Brain Tumor Detection: A Comparison among Various

Download PDF
0 downloads 0 Views 939KB Size Report
Comment
Dec 30, 2017 - Keywords: Brain tumor, MRI, MATLAB, Bhattacharyya coefficient, ... the purpose of processing 'imread' command of MATLAB will have to use.
Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9

Brain Tumor Detection: A Comparison among Various Techniques Md. Abdullah Al Mahmud, A. H. M. Zadidul Karim, M. Abdullah Al Amin, Md. Subbir Reza and Munmun Akter Brain tumor, a severe disease that has devastated many human lives. Day by day the numbers of affected people are increasing and researchers are becoming more concerned to detect these types of severe disease with accuracy. They are now combining their knowledge from medical to mathematical science to detect this disease with an appropriate way. The main objective of this paper is to form 3 types of algorithm for detecting tumor. First algorithm is based on watershed nd segmentation method, 2 algorithm is based on pixel intensity method and 3rdalgorithm is based on bounding box method using Bhattacharyya coefficient. Finally a comparison is discussed to give an explanation so that anybody can easily understand which method is better than other.

Keywords: Brain tumor, MRI, MATLAB, Bhattacharyya coefficient, Morphological operation, Pixel intensity.

I.

Introduction

There have a lot of diseases in human body. Diseases of brain are very common in human body. It takes place in different forms. Some of the major classes of brain diseases are tumors, infections, stroke, trauma and seizures[1]. An infamous disease has affected and devastated many lives which is brain tumor. This type of disease has been the intermediate of outlook of thousands of researchers over many years around the world. Brain Tumor: There are predominantly two types of brain tumor. Those are starting the brain (primary) and those are spread from cancer somewhere else in the body(metastasis).Primary brain tumors such as glioma happen very rare and when they do, they are commonly malignant(cancerous).Amalignant tumor is a clump of cancer cells that stands growing; except feed off the body it doesn't do anything[2-3].MRI: Magnetic resonance imaging (MRI) of brain is a safe and smooth test that uses a radio waves and magnetic field to produce manifest images of the brain and the brainstem [4].An MRI scanner build with a largedoughnutshapedmagnetthatmostlyhasatunnelinthecenter.Patientsarelyingonatable that slides into the tunnel. During the experiment, radio waves manipulate the magnetic position of the atoms of the body, which are collected by a powerful antenna and then sent to a computer. The computer carry out millions of calculations for resulting in clear cross-sectional black and white images of the body[5]. These images can be modified by three- dimensional (3-D) pictures of the scanned area. This helps identify the problems in the brain when the scan focuses on a specific areas. In this paper a number of automatic brain tumor detection algorithm is proposed. Every proposed algorithm will detect tumor and finally tumor area will be shown by the designed system. ___________________________________________________________________ Md. Abdullah Al Mahmud, A. H. M. Zadidul Karim and M. Abdullah AlAmin, Department of EEE, University of Asia Pacific University of Asia Pacific(UAP), Dhaka-1215, Email: [email protected], [email protected], [email protected], Md. Subbir Reza and Munmun Akter, Department of EEE, University of Asia Pacific University of Asia Pacific (UAP), Dhaka-1215, Email: [email protected], [email protected]

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 Methodology To detect brain tumor in an accurate way three methodology have been proposed in this paper. The following methodologies are2.1 Watershed segmentation basedmethodology[6-12] 2.2 Pixel intensity basedmethodology[13-18] 2.3 Bounding box methodology basedon Bhattacharyyacoefficient[19-20] Every methodology contains a strong algorithm. The following procedure is designed for watershed segmentation based methodology      

Considering MRI image as ainput. Gray scaleConversion Application of High passfilter Application of Medianfilter Thresholdsegmentation. Watershedsegmentation Morphologicaloperation.



Detection of tumor as anoutput

Pixel intensity based methodology contains the following algorithm• • • • • • • • •

ImageAcquisition Grayscaleimages Adjusting the MR image using peak intensity level Conversion of MRI image into binary image to detect theboundary Area of the head crosssection Conversion of adjusted image into binary image to detect the boundary of head crosssection Detecting the tumor affected region byeliminatingboundary Performance parameter of tumor the affected image Tumor Areacalculation

And finally the bounding box methodology based on Bhattacharyya coefficient contains following algorithm-

• • • • • •

As an input taking a MRIimage Normalize theimage SkullDetection Maskcreation Vertical and Horizontal Scanning For computing BC based scorefunction Detecting tumor by bounding a box around the tumor cell and plotting graphicalinterpretation

Now every methodology will be explained clearly in this part for understanding the accuracy of detection. First watershed segmentation methodology will be discussed under the point of 2.1, Second pixel intensity based methodology will be discussed under the point of 2.2 and bounding box methodology will be discussed under the point of2.3. 2.1 Watershed segmentation basedmethodology: A. Considering MRI image as ainput:

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 To read an image for the purpose of processing ‘imread’ command of MATLAB will have to use. The example reads one of the sample images ‘ta1.bmp’ and stores it in an array named I. I = imread('ta1.bmp'); If the file of sample image is not in the actual folder, then the full path with the filename will have to specify. The string ‘bmp’ describes the format of the file by its standard file extension. For example, 'png' is specified for Portable Network Graphics Format files. If the command of imread is not capable to find a file named 'filename', it has a file named 'filename.fmt'.

This is the main image of the brain which have a tumor. Tumor will have to detect in this image of brain.Fig-2 is the representation of considering MR image as an input. B. Gray scaleConversion: J = rgb2gray(I); MRI means Magnetic Resonance Imaging. MRI is as magnetic field and radio frequency rather than ionizing radiation used in XRAY and CT. In calculating, a grayscale image is an image where the value of every pixel is a sample that can carries only deepness of information. Images of this category, known as black and white are composed by the shades of gray, which vary from white to the strongest black at the weakest intensity. Grayscale is nothing but shades of gray without any apparent color. The darkest possible shade is black, which is denoted the total amount absence of transmitted or reflected light. The lightest possible area is white. The total reflection or transmission of light at all visible wavelength should be displayed. Fig-3 is the representation of gray scale conversion. C. Application of High passfilter: H = fspecial('average'); a= imfilter(J,H,'replicate'); To make an image look sharper used high-pass filter is used. In digital image processing high-pass filters are used. It also used to perform image modifications, noise reduction, enhancements etc. Design of High-pass filter design can be done either in the spatial domain or in the frequency domain. High pass filter also increase the brightness of center pixel to neighboring pixel. Fig-4 is the representation of high pass filter application. D. Application of Medianfilter: G = medfilt2(J,[11 11]); Many filter are used to remove the noise from the images. In this types of filter pixel’s value is replaced with its neighborhood value. For removing noise from the image median filter are used. The value of pixel is concluded by the median of the neighboring pixels. Fig-5 is the representation of median filter applications. E. Thresholdsegmentation: The simplest method of image segmentation can be considered as the threshold method, which is a clip level based method.

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 level = graythresh(G); By selecting multiple-levels, the values which are acquired are known as threshold values. Selecting these threshold values is the mandatory part of this method. Now-a-days in the industrial fields some other methods are applied. Like, the Otsu’s method (maximum variance), maximum entropy method. In computer vision the process at which a digital image is divided into multiple segment is called Segmentation. These multiple sets of pixels are also known as super pixels. Simplification is the objective of this process. It converts the image into an easier representation which is also easier to observe the right state of human brain. Detecting and naming a label to every pixel of an image can be done by image segmentation. It also helps detecting lines, curves in the images. The work of a set of segments is covering the whole image. This is the outcome of image segmentation. The pixels are different in texture, color or intensity but for an overall point of view they are closer in characteristics. Marching cubes is a form of interpolating algorithms which helps reconstructing a 3D picture. This can be done only if the image segmentation is ended up. Fig-7 is the representation of threshold segmentation. F. Watershedsegmentation: L = watershed(BW); Depending on the intensities the grouping of pixels for an image can be done and watershed method does it in its best way. It is an excellent method for partitioning a tumor from an image. It can be considered as an operation tool which has both mathematical and morphological operation. It is used as an input segmentation method, but mostly used for monitoring the output result. The problem of suffering from over and under segmentation is the main cause of it.Fig-6 is the representation of watershed segmentation.

Morphology is a large set of image processing activities that process images based on shapes. Morphological operations apply an input image and creating an output image of the same size. In morphological operations, the value of every pixel in the output image which compare the corresponding pixel in the input image with its neighbors. To construct a morphological operation that is conscious of specific shapes in the input image by choosing the shape and size of the neighborhood. Morphological activities are erosion and dilation. Erosion removes pixels to the object boundaries of an image while dilation adds pixels to the boundaries. The number of pixels removed or added from the image depends on shape and size of the element used to process the image. In the morphological erosion and dilation activities, the state of any given pixel in the output image is determined by using a rule to the respective pixel and its neighbors in the input image. Fig-8 is the representation of morphological operation.

H. Detection of tumor as anoutput: Finally output is Fig-9 displayed through an image. The tumor affected part is marked as a white portion and the other’s are considered as a black portion. The white portion has the maximum intensity level and the black portion has the minimum intensity level. The value of each white pixel is 11111111 (255) and the value of each black pixel is0(0). 2.2 Pixel intensity basedmethodology: A. ImageAcquisition: To read an image for processing the ‘imread’ command have been used. The example reads one of the sample images ‘ta1.bmp’ and stores it in an array named I. I = imread(' a1.bmp');

Between many segmentation methods two basic and useful methods are: i.

ii.

G.

As a marker the computed local minima of image gradient is chosen. When this is done then in the next step merging isdone. By using markers watershed transformation utilizes the specifically defined marker positions. Using this marker the position of tumor can be obtained by automatic use of morphologicaltools. Morphologicaloperation: se = strel('disk',1);

Images are achieved by using MRI scan. These scanned images are displayed in a two dimensional (2-D) matrices having pixels as its elements. These types of matrices are dependent on matrix size and its field of view. Images are saved in MATLAB and displayed it as a gray scale image which size are 256*256. These gray scale image are ranging from 0 to 255, where black color defined by 0 and pure white color defined by 255. Between this ranges are vary in intensity from black to white. Fig-10 is the representation of image acquisition.

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 B. Grayscaleimages: Using ‘rgb2gray’ command in MATLAB MRI Images are converted into gray scale images. J = rgb2gray (I) Grayscale is nothing but shades of gray without any apparent color. The lightest possible shade is white which is denoted the total transmission or reflection of light and the darkest possible shade is black denoted the total absence of transmitted or reflected light. Because of the above reasons first of all MRI images are converted into grayscale image. Grayscale is the process to make an image totally colorless, which only have black and white color. This lightness method averages the most distinguished and least distinguished colors: (max(R, G, B) + min(R, G, B)) / 2.The average method simply find the average values of (R + G + B) / 3. The more advanced version of the average method is luminosity method. It also find the average of the values but it forms a weighted average to estimate for human thought. Human eye is sensitive to green than the other colors. The formula of luminosity is (0.21 R + 0.07 B+ 0.72 G). Fig-11 is the representation of gray scale conversion.

‘im2bw’ produces binary images from intensity or RGB images. If it is not already an intensity image by doing this conversion the input image is converted to a grayscale format. Then the grayscale image is converted to binary by fixing threshold. The binary image BW has a value of 0 which denotes to black for all pixels in the input image with luminance less than threshold level and 1 that denotes to white for all other pixels. BW = im2bw (K2, level) converts the intensity image K2 to black and white.Fig-13 is the representation of binary image.

E. Area of the head crosssection: In order to calculate the area of the head cross section of an image, built-in function ‘regionprops’ has been used. Stat = regionprops(BWfill,'Area') The image BW_k2, it can have any dimension. By using ‘For–loop’ the total volume of head is calculated. Using ‘bwperim’ command which can be returned to a binary image containing only the perimeter pixels of objects in the input image. For loop will calculate the total area of head crosssection. It will show the boundary of the brain.

C. Adjusting the MR image using peak intensitylevel: In order to adjust the gray scaled image a built-in function “imadjust” have been usedK1 = imadjust(J,[0.55 0.8],[]); Intensity adjustment is a process for mapping an image's intensity level to a new range. In addition to decreasing or increasing contrast, a wide variety of other image improvements with imadjust can be performed. By this “imadjust” command, intensity of image can be decreased or increased. In this formula, J is the new value which is used to increase the contrast of image K1. Fig-12 is the representation of adjusted image.

D. Conversion of MRI image into binary image to detect the boundary: In order to convert the image of peak intensity intobinary image abuilt-infunction ‘im2bw’ have beenused. K2 = imadjust(J,[0.1 0.5],[]); BW =im2bw (K2, level);

head_area =0; for i = 1:kk head_area = head_area+ stats (i).Area; end Fig-14 is the representation of head cross section.

F. Conversion of adjusted image into binary image to detect the boundary of head crosssection: In order to detect the boundary of head cross section of the image, built-in function ‘im2bw’ has been used. BW_K1 = im2bw (K1, level) This function converts a image into a binary image again. After that morphologically binary image is opened using ‘bwareaopen’ command. BW_2 = bwareaopen(BW_K1, tuning+5); It removes all connected components that have fewer than tuning+5 pixels from a binary image that produce another binary image BW_2.

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 G. Detecting the tumor affected region by eliminating boundary:

‘for’ command executes block of code specified number of times and ‘end’ command terminate block of code indicate last array index. Fig-17 is the focused of tumor area.

In order to detect the tumor affected area of an image, two subtracting process have been usedFirst -when boundary area is open. Second -when image is processed by Imdilatecommand. sub1 = BW2 –BWsdil; BWsdil = imdilate(BWoutline, SE); The argument ‘SE’ is an array of structuring element objects, returned by the ‘strel’ function. If ‘BW outline’ is logical and the structuring element is also flat, ‘imdilate’ performs binary patulousness; otherwise it performs grayscale patulosion. If SE is a layout of structuring elements, ‘imdilate’ executes multiple patulosions of the input image, using each structuring element in SE in succession. BWoutline = bwperim(sub2);

2.3 Bounding box methodology based on Bhattacharyya coefficient: A. As an input taking a MRIimage: For processing the ‘imread’ command have been used to read an input image. ‘imread’ command reads a Grayscale or Color image from the specified filename. If the imread can’t find a file named ‘filename’, it looks for a file named ‘filename.fmt’. Filename must be a string and .fmt is a string that specifies the format of the file. If the file is not in the current path, specify the full pathname of the location on the system. The output of ‘imread’ is an array containing the image data, when the file contains a grayscale image, it will be an M-N array. Again when the file contains a color images, it will be an M-N-4 array. This is read the sample images ‘t.bmp’ and it stored in an array namedI.

Fig-15 is the representation of tumor detection. I=imread(‘t.bmp’); H. Performance parameter of tumor affected image: For the technique of image processing performance parameter image profile and image histogram have been used. Image is represented by ‘imhist’ command of MATLAB for knowing about image histogram of every processed image. It is a chart that shows the distribution of intensities in a grayscale image. ‘Imhist’ command creates a histogram plot by making a number of equally spaced bins, each representing a range of data. Then calculates the number of pixels within each range byit. Again ‘improfile’ computes the intensity values along a multiline path in an image. It selects spaced points along the path is marked and to find the intensity level for each point uses interpolation.Fig-16 is the representation of the performance parameter of tumor affected image.threshold segmentation. Fig-16 is the representation of the performance parameter of tumor affected image. I.

Tumor Areacalculation:

Largest elements along different dimensions of an array of a tumor is calculated by using ‘max’ command in MATLAB. For calculating the area of tumor cross-section a loop is used. fori = 1: kk tumor_area = tumor_area+stats(i).Area ; end

By its standard file extension, the text string ‘.bmp’ specifies the format of the file. The format files can be specified by ‘bmp’ for windows bitmap. As it is already mentioned that the return value of I is an array containing the images data. It will be an M-N array, if the file contains a grayscale image I. I is an M-N-3 dimensional array, if the file contains a true color image. I is an M-N-4 dimensional array, if TIFF files taking on color images that use the CMYK color space. On bit-per-sample of the image data, the class of ‘A’ depends. Which is rounded to the next byte boundary. Fig-18 is the representation of MRI image as an input. B. Normalize theimage: Normalization is an important pre-processing step specially for the analysis of Magnetic Resonance Images (MRI) of human brains. Intensity normalization takes a very significant role for automatic image segmentation and classification is done on the basis of their assumptions regarding the intensity distributions on a standardized intensity range. There are many approaches for intensity normalization. Nyul and Colleagues invented one of the fast and accurate approach. A comprehensive validation of this procedure in real clinical domain is used for intensity dissimilarity correction that is responsible for scanner-specific antiquities. MRI volumes can be affected from various causes such as data dissimilarity resulting from multi-site and multiscanner antiquities. The presence of multiple sclerosis lesions and the stage of disease movement in the brain can also affect

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 MRI volume. The usefulness of the normalization in rendering under the distributional presumption of segmentation approaches have been estimated. Again intensities, that are more homogenous for the same types of tissue while simultaneously resulting in the better tissue type separation by using the distributional divergence criteria have been done. The convenience of the piece-wise linear method have been demonstrated on the task of multiple sclerosis lesion segmentation against a linear normalization method over three image segmentation algorithms:

The Bhattacharyya coefficient (BC) is an estimate measurement of the amount of overlap between two samples. To determine the relative closeness of this two samples being considered, this coefficient can be used. Calculating the Bhattacharyya coefficient involves a basic fundamental form of integration of the overlap of this two samples. The interval of the values of this two samples is split into chosen number of partition and the number of members of each samples in each partition is used in the following formula,

i. A standard Bayesianclassifier ii. An outlier detection basedapproach iii. A Bayesian classifier with Markov Random field (MRF) based onpost-processing. Fig-19 is the representation of normalize image. C. SkullDetection: In order to detect Skull a loop is constructed to find out the identity matrix of the image. For this purpose image is converted to binary and a threshold level is fixed. Converting this level to a matrix a special type of loop (i.e. for loop) is used to calculate the total number of pixels. After that using “double” command image is filtered to double precision. Performing that image is scaled by using gray color map. Then aspect ratio of image specially x-y and z axis are automatically adjusted and plotted in a box to fit lightly. Defuzzify membership function “centroid” is used to detect the center of the images. Image registration technique is applied to rotate the image and finally Skull is detected.Fig-20 is the representation of skull detected image. D. Maskcreation:

pi qi

Where, Samples are p and q, Number of partitions is nand The numbers of members of samples p and q in the ithpartition are piand qi Fig.1 explains the notations. In Fig.1 I and R represent the object and the source images respectively, which have same height hand same width w. The rectangular area D = [lx, ux]×[ly, uy] illustrates the area of tumor or edema that are looking for tumor between these images I and R. FBB algorithm finds the rectangle area D, which have four parameters have to know i.e. lx, ux, ly and uy in two linear passes of the source and object image. In a vertical sweep first it finds the ly and uy values and then finds lx and ux in a

For creating mask new variable “STATS” and “Midx” have been declared. Under STATS variable a new command “regionpnp” have been used. Which determine the property of image region. All pixels should be considered as non-zero matrix. Then the result of “STATS” variable is rounded to the nearest integer. For the purpose of filtering in an accurate way image was rotated to 900 either clock wise or anti clock wise direction. For masking clearly “Midx” variable have been selected. It’s range is fixed from -1 to 1. Then the value is plotted to rotated or registrated image. After that Left part is masked. Again for masking Right part, “Midx” variable has again selected. But the range is up to end. It is also plotted to rotated/registrated image. Finally brain tumor part has been masked to cross match the similarity between two parts. Fig21 is the representation of creation mask. E. Vertical and Horizontal Scanning For computing BC based scorefunction:

Fig.1 (a) Finding violation D from test image I using reference image R. (b) Energy function plot.

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 horizontal sweep over the object and source images. A score function is used to the FBB algorithm in each sweep. Here the vertical score function is described here because of the horizontal score function coincides with the vertical score function which applied to the interchange of the images. Let T(l) is the “top” sub-rectangles of the image and B(l) is the “bottom” sub-rectangles of the image, separated at a distance l from the top to bottom of the image:-

According to the figure-1 (a) T(l)=[0, w]×[0, l] and B(l)=[0, w]×[l, h] Vertical score function can be defined as: E(l)=BC(PIT(l) ,PRT(l))−BC(PIB(l),PRB(l) )

(1)

Where, T(l)

PI is the normalized intensity histogram of the image I within the area T(1). PRT(l), PIB(l) and PRB(l) are defined respectively. BC(a, b)= ∑i√ {a(i) b(i)}∈[0, 1] denotes the Bhattacharya coefficient between two normalized histograms a(i) and b(i) by indicating a histogram bin. The Bhattacharya coefficient calculates the affinity in the midst of two normalized intensity histograms. BC has already used very successfully in different computer perspective applications, such as edge detection[21], object tracking[22], registration[23].

in Fig-22. Fig-22(a, a') shows that a tumor on the left side and right side of the brain on two T1C MRI slices respectively. After automated global thresholding and subsequent postprocessing Fig-22(b, b') shows that the cranium border is detected. Now the cranium border is rounded by an allegory. By computing angle between the main axis of allegory, cranium border and vertical axis is detected and the angle of the skull is founded. Then the cranium is revolved by a precise angle that’s why the major axis becomes vertical. After that a vertical line is drawn (i.e. consider an approximate line of symmetry) through the centered of the cranium mask (as Fig22(b, b')) that splits the whole image into two pieces: the true side of ALOS is considered as the mentioned image R and the left side of ALOS is viewed as the trial image I. Fig-22(c, c') the steep account function plots against the distances from the top of the image. Both plotting figures show the risingdropping- rising nature of the score plots. The positive peak and the negative peak corresponding to the major, minor, true and left bounds of the bounding box are superimposed in Fig.22(a, a') and are also displayed by burnet dots in Fig.22(c, c') and 22(d, d'). To estimate whether the true half image or the left half contains the tumor, the moderate severity within the bounding boxes have been placed on left and right sides. Finally the tumor is disclosed where the bounding box is displayed. Fig-22 is the representation of tumor detected image by a box and a graphical interpretation is plotted in the figure.

II.

SIMULATIONRESULTS

3.1 Simulation results of watershed segmentationbased methodology: Here –

BC(PIT(l),PR

Original image

) = ∑i√ {(PIT(l)(i))(PR

T(l)

T(l)

(i))}

[2]

(i))(PRB(l) (i))}

[3]

and BC(PI B(l), PR B(l))=∑i√{(PI

B(l)

Fig.2:- Original MRI image

When the established scatter diagram are the same, then BC = 1 and when two established scatter diagram are completely different from each other, then BC = 0. Thus the scoring rule E(l) (from equation -(1)) is broad when the domain region T(1) in two images are akin and the base regions B(1) are totally contrasting.

F. Detecting tumor by bounding box around the tumorcell and plotting graphicalinterpretation: Fig.3:- Gray scale Conversion

The step by step tasks associated with automated localization of a square box on an MRI slice which is shown

Proceedings of 14th Global Engineering and Technology Conference 29-30 30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9

Fig.4:- Application of high pass Filter

Fig.8:- Morphological operation

Fig.9:- Tumor detection as an output Fig.5:- Application of Median filter

3.2 Simulation results of pixel intensity basedmethodology:basedmethodology:

Fig.6:- Threshold segmentation

Fig.10:- Image acquisition

Fig.7:- Watershed Segmentation

Fig.11:- Grayscale Conversion

Proceedings of 14th Global Engineering and Technology Conference 29-30 30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 4

originalhistogram

x10 12

numberofpixels

10

8

6

4

2

0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Fig.16:- Performance parameter of tumor affected image Fig.12:- Adjusting the MRI image using peak intensity level

Output command of MATLAB window is: The area of the tumor is:1.7644 percent of the area of the head cross-section. section.

Fig.13:- Conversion of MRI image into binary image to detect the Boundary

Fig.17:- Final output of the processed image and it’s histogram representation

3.3

Simulation results of bounding box method based on Bhattacharyyacoefficient:

Fig.14:- Conversion of adjusted image into binary image to detect the boundary of head cross section

Fig.18:- MRI image as an input

Fig.15:- Detecting the tumor affectedregion

Fig.19:- Normalize the image

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9

a'

b'

c'

d'

Fig.20:- Detect Skull

Fig. 22(a, a') Output of MR image (b, b') Line of symmetry (c, c') Score plot for vertical direction (d, d') Score plot for horizontal direction

III. OVERALLCOMPARISONANDDISCUSSION

Fig.21:- Mask creation

a

a

b

c

d b

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 intensity based technique and finally bounding box technique have been reviewed. It has been tried to show a comparison among them. Through using these technique it is possible to detect tumor but in future some extra features will be added. The above techniques are not only capable to say which part of brain has been affected by tumor but also in future any one of the technique will capable to say which part of the brain has been affected bytumor.

REFERENCES c

Fig.23 (a) Output of watershed segmentation based methodology (b) Output of pixel intensity based methodology (c) Output of bounding box based on BC Comparing three techniques it is seen that watershed segmentation technique can detect tumor but its detection is not appropriate. Although it can detect some part of tumor affected region but whole part is not detected. Though in this paper it is seen that watershed segmentation method have detected tumor in an efficient way but when this method is applied to other high resolution image, it is seen that this methodology is not capable to detect the tumor in an appropriate way. It can detect only a small part of tumor or in some cases it detects an inappropriate portion as a tumor. On the contrary pixel based intensity technique is capable to detect tumor in appropriate region and it will also show the area and image profile of tumor affected image but it’s main limitation is it can detect tumor only from top view. It is observed that when MR image from other view is placed as an input under this methodology it can’t detect tumor in an efficient way. As this method detect tumor on pixel intensity so when it found some portion of MR image in high pixel intensity it considered that portion as a tumor. But clinically high intensity portion of pixel should not be considered as a tumor in all cases. It can’t detect tumor from other view (i.e. left view, right view and back view). Finally bounding box technique is applied. It is found that this technique is more efficient than previous two techniques. Bounding box technique is more efficient in the sense that it can detect tumor from any side view (i.e. left side, right side and back side view).Besides previous techniques are applicable only for top view but this technique is applicable for any view. It will detect the tumor by bounding a box showing its graphical interpretation.

IV. CONCLUSION In this paper various algorithm for detecting tumor have been reviewed and tumor detection technique from MRI have been investigated. First watershed segmentation, second pixel

[1] http://www.webmd.com/brain/brain-diseases#1 [2] http://www.webmd.com/cancer/brain-cancer/brain-tumors-in-adults [3] http://www.nhs.uk/Conditions/brain-tumours/Pages/Introduction.aspx [4] http://www.radiologyinfo.org/en/info.cfm?pg=bodymr [5] https://medlineplus.gov/ency/article/003791.htm [6] Rajesh C. Patil, Dr. A. S. Bhalchandra, Brain Tumour Extraction from MRI Images Using MATLAB ,International Journal of Electronics, Communication& Soft Computing Science and Engineering, ISSN: 2277- 9477, Volume 2, Issue 1 [7] AnamMustaqeem, Ali Javed, Tehseen Fatima AnEfficientBrainTumor Detection Algorithm Using Watershed &Thresholding Based Segmentation, I.J. Image, Graphics and Signal Processing, vol.10 PP.3439,2012 [8] W. Gonzalez, “Digital Image Processing”, 2nd ed. Prentice Hall, Year of Publication 2008, Page no 378 [9] S. Murugavalli, V. Rajamani, “A high speed parallel fuzzy c-mean algorithm for brain tumour segmentation”,” BIME Journal”, Vol. no:06, Issue (1), Dec., 2006 [10] Mohamed LamineToure, “Advanced Algorithm for Brain Segmentation using Fuzzy to Localize Cancer and Epilepsy Region”, International Conference on Electronics and Information Engineering (ICEIE 2010),Vol. no 2. [11] Dr.G.Padmavathi, Mr.M.Muthukumar and Mr. Suresh Kumar Thakur, “Non linear Image segmentation using fuzzy c means clustering methodwiththresholding for underwater images”, IJCSI International Journal of Computer Science Issues, Vol. 7, Issue 3, No 9, May 2010 [12] MateiMancas, Bernard Gosselin, Benoîtmacq, “Segmentation Using a Region Growing Thresholding” T .Logeswari and M.Karnan “An improved implementation of brain tumor detection using segmentation based on soft computing” Journal of Cancer Research and Experimental Oncology Vol. 2(1) pp. 006-014, March, 2010 [13] Md. Abdullah Al Mahmud, A.H.M Zadidul Karim & Md. Mashiur Rahman , “Brain Tumor Detection using MR Images Through Pixel based Methodology” Global Journal of Computer Science and Technology: Graphics & Vision ,Volume 15 Issue 4 Version 1.0 Year 2015 , Publisher: Global Journals Inc. (USA) ,Online ISSN: 0975-4172 & Print ISSN: 0975-4350 [14] Devos, A, Lukas, L., “Does the combination of magnetic resonance imaging and spectroscopic imaging improve the classification of brain tumors?” On Page(s): 407 – 410, Engineering in Medicine and Biology Society, 2004. IEMBS '04. 26th Annual International Conference of the IEEE, 1-5 Sept. 2004. [15] Farmer, M.E, Jain, A.K. , ―A wrapper-based approach to image segmentation and classificationǁ, Page(s): 2060 - 2072 , Image Processing, IEEE Transactions on journals and magazines, Dec. 2005. [16] Gopal,N.N. Karnan, M. , ―Diagnose brain tumor through MRI using image processing clustering algorithms such as Fuzzy C Means along with intelligent optimization techniques ― , Page(s): 1 – 4, Computational Intelligence and Computing Research (ICCIC), 2010 IEEE International Conference, 28-29 Dec. 2010. [17] Joshi, D.M.; Rana, N.K.; Misra, V.M. i ,ǁ Classification of Brain Cancer using Artificial Neural Network ― , Page(s): 112 – 116, Electronic

Proceedings of 14th Global Engineering and Technology Conference 29-30 December 2017, BIAM Foundation, 63 Eskaton, Dhaka, Bangladesh ISBN: 978-1-925488-60-9 Computer Technology (ICECT), 2010 International Conference, 7-10 May 2010 [18] Ming niwu,chia-chen Lin and chin-chenchang, ―Brain Tumor Detection Using Color-Based K-Means Clustering Segmentationǁ , Page(s): 245 – 250 , Intelligent Information Hiding and Multimedia Signal Processing, 2007. IIHMSP 2007. Third International Conference,26-28 Nov. 2007 [19] BaidyaNathSaha, Nilanjan Ray, Russell Greiner, Albert Murtha, Hong Zhang, “Quick detection of brain tumors and edemas: A bounding box method using symmetry”, Computerized Medical Imaging and Graphics 36 (2012) 95– 107 [20] Brummer ME, Mersereau RM, Eisner RL, Lewine RJ. Automatic detection of brain contours in MRI data sets. IEEE Transactions on Medical Imaging 1993;12(2):153–66

[21] Comaniciu D, Ramesh V, Meer P. Real-time tracking of non-rigid objectsusingmeanshift.ProceedingofCVPR2000;2:142–9 [22] Konishi S, Yuille A, Coughlan J, Zhu S.” Fundamental bounds on edge detection: an information theoretic evaluation of different edge cues.” IEEE CVPR 1999:573–9 [23] Viola P, Wells W. “Alignment by maximization of mutual information”. International Journal of Computer Vision 1997;24(2):137–54