the year 2008 and Master of Computer Applications (MCA) degree in the year 2011
from Gandhigram Rural Institute - Deemed University. He is currently pursuing
Ph.D. degree in Gandhigram Rural Institute- Deemed University. His research
focuses on Brain Tissue Segmentation in MRI Head scans.
P. Sriramakrishnan received his Bachelor of Science (B.Sc.) degree in 2011 from
Bharathidasan University, Trichy, Tamilnadu, India. He received Master of
Computer Application (M.C.A) degree in 2014 from Gandhigram Rural Institute-
Deemed University, Dindigul, Tamilnadu, India. He worked as Software Developer
in Dhvani Research and Development Pvt. Ltd, Indian Institute of Technology
Madras Research Park, Chennai during May 2014 – March 2015. He is currently
pursuing Ph.D. degree in Gandhigram Rural Institute- Deemed University. His research focuses
on Medical Image Processing and Parallel Computing. He has qualified UGC-NET for
Lectureship in June 2015.

A SIMPLE IMAGE PROCESSING APPROACH TO ABNORMAL SLICES DETECTION FROM MRI TUMOR VOLUMES

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to A SIMPLE IMAGE PROCESSING APPROACH TO ABNORMAL SLICES DETECTION FROM MRI TUMOR VOLUMES (20)

Recently uploaded (20)

A SIMPLE IMAGE PROCESSING APPROACH TO ABNORMAL SLICES DETECTION FROM MRI TUMOR VOLUMES