Medical a vital role in accurate diagnosis
image processing is one of the most challenging fields. Detecting brain tumor
is a recent challenge faced by medical imaging researchers, it’s challenging
because of complex anatomy of brain. Brain tumor is abnormal growth of cancerous
cells in the brain that can even cause death.
This paper provides a brief overview
of multiple methods and techniques used for brain tumor detection along with
classification algorithms. These well-known methods use MRI scanned images for
tumor detection purpose, that provides better results than CT scan,
ultrasound and X-ray. By using MATLAB software and applying various
image processing techniques discussed i.e Image pre-processing, Image
enhancement, segmentation and feature extraction, brain tumor can be detected
from MRI images of the brain.
Processing, MRI, Brain Tumor, Extraction, Segmentation, Classification
medical image processing is a demanding and emerging field, which helps
surgeons or doctors in analysis and diagnosis of complex diseases such as
cancer, kidney or bladder stones and brain tumor etc. the diagnosis or detection of brain tumor has
vital importance because it is one of the most common brain diseases that can eve
cause death 1. Brain tumor detection is challenging because of brain’s
complex structure. Brain is a central part of human body that contains many
cells, that grows and divide. Brain tumor is intracranial mass due to abnormal
growth of cells in the brain. The brain tumor can either be cancerous or benign.
Primary brain tumor starts from the brain whereas, secondary brain tumor
expands into brain through some other body part.
imaging plays a vital role in accurate diagnosis of brain tumors. Different
techniques are used for analysis and detection of brain tumor such Magnetic Resonance
Imaging (MRI), Computed tomography (CT) scan, Ultrasound and X-ray. MRI is
mostly and widely used technique in medical imaging because it provides better
results i.e high quality images than CT scan and X-ray, and lack side effects
on body tissues. Magnetic field and Radio frequency pulses are used in MRI
imaging to scan and generate images of structure and organs of body. It helps
doctors to perfectly visualize anatomic structures of brain by creating a
high-resolution image. Detection of brain tumor at an early stage is a major
issue, By using MATLAB tool and image processing techniques, not only detection
but classification of brain tumor is also possible.
tumor detection is very crucial task because of the complex structure of brain,
most of the patents die because of
inaccurate detection 1. Various image processing
techniques are being used for accurate detection, identification and
classification of brain tumor. 2 proposes a tumor detection, identification and
classification method, that detects the tumor by segmentation and feature
extraction methods with the aid of pixel intensity. This technique uses consecutive
smoothing stages to remove noise and high frequency components, non maximum suppression
and region of interest (ROI) detection through thresholding. It helps to
measure tumor position and presents a method to prevent the spreading of tumor.
3 focuses on detection of brain
tumor from MRI images. This paper proposes use of region growing method that
define boundaries of brain tumor and provides precise segmentation and
identification of brain tumor. In this paper salt & pepper noise is added
during identification process and then filtered out by median filter and lastly
tumor is located using segmentation.
4 proposes a method based on neural
network and discrete wavelet transform.
Neural networks are used for
identification of brain tumor, the neural networks are trained for selected
features which are extracted from the image and tumor can be detected. Here
fusion method is used for detection by using multimode scanning of images which
gives relatively good results, and discrete wavelet transform is applied at
image to get coefficient values. A fully automatic segmentation process for
tumor is proposed, in which the algorithm used, integrate the images first and
all noise is removed and then transformed into a new image.
5 This paper proposes a strategy for detection and extraction of brain tumor
from MRI images of patient’s brain. The method consists of various noise
removal functions, morphological operations and segmentation. MATLAB software
is used for detection and extraction through MRI scanned images. Two staged algorithm
is applied, at first stage preprocessing of images is done and segmentation is
done in second stage. Morphological operations are performed later. They state that the stage of tumor
is based on the area of the tumor. So, for this, size of the tumor is
easily calculated by calculating number
of white pixels in tumor binary image. Brain Tumor can be classified according
to its type.
Numerous other Image processing methodologies are also being used for detection of brain tumor
6 presents a simple strategy for detection, using image pre-processing,
segmentation and extraction method. They proposed Histogram thresholding,
K-means clustering and Fuzzy C-means & support Vector Machine (SVM)
methods. The methods presented, includes different pre-processing steps such as
noise removal and RGB to Gray conversion.
image processing techniques are used for brain tumor detection, the techniques
are applied on MRI scanned images, techniques used mainly consists of following
4 steps: Pre-processing, segmentation, extraction and classification. 7
and Enhancement of an Image
and image enhancement is first step of image processing, for accurate detection,
even the finest details of the image are enhanced and noise is removed Fig 2.
Firstly, film artifacts are removed from MRI images (x-ray marks and labels
etc) then different filters are used according to requirement such as isotropic
filter is used for removal of background noise, weighted median filter to
remove salt and pepper noise etc. weighted median filtering technique gives
better results than median, spatial or adaptive filters 3-5
Pre-processing and enhancement 7
segmentation is a necessary and crucial step in image analysis, it is used to
extract different features of image. It’s a process of separating or dividing
an image into multiple regions having similar properties such as color, gray
level, brightness or contrast etc. it separates or divide an image into multiple specific
regions, pixels in each region exhibit high similarity and pixels that lie
between the regions exhibit high contrast 7-8. In brain tumor
segmentation tumor tissues are separated from normal brain tissues. Then to
improve the quality of images and limit the risk of distinct regions fusion in
the segmentation phase an enhancement process is applied. Various techniques for
image segmentation are used such as Thresholding approach, edge approach,
region approach and clustering etc 8 each of which have several advantages
and disadvantages, therefore, they are selected or used based on requirement.
The main approaches to segmentation are as follows:
is the most commonly used segmentation method in which image pixels are
allocated to categories according to the range where pixel lies. It uses intensity
histogram and find the intensity values i.e threshold values. Therefore, the
image is segmented based on threshold value. Eq 1
‘v’ is the gray value and ‘t’ is the threshold value. After thresholding, the
image is segmented into two values 0 and 1 and gray scale values are converted
this approach, edge filter is applied to detect edges in an image, edges are
assumed to represent object boundaries and help identify objects. Edges
are identified by rapid transition of intensity, after identification pixels are linked together to
represent a boundary.
In this approach edge position is given by either first order derivative or by
zero crossing in second order derivative.
approach focuses on finding the
object region rather than edges, it’s based on an assumption that bordering
pixels in one region that have same values. Pixels are compared with the
neighbors to see if congruence criteria satisfy, if it does so, the pixel is
set to belong to the cluster as one or more of its neighbor.
are two types of region approaches: Region growing and Region splitting. In Region
growing an initial point (seed point) is defined, all the other surrounding
pixels having same intensity value as seed point are connected to the seed
point. Whereas, in region splitting approach no seed point is defined and the
image is divided into unconnected regions, which are connected later based on
based approach uses different clustering algorithms, such as:
It is an unsupervised, iterative clustering
method, also known as hard clustering algorithm. It is a widely used method,
that separates a given dataset into k number of datasets or clusters, where each
cluster is defined by its centroid, that is a point in which sum of distance
from all the objects in that cluster is minimized, therefore, Purpose of this
algorithm is to minimize the sum of distances of all the objects to their
cluster centers and the objects are set to belong to certain
cluster 8, hence called hard clustering. It is a fast algorithm and robust to implement but has a drawback
that it may not successfully find overlapping clusters and may also fail to
cluster non-liner datasets or noisy data.
clustering is also known as soft clustering. As boundary of tumor tissue is
irregular, this fuzzy clustering technique can be helpful in getting better results.
In this approach objects located on boundaries of clusters can be member of
many clusters, Objects are classified into different groups such as the pixel
value of an image can belong to many clusters unlike K-means, the objects are
not forced to belong to certain clusters. Comparison between K-means and Fuzzy
is given in Table 1.
It is an optimization technique,
consisting of three main operators: Recombination, Mutation and Selection
operation. It operates on population of strings, at the start number of solutions
or populations are available, the solution from one population is utilized to
form a new population which is superior than the old one. If some condition
satisfies this process is repeated. It can easily be implemented and can solve
higher non-linearities but its computational cost is a major drawback.
Particle swarm optimization(PSO)
is a population based search technique, it is initiated by randomly selected
populations (particles) each of which have individual fitness value, that can
be calculated by fitness function. In segmentation optimal cluster centers are
determined by PSO. Unlike GA,
lacks recombination operator. Comparison between GA and PSO is given in Table 2.
extraction of a tumor is a critical task because of brain’s complex structure 8.
Shape, location of tumor, size and composition are some major parameters that
are considered for feature extraction.
of tumor is then done based on the results obtained by feature extraction.
It is an extended form of hard clustering
It is known as hard clustering
Objects may be linked with multiple clusters
Objects are set to belong to certain clusters
Cluster center is based on distance between data points
Each cluster has a center point i.e centroid
Used for analysis based on only distance between various
input data points
Used for analysis based on location and distance between
various input data points
Implements three operations: Recombination, mutation and
Lacks recombination operator.
Operations are not labeled like GA
Initially a discrete technique suitable for combinatorial
Continuous technique, not well suitable for combinatorial
High computational cost and more parameters to adjust
Fewer parameters to adjust, Therefore, easier to
Comparison between Genetic Algorithm and PSO
MRI imaging is very
helpful for analysis, diagnosis, and treatment of brain tumor & provides
better results than CT scan, ultrasound and X-ray. It helps in brain tumor
detection by segmentation, which is a critical step because wrong
identification may lead to severe consequences. This paper gives an overview of
several state of the art methodologies used for the detection of brain tumor
such as Image pre-processing, enhancement and extraction, and also various
algorithms used for classification