Morphological Transformation Based Fractal Texture Segmentation of MR Brain Tumour Image Slices

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org

Volume 8, Issue 1, January 2019 ISSN 2319 - 4847
Morphological Transformation Based Fractal

Texture Segmentation of MR Brain Tumour
Image Slices
R. Usha1, and K. Perumal2
1
Research Scholar, Department of Computer Applications, School of Information Technology,
Madurai Kamaraj University, Madurai, Tamil Nadu, India.
2
Professor, Department of Computer Applications, School of Information Technology,
Madurai Kamaraj University, Madurai, Tamil Nadu, India.
ABSTRACT
This paper presents the morphological transformation based fractal texture segmentation for the multi-model view of MR brain
tumor region extraction. The tumor image segmentation of this approach consists of morphological image transformation,
fractal texture based image segmentation, tumor region extraction. The segmented tumor image results are evaluated with the
ground truth image with use of jaccard index and dice coefficient measures. Then, the proposed image segmentation technique
is also compared with the conventional SFTA (Segmentation based Fractal Texture Analysis) based image segmentation in
terms of a number of image decomposition for tumor extraction and computation time.
Keywords: segmentation based fractal texture analysis, minima transform, image complement, morphological
transform.
1. INTRODUCTION
The brain is the most complex organ in the human body that has a stunning one hundred billion nerve cells or neurons.
Any issue in a brain triggers major problems even it might prompt the loss of patient’s life. The numerous distinct
fractal texture based approaches are developed and discussed for automatic medical image analysis of image
segmentation. Texture has an important role in medical based image analysis to discriminate the tissue
characterization. According to, the different aspects of geometrical fractal approaches [1] such as piecewise threshold-
box-counting (PTBC), piecewise modified box-counting (PMBC), and piecewise triangular prism surface area (PTPSA)
were analyzed to recognize the MR brain slices. Another approach of fractal texture feature based microcalcifications
[2] analysis was discussed in mammogram images to identify the various subjects, which has given a maximum of
accuracy other than LBP (local binary pattern). The new technique of wavelet based SFTA approach has introduced
[3] to classify the normal images into abnormal MR brain slices. SFTA is one of the most frequently used algorithms to
analyze medical images. This approach made use of these purposes like segmentation, feature extraction, and analysis
of image texture. However, SFTA [4] takes less computational time for extraction of image features rather than
Haralick, GLCM techniques. But, it aggregates more number of image decomposition to analyse the image details.
Sometimes, it may lead to the problem of over-segmentation when analyzing the image texture. Another approach of
the combined gray-scale morphology and marker-controlled watershed based image segmentation [5, 6], texture based
region merging approach [7] were introduced to avoid the over-segmentation. The combination of thresholding and
morphological image analysis technique was presented for automatic image segmentation to detect the brain tumor
from MR slices [8]. This study presents the brain image segmentation of MR brain abnormal tissues from normal
tissues such as gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) using morphological
transformation based fractal texture image segmentation approach.
2. RESEARCH METHODOLOGY
This section defines the methodology of morphological transformation based MR brain image segmentation. Totally, it
contains the following main phases such as (i) pre-processing steps of hierarchical image transformation, (ii) image
complement and minima transform, (iii) fractal texture-based image segmentation, and (iv) the post-processing steps of
Volume 8, Issue 1, January 2019 Page 1

area opening and ROI extraction. The overall process flow diagram of this method is illustrated in Figure 1.
Figure1. Process flow diagram of morphological transformation based image segmentation

2.1 Pre-processing
In this stage, the collected MR brain slices are pre-processed by the hierarchical transformation technique (HTT) [9] for
enhancing the original images. It incorporates the main phases, the optimum size of the structural element (SE)
selection and image enhancement. The selection of optimum SE size of each image is desired by the measures of
highest PSNR (Peak Signal-to-Noise Ratio) and lowest AMBE (Absolute Mean Brightness Error) value. Along with
this suitable disk-shaped structural element selection is desired to do the morphological transformation of top-hat and
bottom-hat to promote the well-qualified image for successive image analysis.
2.2 Image Complement
The morphological image complement operation has applied on the resultant pre-processed image, which is used to
enhance the image depends upon intensity or binary. For an intensity image, the block subtracts each pixel value from
the maximum pixel value of an image. The resultant image, light areas become darker and vice versa. The image
complement operation is mathematically defined by the equation of,
(1)
Where, A (i, j) and B (i, j) represents the output and input images respectively.
2.3 Minima Transform
Consequently, the extended minima transformation is applied to detect the intensity valleys deeper than a specific
threshold value with this function of [10],
(2)
Where, s delineates the given depth and M illustrates the reconstruction and erosion operators correspondingly. With
the use of H-minima transform, all minima whose depth is minimum or equivalent to the presented i-value are
destroyed.
2.4 Image Segmentation
The fractal texture segmentation approach of SFTA [4] algorithm has been followed by the transformation of an
enhanced image of complement and extended minima transform. Then, the morphological operation based area
opening and ROI calculation operations have been done to acquire the segmented image results of SFTA for further
detection of a tumor.
3 EXPERIMENTAL RESULTS
The morphological transformation based fractal texture segmentation result consequences of multi-model views in MR
abnormal slices is demonstrated in Figure2. At first, an enhanced image of hierarchical transformation is executed on
input abnormal brain slices. Then, the morphological based image complement, which is for diseased tissue portion
extraction and extended minima transforms that, is for highlights the brain abnormality these results are represented in
the column of 3 and 4. Afterward, multi-level Otsu threshold based image decomposition has done as mentioned in the
SFTA algorithm to extract the correct influenced tumor zone. The segmented result of this proposed approach is
depicted in the column of 5. Moreover, the morphological based post-processing tasks are executed like area opening,
and region of interest computation, which are shown in column 6 and 7. Furthermore, the extracted affected brain cells
have been established in the column of 8 at figure 2.

Figure2. Proposed abnormal image segmentation consequences of multi-model view of

MR brain images
3.1 Performance evaluation
The sample segmented abnormal image results are compared and analyzed with ground truth slices of brain image that
are shown in fig 3. Subsequently, the segmented tumor results are assessed with the appropriate ground truth slices
with use of jaccard index and dice co-efficient [11].
Image name Original MR Actual tumor Proposed image Tumor ROI
Image region as red segmentation extraction
Image1
Image6
Figure3. Comparison of extracted tumor image results with ground truth slices

If a= (a1, a2, a3 …an) and b= (b1, b2, b3 … bn) are two vectors (ai, bi) ≥ 0, then the jaccard coefficient is
mathematically depicted by,
(2.3)
Where xi and yi are ground truth of tumor extracted image, and the jaccard distance can be defined as
(2.4)
Where 0 ≤ J (a, b) ≤ 1. If a and b are empty, J (a, b) = 1. The range of jaccard’s distance lies between 0 and 1. Jaccard
distance goes its best value at 0 and worst value at 1.
According to these two co-efficient, image 2 and 3 attained the highest value among the image samples. Hence, the
average performance of jaccard and dice coefficient is accomplished by the morphological transformation based fractal
texture segmentation as 0.93 and 0.94 correspondingly.
Table1. Performance evaluation in terms of image similarity

Images / Jaccard Jaccard Dice
Measures distance coefficient coefficient
image1 0.09 0.91 0.89
image2 0.02 0.98 0.94
image3 0.04 0.96 0.99
image4 0.05 0.91 0.94
image5 0.1 0.9 0.93
image6 0.07 0.93 0.92
Average 0.06 0.93 0.94
4. COMPARISON OF MORPHOLOGICAL TRANSFORMATION BASED IMAGE SEGMENTATION WITH

SFTA SEGMENTATION APPROACH
In this section, an alternative approach of the mathematical morphological based image segmentation strategy has
been compared with segmentation of the SFTA algorithm. The comparison of two approaches has been represented
with the use of different view of sample abnormal image 1, 2. The aggregate number of an extracted threshold value of
these two techniques is illustrated in Table 2 and Table 3 when the size of a total number of an extracted threshold is
declared as 2.
Table 2. Extracted threshold values for sample abnormal images using multi-level Otsu thresholding of SFTA algorithm for STBD
and TTBD image decomposition
Input Size of Threshold Extracted threshold
image feature set (T) value
vector
image1 2 T(1) 0.0938
T(2) 0.2617
image2 2 T(1) 0.0859
T(2) 0.2148
Table 3. Extracted threshold values for sample abnormal images using multi-level Otsu thresholding of proposed approach for
STBD and TTBD image decomposition
Input Size of Threshold Extracted threshold
image feature set (T) value
vector
image1 2 T(1) 0.5000
T(2) 0.7539
image2 2 T(1) 0.4570
T(2) 0.7617

4.1 Threshold based image decomposition
The outcomes of the proposed morphological segmented abnormal image outcomes about are compared and analyzed
with the algorithm of SFTA as far as total time is taken and the several iterations. The binary image decomposition of
an input abnormal images sample 1, and 2 based on the two techniques such as conventional SFTA algorithm and the
proposed approach resultant images are compared when the size of threshold set (T) has been defined as 2; they are
shown at Figure 4.
Single Threshold based Image Decomposition (STBD)
Conventional SFTA results Proposed Technique results
Threshold Decomposed Decomposed Decomposed Decomposed Decomposed Decomposed
value Image 1 Image 2 Image 3 Image 1 Image 2 Image 3
T(1)
T(2)
Two Threshold based Image Decomposition (TTBD)

Conventional SFTA results Proposed Technique results
Pairs Decomposed Decompose Decomposed Decomposed Decomposed Decomposed
threshold Image 1 d Image 2 Image 3 Image 1 Image 2 Image 3
tl tu
T(1) T(2)
T(2) T(3)
Figure4. Comparison of conventional SFTA and proposed segmentation techniques results based on the STBD and
TTBD image decomposition
The Figure4 depicts the comparison of two segmentation approaches based on the multi-level Otsu thresholding
algorithm. The SFTA algorithm works better and faster rather than other feature extraction algorithm of GLCM,
Gabor, and Haralick [9] with respect to the feature extraction time. However, the traditional SFTA required the
maximum number of iteration to extract the texture features from an image and to segment the diseased tissue of the
brain to analyze the corresponding image patterns. In the event that the quantity of iteration goes high, the calculation
time likewise increases to analyze the appropriate image pattern of the brain images. These can be illustrated in Fig. 4
with the minimum of binary image decomposition when a size of the extracted threshold limit esteem is 2. The
representation of a multi-model view of abnormal image sample1, sample2 based on the SFTA algorithm is shown in
the first column of STBD and TTBD results. From these image decomposed results of SFTA, there is no detailed
pictorial information regarding the brain abnormality. The conventional SFTA still needs a most extreme number of
iteration for decomposing the image to describe the brain abnormality from the tissue of a normal one.
As contained in figure4, the proposed segmentation method yields preferable segmented outcomes rather than
traditional SFTA algorithm. And it also provides a detailed description of image representation with a limited number
of iteration. The proposed technique takes a most of two or three iterations or the size of threshold setting might be 2 or

3, has enough to separate the diseased tissues from normal tissues at brain image that has shown in the second column
of STBD and TTBD for axial view of image 1, coronal view of image 2 and sagittal view of image 3. But, in the
segmentation based SFTA does not give the information about unhealthy tissue of abnormal brain in the restricted size
of the threshold value set as 2 or 3. It requires the greatest number of extracted thresholds to extract the diseased tissue.
Figure5. Segmented image result of SFTA and Proposed techniques

for sample multi-model views of MR abnormal brain
Some of the post-processing of area opening and region of interest morphological operations are easily handled on the
proposed approach to separate the tumor area from the skull which has been appeared in Fig. 2. Relatively, the similar
process of post-processing operations on the segmented result of SFTA is too complex in the limited number of a
threshold value in threshold set based image segmentation. From the segmented results of these two techniques is
outlined in Fig.5. As per this figure, the segmentation results of conventional SFTA algorithm look like more complex
to differentiate the diseased tissue from normal tissue. In any case, the proposed segmentation technique offers better
image representation to differentiate the image irregularity from a normal one.
5. CONCLUSION
The present work gives a morphological transformation based fractal texture image segmentation of MR brain
slices. The extracted tumor region of the brain has also analyzed and compared with the appropriate ground truth
slices. And the total number image decomposition of conventional SFTA technique is reduced with the use of this
proposed approach that has discussed in the experimental result analysis. Similarly, the segmented image details of the
proposed approach provide better visual results other that of existing SFTA based segmented result in a limited number
of image decomposition.
References
[1] Iftekharuddin, K.M., Jia, W., and R. Marsh, “A Fractal analysis approach to identification of tumor in brain MR
images”, IEEE, ISBN: 0-7803-6465-7, Engineering in Medicine and Biology Society, 2002.
[2] Cardona, H.D.V., Orozco, A., and M.A. Alvarez, “Automatic Recognition of Microcalcifications in
Mammography Images through Fractal Texture Analysis”, Springer © Advances in Visual Computing, 2014,
841-850.
[3] Saraswathi, D., Srinivasan, E., and G Sharmila, “An Automated Diagnosis system using Wavelet based SFTA
Texture Features”, IEEE, 20014, ISBN: 978-1- 4799- 3834-6.
[4] Costa, A.F., Mamani, G.H., A. J. M. Traina, “An Efficient Algorithm for Fractal Analysis of Texture”, Graphics,
Patterns and Images (SIBGRAPI), 2012, 39-46.
[5] Parvati, K., Rao, B.S.P., and M. Mariya Das, “Image Segmentation using Gray-Scale Morphology and Marker-
Controlled Watershed Transformation”, Discrete Dynamics in Nature and Society, 2008.
[6] Punitha M., and K. Perumal, “Multileyer Perseptron ANN based Watershed Segmentation to Detect Mammogram
Tumor”, IPASJ International Journal of Information Technology (IIJIT), Vol. 6 (10), pp. 1-9, 2018.
[7] Ng, H.P., Huang, S., Ong, S.H., Foong, K.W.C., Goh, P.S., and W.L. Nowinski, “Medical Image Segmentation
using Watershed Segmentation with Texture-based Region Merging”, IEEE, Engineering in Medicine and
Biology Society, 2008.
[8] Sujan, MD., Alam, N., Noman, S.A., and M. Jahirul Islam, “A Segmentation based Automated System for Brain
Tumor Detection”, International Journal of Computer Applications, 2016, 153(10): 42-49.
[9] Usha, R., and K. Perumal, “Image Enhancement of Medical Images Based on an Efficient Approach of
Morphological and Arithmetic Operations”, International Journal of Computer Trends and Technology, Vol. 49
(5), pp. 253-258, 2017.
[10] Jung, C., and C. Kim, “Segmenting Clustered Nuclei Using H-minima Transfrom-Based Marker Extraction and
Contour Parameterization”, IEEE Transactions on Biomedical Engineering, 2010, 57(10): 2600-2604.
[11] Mohan, J., and V. Krishnaveni, “Automated Brain Tumor Segmentation on MR Images Based on Neutrosophic
Set Approach”, Proc: IEEE Sponsored 2nd International Conference on Electronics and Communication Systems
(ICECS 2015), Coimbatore, India, 26-27 Feb. 2015.
AUTHORS
Miss. R. Usha [1] is pursuing Ph.D in the Department of Computer Applications, School of Information
Technology, and Madurai Kamaraj University, India. She has completed M.Phil in Computer Applications
at Madurai Kamaraj university, India in 2014. She has Contributed papers more than 8 in International
Journals, National and International Conferences. Her Research work is focused on Medical Image
Processing.
Mr. K. Perumal [2] is working as an Associate Professor in the Department of Computer Applications,
School of Information Technology, Madurai Kamaraj university, Madurai. He has contributed more than 35
papers in International Journals and Conferences. He has Guiding 6 Scholars. His Interest includes Data
mining, Image Processing, and Medical Imaging.

Morphological Transformation Based Fractal Texture Segmentation of MR Brain Tumour Image Slices

Uploaded by

Document Information

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Morphological Transformation Based Fractal Texture Segmentation of MR Brain Tumour Image Slices

Uploaded by

Copyright:

International Journal of Application or Innovation in Engineering & Management (IJAIEM)

Web Site: www.ijaiem.org Email: editor@ijaiem.org

Morphological Transformation Based Fractal

Volume 8, Issue 1, January 2019 Page 1

Figure1. Process flow diagram of morphological transformation based image segmentation

Volume 8, Issue 1, January 2019 Page 2

Figure2. Proposed abnormal image segmentation consequences of multi-model view of

Volume 8, Issue 1, January 2019 Page 3

Table1. Performance evaluation in terms of image similarity

4. COMPARISON OF MORPHOLOGICAL TRANSFORMATION BASED IMAGE SEGMENTATION WITH

Volume 8, Issue 1, January 2019 Page 4

4.1 Threshold based image decomposition

Two Threshold based Image Decomposition (TTBD)

Volume 8, Issue 1, January 2019 Page 5

Figure5. Segmented image result of SFTA and Proposed techniques

Volume 8, Issue 1, January 2019 Page 7

You might also like