More Related Content
Similar to An overview of automatic brain tumor detection frommagnetic resonance images
Similar to An overview of automatic brain tumor detection frommagnetic resonance images (20)
An overview of automatic brain tumor detection frommagnetic resonance images
- 1. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
61
AN OVERVIEW OF AUTOMATIC BRAIN TUMOR DETECTION
FROM MAGNETIC RESONANCE IMAGES
Mayur V. Tiwari, D. S. Chaudhari
Electronics and Telecommunication Department,
Government College of Engineering, Amravati.
ABSTRACT
Magnetic resonance (MR) images are very useful tool to detect the tumor growth in
brain but precise brain image segmentation is a complex and time consuming process.
Manual brain tumor detection may prone to human error. In this paper different technique of
automatic brain tumor detection with our proposed method has been discussed.Using one or
two level of wavelet transform and Gabor filters, brain tumor detection is possible.
Key Words: Magnetic Resonance Imaging (MRI), Computed Tomography, UDWT, Gabor
Wavelets, Segmentation.
I. INTRODUCTION
Brain tumor is very hazardous disease to human beings due to which death may
occur. In the field of medical engineering brain tumor segmentation and detection from MR
images has been recent area of research [1]. Most of the researches from this field have
shown their interest to develop an automatic brain tumor segmentation and detection
system.Brain diseases and disorders have been commonly studied using magnetic resonance
imaging (MRI) technique [4]. For automatic detection of brain tumors, Magnetic resonance
Imaging gives better results than computed tomography (CT) since greater contrast between
different soft tissues of human body has been provided MRI [1]. Hence MR images are
widely used in detection of brainand cancerous tumors [1].
Automatic image segmentation is most important process in medical imaging for
analyzing different types of pathological and healthy tissues. In MRI segmentation problem is
to label voxels according to their tissue types which includes white matter (WM), grey matter
(GM), cerebrospinal fluid (CSF) and pathological tissues (tumor), etc [3]. Radiologist can
diagnose and find location of lesions based on visual diagnosis with help of available
INTERNATIONAL JOURNAL OF ADVANCED RESEARCH IN
ENGINEERING AND TECHNOLOGY (IJARET)
ISSN 0976 - 6480 (Print)
ISSN 0976 - 6499 (Online)
Volume 4, Issue 2 March – April 2013, pp. 61-68
© IAEME: www.iaeme.com/ijaret.asp
Journal Impact Factor (2013): 5.8376 (Calculated by GISI)
www.jifactor.com
IJARET
© I A E M E
- 2. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
62
software. While performing visual diagnosis, error may be introduced in finding location of
tumor. To remove this problem there is need to develop a computer based automatic
technique for detecting tumor from brain MR images [4].
For different surgical evaluations, Fluid Attenuated Inversion recovery (FLAIR)
magnetic resonance images are most widely used from different kind of MR images. FLAIR
images are widely used in study and analysis of brain maturation, acute hemorrhage and in
diagnosis of intra cranial tumor, periventricular lesions, head injury, multiple sclerosis etc. A
FLAIR sequenceproduces heavily T2 weighted MR images in which cerebrospinal fluid does
not appear. In FLAIR images, subtle lesions near CSF appear against a background of
attenuated CSF which looks dark and lesions, tumors and edematous tissues looks bright [4].
Automated tumor segmentation becomes difficult since it is tedious job to create a pre
designed model of expected size, shape, location or image intensity [3]. Brain tumors are of
different sizes, locations and positions. They also have overlapping intensities with normal
tissues [3].Accurate brain tissue segmentation and detection is a very vital issue for the
diagnosis of brain tumors and study of brain diseases. Brain tumor detection is a time
consuming process since tumor location and size is manually detected by expert because of
which, it is more prone to human errors. Therefore, it is necessary to develop fully automated
brain tumor segmentation and detection tools [2].
There are some techniques available such as classical, fuzzy, neural network etc, for
automatic brain tumor segmentation and detections. Classical techniques uses thresholding
(part below threshold becomes bright otherwise dark), edge and region based methods which
are known as standard image processing. In region based technique the region with similar
pixel values are extracted. The brain tumor from MR images consist of similar pixel values.
The edges consist of high frequency information of an image. Due to the lack of information,
final segmentation using classical methods are highly sensitive to noise and usually do not
result in continuous regions. Fuzzy segmentation technique applied for multichannel images
but not for single channel images. Neural network techniques are applied only for some types
of images [3].
II. BRAIN TUMOR DETECTION SYSTEMS
A. Usmanet al. have designed an automated computer aided system which segment
and detect the brain tumor. With the help of this system, brain tumor can be extracted
usingpreprocessing, global thresholding and post processing. Pre processing is used to
reduced the noise and enhance the brain MR image for further processing. In this stage the
image quality get enhance with more surety and ease in detecting the tumor. Segmentation is
the next process after improving the image quality. It separates similar pixel values from
different pixel values that means separate foreground from its background. Segmentation is
done with the help of technique called global thresholding in which threshold value can
select. Depending upon pixel value either it is above or below of threshold value considered
as foreground or background respectively. The remaining part is post processing operations
which get applied on the image to clearly locate the tumor part in the brain. It can include
morphological operations (erosion, dilation, masking etc) and windowing technique [1].
This technique of brain tumor detection is very simple and useful for some particular
kind of MR images. This technique used the global thresholding in order to segment an
image. Due to this false result may get obtained while detecting the tumor, since all whitish
part of the brain MR image is not always tumor.
- 3. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
63
Automatic detection and segmentation of brain tumor using fuzzy classification
and deformable models was introduced by W. Yang et al. The proposed method is fully
automatic which is a combination of region based and contour based methods. It requires
MR image of brain tumor which is applied to the improved kernel Fuzzy c-mean
(IKFCM) method for rough detection of brain tumor. For precise tumor segmentation,
this rough detection is used as the initial value of a deformable model. Classification can
be done using IKFCM and Morphological operations. Improved kernel fuzzy c-means is a
combination of Fuzzy c-means (FCM) and kernel method. For data classification point of
view both membership and typicality are mandatory for data structure interpretation. It
computes these two factors simultaneously. Noise sensitivity defect of FCM removed by
IKFCM. For tumor detection and labeling histogram based IKFCM is used, since it is
faster than classical FCM implementation. The extracted brain image can be classified
into five classes, cerebrospinal fluid, white matter, tumor, gray matter and background.
The mean of cerebrospinal fluid, white matter and gray matter are used as the centers of
their classes. Zero value is used for background. Tumor has highest intensity among the
five classes. Some morphological operations such as opening, erosion, largest component
selection, etc are then applied to the tumor region in order to correct misclassification
errors.Segmentation results were performed through a quantitative comparison with the
results of fuzzy possibilistic c-means (FPCM). To evaluate the results two measures are
used - ratio of correct detection (TP) and ratio of false detection (TF). The TP shows how
much of the actual tumor has been correctly detected, while FP shows how much of the
detected tumor is wrong [2].
The standard images or methods are not present for quantitative evaluation of the
results of segmentation. Hence more tests are however necessary to further validate the
approach.
In some researches it was proposed that the tumor edema and healthy tissues can
be detected from FLAIR MR Images with the help of higher order wavelets and statistical
parameters has been detected. Intracranial brain images and feature vectors get extracted
from the blocks of 4ൈ4 pixels of images. Feature vectors have developed high order
functions of statistics and wavelet functions of horizontal and diagonal bands of each
block. A wavelet transform can decompose an image into some simpler images. These
decomposed images have different scale and dimension. Dimension is depends upon level
of wavelet transform. For an instance,8ൈ8 block is decomposed into four frequency bands
of 4ൈ4 coefficients using wavelet transform, out of which one low frequency band and 3
high frequency bands. The edges of the images consist of high frequency bands.The high
frequency bands consist of texture properties which is useful in segmentation process.
Due to this, feature vectors were derived from coefficients of high frequency bands.
These feature vectors are considered as input and get clustered into 3 segments
corresponding to tumor tissues, edema tissues and healthy tissues using fuzzy c- means
algorithm. Brain is analyzed for healthy and pathological tissues. Pathological tissues
percentage in intra cranial brain is calculated for further process. For better human
visualization randomcolouring of segmented regions are done. Fuzzy back propagation
algorithm is used for supervised training of Artificial Neural Network(ANN) with results
of feature vectors and edema. Finally tumor and edema are detected from converged
values of fuzzy ANN [4].
- 4. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
64
S. Ghanavati et al. have proposed an automatic algorithm for tumor detection by
using different MRI modalities such as T1, T2-weighted and T1 with gadolinium contrast
agent that are mostly used for diagnosis. These modalities can provide better visualization of
abnormalities. The intensity, shape deformation, symmetry and texture features were
extracted from each image. The Adaboost classifier is used to select most discriminative
features and also to select the tumor. Adaboost classifier improves classification accuracy. In
the next step, the simulated tumor is used for training and validation of classifier. The data
preprocessing is the first step in which images are corrected for the intensity in homogeneity
using non parametric non uniform intensity normalization method. Brain extraction tool
(BET) is used to remove the non brain tissue like skull, eyes etc. Next process is the feature
extraction in which the parameters like intensity, symmetry, shape deformation and texture
are extracted. Classification is the last step through which the tumor is detected. Here
Adaboost classifier selects and combines the most discriminative features during learning
process [5].
In some of the researches, it was studied that brain tumor detection can be possible
using colour converted k – means clustering segmentation. This approach first convert the
input gray – level image into a color space image and operating the image labeled by cluster
indeed. Clustering is a process in which date points with similar feature vectors can groups in
a single cluster while data points with dissimilar feature vectors placed in different clusters.
The data points that are nearly similar to the feature space are clustered together follows with
further clustering of similarity feature space. Similarity of feature vectors can be represented
by as Euclidean or Mahalanobis distance. The clustering is reckoned approach for
partitioning d-dimensional data into k – clusters. K – means clustering algorithm places k –
clusters for arbitrarily selected cluster centroids ‘vi’ where i = 1,2,---,k and modifies centroid
for new cluster shape formation. Colour- converted segmentation test is the next step to
segment an image. The colour converted segmentation is a method that converts colors into a
single integer index where a specified color can be selected from a list of colours in a data
collection. From final segmented image, the brain tumor is detected [7].
J. Jayakumariet al. had proposed method for automatic detection of brain tumor based
on magnetic resonance image using computer aided diagnosis system with watershed
segmentation. For automatic brain tumor detection,computer aided diagnosis system can
provide better accuracy, simple algorithm, precision and good performance. Preprocessing
and Enhancement techniques are the first step of computer aided diagnosis system. It is used
to improve image quality for segmentation and detection of suspicious or tumor regions in
MR images. The enhancement stage can eliminate image uncertainty such as particular labels
and marks on MRI image and high frequency components. Generally low pass filter, median
filter, Gabor filter etc. may be used as enhancement techniques. In this stage noise can also be
removed. To separate tumor region from background MR image the process called
segmentation is used. Image segmentation is the process of partitioning a digital image into
multiple segments. Segmentation is based on discontinuity and similarity. Segmentation finds
the Region of Interest (ROI) in an image. Here watershed segmentation technique is used to
segregates any image as different intensity portions and tumor cells consist of cerebrospinal
fluid which has very high intensity.A watershed is a basin like structure defined by
highpoints and ridgelines. The major idea of watershed segmentation is based on topographic
representation of image intensity.Watershed segmentation can segregate tumors and high
intensity tissues of the brain [6].
- 5. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
65
In recent studies it was found that undecimated wavelet transform beneficial in
automatic detection of the brain tumor. Wavelet decomposes data into different frequency
components, due to which each component can be studied with a resolution matched to its
scale. Hence wavelet transform is best tool for image feature extraction. The undecimated
wavelet transform also called as stationary wavelet transform (SWT) leads to a redundant
input signal representation, since absence of decimator. Basically UDWT convert the data
into the appropriate low and high frequency components at each level. The resultants
have the same length as the original sequence. By modifying the filter at each level
multiresolution is achieved. The 2 – D UDWT diagram is shown in figure1.
input
image
LowHigh
LowHigh
columns
High
columns
Low
rows
rows
columns
columns
HH HL LH LL
horizontalverticaldiagonal
Figure 1 Two-dimensional UDWT
The Gabor wavelets provide multi-channel filtering approach of texture analysis.
The Gabor filters are basically a band - pass filters. Such band-pass filters efficiently
captured the characteristics of tumor tissue. The decomposed image has been passed
through Gabor filter. Here region-based segmentation is possible through multi-channel
filtering technique. The peaks within the array of a feature vector are determined by peak-
finding algorithm. Finally using k-means clustering algorithm the segmented brain tumor
is produced [3].
III. PROPOSED ALGORITHM
Proposed technique of automatic brain tumor detection can effectively detect the
tumor part of the brain from magnetic resonance images. Wavelet transform and Gabor
filter are used to preprocessed the input image. The resultant image is then passed through
further algorithmic steps from which finally brain tumor is detected.The proposed work
flow of brain tumor detection is shown in figure 2.
- 6. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
66
Figure 2 Proposed work flow of automatic brain tumor detection from MR images
After applying some morphological operations like thresholding, dilation, etc. the resultant
image is converted to binary image. Parameters such as area, centroid, perimeter, etc. are
computed from white patch or blob also called as Region of Interest (ROI). When the
parameters of ROI found within the specified range brain tumor can be detected.
- 7. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
67
IV. CONCLUSION
Several technologies for automatic brain tumor detection have been discussed in
this paper. MR image consists of different gray levels and intensities. Also brain tumor is
varied in size and shape. One can develop a system which detects the tumorwith
precision.Wavelet transform is one of the tools used in detection of brain tumor.Wavelet
transform provide multiresolution approach since it decomposes an image into several
images. Each decomposed image consists of different frequency information.The
proposed method can segment and detect the tumor from surrounding brain tissue.
REFERENCES
[1] M. Akram, A. Usman, “Computer Aided System for Brain Tumor Detection and
Segmentation,” IEEE International Conference on Computer Networks and
Information Technology, pp. 299-302, 2011.
[2] W. Yang, M.Siliang, “Automatic detection and segmentation of brain tumor using
fuzzy classification and deformable models,”IEEE,4th International Conference on
Biomedical Engineering and Informatics, pp. 1680-1683, 2011.
[3] G. Mirajkar, B. Barbadekar, “Automatic Segmentation of Brain Tumors from MR
Images using Undecimated Wavelet Transform and Gabor Wavelets,”IEEE
International Conference on Electronics, Circuits and Systems, pp. 702-705, 2010.
[4] N. Pradhan, A. Sinha, “Intelligent computing for the analysis of Brain Magnetic
Resonance Images,”IEEE, First International Conference on Integrated Intelligent
Computing, pp. 211-217, 2010.
[5] S. Ghanavati, J. Li, T. Liu, P. Babyn, W. Doda, G. Lampropoulos, “Automatic Brain
Tumor Detection in Magnetic Resonance Images,” IEEE, International Symposium
on Biomedical Imaging, pp. 574-577, 2012.
[6] K. Viji, J. Jayakumari, “Automatic Detection of Brain Tumor based on Magnetic
Resonance Image using CAD System with watershed segmentation,” IEEE,
Proceedings of International Conference on Signal Processing,Communication,
Computing and Networking Technologies, pp. 145-150, 2011.
[7] L. Juang, M. Wu “MRI brain lesion image detection based on color-converted k-
means clustering segmentation,” Measurement 43, pp. 941-949, 2010.
[8] B.Venkateswara Reddy, Dr.P.Satish Kumar, Dr.P.Bhaskar Reddy and B.Naresh
Kumar Reddy, “Identifying Brain Tumour from MRI Image using Modified FCM
and Support Vector Machine”, International journal of Computer Engineering &
Technology (IJCET), Volume 4, Issue 1, 2013, pp. 244 - 262, ISSN Print: 0976 –
6367, ISSN Online: 0976 – 6375.
[9] Selvaraj.D and Dhanasekaran.R, “MRI Brain Tumour Detection by Histogram and
Segmentation by Modified GVF Model”, International journal of Electronics
and Communication Engineering &Technology (IJECET), Volume 4, Issue 1, 2013,
pp. 55 - 68, ISSN Print: 0976- 6464, ISSN Online: 0976 –6472.
- 8. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 4, Issue 2, March – April (2013), © IAEME
68
AUTHORS’ INFORMATION
Mayur V. Tiwari received the B.E. degree in Electronics and
Telecommunication Engineering from H. V. P. M College of Engineering
and Technology, Amravati University in 2009. He is currently pursuing
M.Tech in Electronic System and Communication from Government
College of Engineering, Amravati.
Dr. Devendra S. Chaudhari obtained BE, ME, from Marathwada
University, Aurangabad and PhD from Indian Institute of Technology,
Bombay, Mumbai. He has been engaged in teaching, research for period of
about 25 years and worked on DST-SERC sponsored Fast Track Project
for Young Scientists. He has worked as Head Electronics and
Telecommunication, Instrumentation, Electrical, Research and incharge
Principal at Government Engineering Colleges. Presently he is working as Head, Department
of Electronics and Telecommunication Engineering at Government College of Engineering,
Amravati. Dr. Chaudhari published research papers and presented papers in international
conferences abroad at Seattle, USA and Austria, Europe. He worked as Chairman / Expert
Member on different committees of All India Council for Technical Education, Directorate of
Technical Education for Approval, Graduation, Inspection, Variation of Intake of diploma
and degree Engineering Institutions. As a university recognized PhD research supervisor in
Electronics and Computer Science Engineering he has been supervising research work since
2001. One research scholar received PhD under his supervision. He has worked as Chairman
/ Member on different university and college level committees like Examination, Academic,
Senate, Board of Studies, etc. he chaired one of the Technical sessions of International
Conference held at Nagpur. He is fellow of IE, IETE and life member of ISTE, BMESI and
member of IEEE (2007). He is recipient of Best Engineering College Teacher Award of
ISTE, New Delhi, Gold Medal Award of IETE, New Delhi, Engineering Achievement Award
of IE (I), Nashik. He has organized various Continuing Education Programmes and delivered
Expert Lectures on research at different places. He has also worked as ISTE Visiting
Professor and visiting faculty member at Asian Institute of Technology, Bangkok, Thailand.
His present research and teaching interests are in the field of Biomedical Engineering, Digital
Signal Processing and Analogue Integrated Circuits.