Segmentation and Isolation of Brain Tumors Using Different Images Segmentation Methods

are categorized into the benign and malignant


Introduction
A tumor is abnormal cell growth in the body, it is a very dangerous disease which affected human life.So, it is important to detect tumors early to save life.Tumors may have occurred in various human bodies as the tumor of brain.A tumor of brain has various sizes as well as shapes, and they required different treatment procedures.More than 120 types of brain's tumor are existing classified as primary and metastatic.Primary brain tumors start in brain tissues.The secondary brain tumors start in other body tissue and spread to brain.It is found the primary tumor is developing more in children.But the metastatic tumor is more common in adults.Tumors are categorized into the benign and malignant depending on the tumor's characteristics.The benign tumors are sluggishly growing and less aggressive while the malignant tumor is rapidly growing and life threatening 1 .
The image of a normal brain contains many tissues such as: White Matter (WM); Gray Matter https://dx.doi.org/10.21123/bsj.

Magnetic Resonance Imaging (MRI)
The MRI is defined as the non-invasive method of imaging the internal structures and certain facets of function in the human's body.MRI using the Radio Frequency (RF) with the presence of strong magnetic fields to generate high three dimensions correctional MRI images of the human's body.The MR Image is obtained by placing the patient on a table as illustrated in Fig 2 and applying the magnetic field that affects the nuclei of many atoms in the body (especially Hydrogen atoms) and aligns them with the external magnetic field.Then applied RF signal to the specific part that will be imaged.When the application of RF signal is stopped, the energy is released from the human's body.The released signals are detected and used to construct the MR image by the computer 3 .

Brain Tumor Segmentation
The basic benefit of MRI is the visualization of soft tissues.But the MR images need enhancement due to the presence of noise.Then, the segmentation method isolates the brain tumor in MR images, after that, the segmented tumor's region analysis and classify as malignant and benign.To detect brain tumor, MR imaging is combined with automatic; manual; and semi-automatic brain segmentation.The manual segmentation is performed computer's software, but it has variation problems as it takes longer and gives inaccurate results that differ from one doctor to another.Semi-automatic methods are affected by personal interference, although automatic segmentation is depending on prior information for the detection of brain's tumor 5

Segmentation Methods
In general, the segmentation process divides the image into various regions based on pixel characteristics.MRI brain image, the separating of tumor tissues from different normal tissues is done by manual segmentation process.Segmentation of brain's tissues is very important to detect and diagnose the brain diseases.But it's hard task due to the unpredictable anatomical brain's structure, existence of intensity inhomogeneity, partial volume effects, and noise.The disorders of brain can be identified exactly by the segmentation of the brain's tissues, such as WM, GM, and CSF 6,7 .In this study many segmentation methods are used to segment MRI brain images are: edge detection, thresholding, and K-Means methods.

Edge Detection Segmentation Method
Edge detection is simple method used in image processing to represent the sharp discontinuities of the image intensity (called edges).Edges are produced because of the rapid changes in the intensity of image's pixel, which represents boundaries of an object in an image.Different types of operations are available and classified into two categories: first and second order derivatives 6 .First and second order derivative methods are used in this study to segment MRI brain images.The first order segmentation process is done by using a special mask to segment input image and generate a gradient image.Sobel and Prewitt 6 represent the first order operators and are called gradient operators.They are detecting edges by identifying the maximum and minimum pixel's intensity values, and inspecting the intensity distribution in the neighborhood of a given pixel, and then decided if a pixel is represented by edge 8 .
Sobel, Prewitt, and Canny 6 detectors are convolving the raw image with their special convolution masks to generate the segmented image (gradient), then use specific threshold values to detect edges 9 .A result of this procedure is based on threshold values 10 .MATLAB program implementation these operators using special threshold technique depending on the Root Mean Square to estimate noise in the image 6 .

Thresholding Segmentation Method
Thresholding segmentation is one of the important segmentation methods.This segmentation method partitions the input image's pixels depending on their intensity level.The thresholding segmentation technique can be applied for images having dark objects on the white background and vice versa.
The selection of the threshold value (T) is the important step in the thresholding segmentation, which can be done manually or automatically.Depending on image's features, three types of thresholding methods are available, these are: Global, Variable, and Multiple Thresholding 11 .In this study, the global thresholding method is used.The global thresholding method is applied using an appropriate T, and it will be constant for the whole MRI image.The resultant segmented image q(x,y) is obtained from the original image p(x,y) using Eq 1 12 .

K-Means Segmentation Method
K-Means segmentation process is a supervised clustering segmentation technique; this type of clustering includes human interaction to input the cluster numbers (K).The segmentation process by K-Means is done by grouping the data vectors into predefined K. Firstly; the center of each cluster is selected randomly.Then pixels are assigned to the cluster depending on the nearest distance between pixel and the center of cluster (using Euclidian distance).When all pixels of image are clustering, cluster's mean is then calculated, and repeated this process until no changes result in cluster's mean 13 .The feature vectors (X) derived from l dataset be: The cluster centroids (C) is given by Eq 3.
The feature vectors are grouped into (k) clusters using the Euclidean distance as shown in Eq 4.
K-Means clustering techniques are considered relatively high-quality clusters, and present very good results in low level of computation.Generally, the aims of k-Means segmentation method are to minimize the sum of squared distances between all images' pixels and the cluster center.The K-Means method produce good results with many data sets, but its good performance is limited mainly to compact groups.On the other hand, the disadvantages of K-Means method include the determination of K, producing different results based on different initial conditions and the centers away from optimum location 14 .
Many segmentation methods were used in pervious works to segment brain tumors 15 , and gave different results.In our study, we collect many segmentations method and applied to the same images to compare and identify whether any method that will give acceptable results.

Methods:
In this study many segmentation methods (edge detection, thresholding, and K-Means) are applied to segment MRI images and isolate the brain tumor using the MATLAB program (R2019a).The images used were obtained from the Internet websites for several people diagnosed with brain tumors as shown in Fig 3.The following diagram is representing the major steps used in this study of segmentation the brain tumor.

Algorithms:
1. Edge's detection: Read MRI images; Convert to RGB; Applying MATLAB code of (Sobel, Prewitt, and Canny) operators; Displayed original and edges detection segmented images;

Thresholding:
Read MRI images; Convert to RGB; Select the T values; Applying MATLAB code of thresholding method; Displayed original and thresholding segmented images;

K-Means:
Read MRI images; convert them to an array; Input the clusters numbers (K); Choose the cluster's center; Determine the distance between each pixel and cluster's center; Applying MATLAB code of K-Means with different numbers of clusters; Displayed original and thresholding segmented images.

Results:
The results of applying the pervious algorithms are displayed in the following figures, where the first column represents the original images and others explained the results of each segmentation method.

Conclusion
The results of segmentation methods show that, the edges detection methods are failed to achieve our purpose, thresholding segmentation methods given good result when using T=120, and when using T=200, the acceptable result were produced.The results of applying K-Mean segmentation method were not succussed to segment and isolate brain's tumor.Only when using K=5, this method was succeed to isolate tumor of brain.
2024.7640 P-ISSN: 2078-8665 -E-ISSN: 2411-7986 Baghdad Science Journal (GM); and Cerebrospinal Fluid (CSF) as illustrated in Fig 1.To diagnose the structure of human's brain many medical imaging techniques are used: Computed Tomography (CT); Positron Emission Tomography (PET); and Magnetic Resonance Imaging (MRI) medical imaging, which provide information from a variety of excitation sequences about the brain tissues 1,2 .In this study MRI brain images are used.

Figure 1 .
Figure 1.Shows the MRI images with GM, WM, and CSF 2 .

Figure 3 .
Figure 3.The MRI brain images are used in this work.