Brain Tumor Detection from Magnetic Resonance Imaging using Image Processing Final Year Project report submitted to Central Institute of Technology, Kokrajhar in partial fulfillment for the award of the degree of Bachelor of Technology in Computer Science and Engineering by Manashi Saharia,Kwmdwn Daimary,Amit Daimary (GAU-C-15/065, GAU-C-15/087, GAU-C-15/L-292) Under the supervision of ‘Mr. Mithun Karmakar (Assistant Professor) Computer Science and Engineering Central Institute of Technology, Kokrajhar 8th Semester, 2019 May 14, 2019DECLARATION We certify that (a) The work contained in this report has been done by us under the guidance of our supervisor. (b) The work has not been submitted to any other Institute for any degree or diploma. (c) We have conformed to the norms and guidelines given in the Ethieal Code of Conduct of the Institute. (d) Whenever we have used materials (data, theoretical analysis, figures. and text) from other sources, we have given due credit to them by citing them in the text of the thesis and giving their details in the references. Further, we have taken permission from the copyright owners of the sources, whenever necessary. Date: May 14, 2019 Manashi Saharia,wmdwn Daimary,Amit Daimary Kresnibor Painagy GAU-C-15/065, GAU-C-15/087, GAU-C-15/L-292 bad Darron — Place: KokrajharCOMPUTER SCIENCE AND ENGINEERING CENTRAL INSTITUTE OF TECHNOLOGY, KOKRAJHAR KOKRAJHAR - 783370, INDIA CERTIFICATE This is wo certify that the project report entitled “Brain Tumor Detection from Magnetic Resonance Imaging using Image Processing” submitted by Man- ashi Saharia,Kwmdwn Daimary,Amit Daimary (Roll No. GAU-C-15/065, GAU-C-15/087, GAU-C-15/L-292) to Central Institute of Technology, Kokrajhar towards partial fulfillment of requirements for the award of degree of Bachelor of Technology in Computer Science and Engineering is a record of bona fide work carried out by them under my supervision and guidance during 8th Semester, 2019. wen Mr. Mithun Karmakar (Assistant Professor) Date: May 14, 2019 ‘Computer Science and Engineering, Place: Kokrajhar Central Institute of Technology, Kokrajhar Kokrajhar - 783370, IndiaBONAFIDE CERTIFICATE This is to certily that the project titled "Brain Tumor Detection from Magnetic Resonance Imaging using Image Processing” is a bonafide recorded of the work done by Manashi Saharia(GAU-C-15/065) Kwmdwn Daimary)(GAU-C-15/087 Amit Daimary(GAU-C-15/L-292) in the partial fulfillment of the requirements for the awards of the degree of Bachelor of technology in Computer Science and Engineering of the CENTRAL INSTITUTE OF TECHNOLOGY, KOKRAJHAR during the year 2018-19. \. Mr. Rahjan M Dr. Pankaj Pratap Singh (Assistant Professor) Project Co-ordinator Head Of the Department Project viva voce ield on 15g Examiner aAbstract Name of the student: §Manashi Saharia,kwmdwn Daimary,Amit Daimary Roll No: GAU-C-15/065, GAU-C-15/087, GAU-C-15/L-292 Degree for which submitted: Bachelor of Technology Department: Computer Science and Engineering Thesis title: Brain Tumor Detection from Magnetic Resonance Imaging using Image Processing ‘Thesis supervisor: Mr. Mithun Karmakar (Assistant Professor) Month and year of thesis submission: May 14, 2019 ‘The Brain tumor is affecting many people worldwide. It is not only limited with the old age people but also detected in early age. Brain Tumor is the abnormal growth of cell inside the brain cranium which limits the functioning of the brain. Early detection of the brain tumor is possible with the advancement of machine learning (ML) and image processing, Medical image processing is the most challenging and emerging field today. Many scientist and researchers are working to develop and add more features to this tool. MRI Imaging plays an important role in brain tumor for analysis, diagnosis and treatment planning. It’s helpful to doctor for determine the previous steps of brain tumor. Brain tumor detection using MRI images is a chal- lenging task, because of the complex structure of the brain. This project is about detecting Brain tumors from MRI images using an interface of GUI in Matlab. MRI image can be processed and the brain tumor ean be segmented using various image iiiiv segmentation techniques. The process of identifying brain tumors through MRI im- ages can be categorized into four sections of image processing: pre-processing, image segmentation, feature extraction and image classification. In this project we have used various algorithms for the partial fulfillment of the requirements to arrive at the best results that can help us to detect brain tumors in early stage.. Keywords: MRI, Image Processing, Matlab.Acknowledgements We would like to express our gratitude and appreciation to all those who gave us the opportunity to complete this report. A special thanks to our final year project coordinator, Mr. Ranjan Maity Asst. Professor,CSE Department,CIT Kokrajhar, whose help, stimulating suggestions and encouragement, helped us to coordinate our project especially in writing this report We would also like to acknowledge with much appreciation the crucial role of the staff of Computer Science and Engineering Lab, who gave the permission to use all required machinery and the necessary material to complete the project. Last but not least, many thanks go to the head of the project, Mr. Mithun Karmakar, Asst. Professor,CSE Department,CIT Kokrajhar who have given his full effort in guiding the team in achieving the goal as well as his encouragement to maintain our progress in track. We would like to appreciate the guidance given by other supervisor as well as the panels especially in our project presentation that has improved our presentation skills by their comment and tips. MANASHI SAHARIA Manarte' Vohosia, (CG au-- 1s/06s) KWMDWN DAIMARY Kumdon Deine (Gou~<-val 0%) AMIT DAIMARY Arut Dornan, Ctiav-e 5] l-292)Contents Declaration i Certificate ii Abstract iii Acknowledgements v Contents vi List of Figures viii Abbreviations ix 1 INTRODUCTION 1 1.1 Introduction . . 1 1.2 Literature Survey... . . 2 1,3 Problem Statement 4 1.4 Motivation and Objective 4 2 PRINCIPLE 5 24 ‘Methadblogy Used «cc ccccc ce sucseeveeeseuanann 5 3 EXPERIMENTAL RESULTS 20 3.1 Input MRI Image... 2 ee 20 3.2 Preprocessing using Median Filter,Ganssian Filter and Sobel Edge Detection ove cece success vee oe 2 3.3 Segmentation and Feature Extr: Mt 21 34 Classification using Artificial Neural Network... . . oe 3.5 Gradient and Validation Check a ORERERR ERIS 3.6 Error Histogram Graph . 3.7 Confusion Matrix . 38 ROC Curve . . . 39° Accuracy We have Obtained fram. Al viContents vii 4 COMPARATIVE STUDY 29 4.1 Comparative Study of Different Brain Tumor Detection and Classifi- cation Techniques Using MRI images... 0. eee ees 29 Conclusion a Future Work 31 Bibliography 32 PROTECT CODING 1S ATTACHED IN ADIGE AT THE Back COVERList of Figures 21 22 2.3 4 3.1 Steps for Image Processing, Median Filter a)Shows pixel intensity value before filtering a Ustons pixel intensity value after filtering . Ganssian Filter Example Discrete Wavelet Decomposition . Working of GLCM ...... 3 Architecture of Neural Network cee Artificial neuron and Biological neuron... Accuracy Table . .Abbreviations DWT GLCM ANN SVM Magnetic Resonance Image Discrete Wavelet Transformation Gray Level Co-occurence Matrix Artificial Neural Network Support Vector MachineChapter 1 NTRODUCTION 1.1 Introduction Brain tumor is the most commonly occurring malignancy among human beings, so study of brain tumor is important. It is the major cause for the increase in mortality among children and adults. Tumor is defined as the abnormal growth of the tissues. Brain tumors are generally classied into two types benign and malignant tumors. MRI imaging plays an important role in brain tumor for analysis, diagnosis and treatment planning. It is helpful to doctor for determining the previous steps of brain tumor(l]. The detections of brain tumor using MRI images is a challenging task becanse of the complex structure of the brain, MRI images provide better results than CT scan, Ultrasound and X-ray because it is an advanced medical imaging technique and it uses powerful magnets to produce high resonance images of all parts of the body. These MRI images can be processed and one can detect the brain tumor using image processing technique by forming an automatic detection process tising various algorithms because the manual detection of tumor from MRI images may give human error. In this project our focus will be mainly to improve the existing approaches of image processing or to design a better approach for the detection of the tumor.Chapter 1:Introduction 2 1.2 Literature Survey 1.Parveen, Amritpal Singh in 2015(2] proposed algorithm is a combination of SVM and fuzzy c-means, a hybrid technique for prediction of brain tumor. Here, the im- age is enhanced using contrast improvement, and mid-range stretel. Double thresh- olding and morphological operations are used for skull striping. Fuzzy ¢-means (FCM) clustering is used for the imiage segmentation. Grey level run length matrix (GLRLM) is used for extraction of feature. Then, Linear, Quadratic and Polyno- mial SVM technique is applied to classify the brain MRI images. Real data set of 120 patients MRI brain images have been used to detect ‘tumor’ and ‘non-tumor’ MRI images. The SVM classifier is trained using 96 brain MRI images, after that the remaining 24 brain MRI images was used for testing the trained SVM. SVM classifier with Linear, Quadratic and Polynomial kernel function give 91,66 %,83.33 % and 87.50 % accuracy. respectively. 2.Astina minz, Prof. Chandrakant Mahohiya in 2017[3] proposed an effective au- tomatic classification method for brain MRI on the dataset of 50 MRI images is projected using the Adaboost machine learning algorithm. The proposed system consists of three parts such as Preprocessing, Feature extraction and Classification. Preprocessing has removed noise in the raw data, it transform RGB image into grayscale, median filter and thresholding segmentation is applied, For feature ex- traction by using GLCM technique 22 features were extracted from an MRI. For classification boosting technique used (Adaboost). It gives 89.90% accuracy and result in normal brain or in Malignant or Benign type of tumor. In future work, we can work of quadratic and polynomial kernel function. The accuracy of the system will be increased by increasing training database images. Also the system can be implement for different types of classes like Glioma and Meningioma. 3. Mukambika P. S., Uma Rani K.[4] in 2017 Proposed Methodology in which Image is processed through: Preprocessing, Segmentation, Feature extraction Clas- sification stages from the dataset of 41 MRI images. In preprocessing, Morphology technique using double thresholding is applied to remove the skull out of the MRI brain images. The present work presents the comparison study of two techniquesChapter L:Introduction 3 used for tumor detection of MRI images. One is based on the Level set method that uses the non parametric deformable models with active contour to segment the brain tumor from the MRI brain images. The other one is the K-means segmenta- tion algorithm. After the segmentation decision making is performed in two stages: Feature extraction using Discrete Wavelet Transform and Gray Level Co-occurrence Matrix, and classification using the Support Vector Machine, Dataset of MRI brain tumor images includes T2 weighted 17 benign and 24 malignant tumor images of different; patients. SVM with Level Set and K-Means segmentation classify image into normal brain, benign or Malignant tumor with 94.12% and 82.35% accuracy respectively. Level Set method gives better results than kmeans segmentation. 4.Rasel Ahmmed, Anirban Sen Swakshar, Md. Foisal Hossain, Md. Abdur Rafiq in 2017[5] proposed method which include stages like image pre-processing, segmen- tation, feature extraction, SVM classification and tumor stage classification using Artificial Neural Network(ANN). In pre-processing three contrast enhancement tech- niques like adjusted, adaptive threshold and histogram imaging using both weiner2 and median? filter is applied. Segmentation is done by TKFCM algorithm which is integration of the K-means and Fuzzy c-means with some modification. Feature extraction is done in two orders. In First order statistic features and in Second order region property based statistic features are derived. Then SVM classify brain MRI image into normal or tumor brain. Brain Tumor stage is classified by ANN classi- fier. The number of the used data for each MRI image of normal brain, malignant tumor, and benign tumor is obtained from 39 images where 3 normal, 9 benign, 17 malignant 1, 6 malignant 11, 3 malignant [, and 1 malignant IV stage tumor brain MRI images. The accuracy of proposed method is 95.4%. 5.K, Sudharani, Dr. T. C. Sarma, Dr. K. Satay Rasad in 2015{6] Proposed Method- ology inchucle methods like Histogram, Re-sampling, K-NN Algorithm, Distance Ma- trix. First, Histogram gives the total number of specified value of pixels distributed in a particular image. Re-sampling re-size image to 629X 839 for proper geometri- cal representation. Classification and identification of brain tumor by using k-NN which is based on training of k. In this work Manhattan metric has applied and calculated the distance of the classifier. The algorithm has been implemented usingChapter 1:Introduction 4 the Lab View, Algorithm has been tested on 48 images. The identification score for all images are about 95%. 1.3. Problem Statement Nowadays we have seen miost of the tumors are life threatening where brain tumor being one of them. As we know that brain tumor can be of any shape, size, location and intensity, therefore it is very difficult to detect tumor and diagnose it. The manual identification of tumor from MRI images is subjective in nature and may vary from expert to expert depending on their expertise and other factors which include lack of specific and accurate quantitative measures to classify the MRI images as it is brain tumor or not. So automated identification of brain tumor from MRI images help in alleviating the major issues and provide better results. Detection of brain tumor from the various symptoms of the patients has always been a major issne for the medical practitioner and pathologist for diagnosis and treatment planning. [t is also a fact that some tests may be time consuming and it gives workloads and difficulty for the pathologists to obtain the accuracy of the presence of the tumor 1.4 Motivation and Objective As the medical practitioner and pathologist face various such types of problems in detecting tumor manually from the MRI image, so there is a need of an automatic detection process. Thus the main aim of our project is to design a framework for automatic detection of the tumor to obtain more accuracy from the imaging dataset which plays a vital role in the diagnosis of tumors by using various image process- ing algorithm in MATLAB. This framework will hopefully help the pathologist to reduce the work-load and minimize human error while maintaining and improving the accuracy to detect the tumor.Chapter 2 PRINCIPLE 2.1 Methodology Used ‘The detection and clusification of tumor from MRI images are a challenging tasie because of the complex structure of the brain. ‘The various stages include pre- processing(enhancement) of MRI images, segmentation of suspicions portion, feature extraction and finally the classification Ficune 2.1: Steps for Image Processing 5Chapter 2:Principle 6 In our project we have written the program code and modeled the output of Brain ‘Tumor in MATLAB R201Ga, Here we have implemented four important stages of image processing that include pre-processing of MRI images, segmentation of suspicious portion, feature extraction and classification. INPUT(MRI IMAGE): The first stage i.e. the image acquisition stage which starts with taking a collection of images from the database. Images are stored in MATLAB will be displayed as a gray scale image. As our MRI DICOM dataset was in .mat file, we have extracted the images using MATLAB code. We have collected our MRI image dataset from: Brain MRI Dataset (BRATS 2015) hittps:/ /in.mathwork.com/matlabcentral/answer/343191-brain mri-cataset-brats-2015. The Brain MRI dataset (BRATS 2015) consists of 2000 images of resolution 200*200 mm of both tumor and non tumor MRI images in DICOM format. IMAGE PRE-PROCESSING:- This is the second stop of image processing where the images are used to enhance the chances of detecting the suspicious region. Finer details of the images are enhanced and noises are removed from the images. Clini- cal MRI when corrupted by noises reduces the accuracy of the images. So various filters are used to remove these noises. It aims to improve the image data by sup- pressing the undesired distortions and enhances some of the image features that will be helpful in further processing. The goal of pre-processing is to remove the noise and to provide Contrast Enhancement to improve the image quality. The functions performed by preprocessing process is - © Gray scale conversion # Noise removal © Contrast Enhancement CONVERSION TO GRAY SCALE:- A grayscale image only consists of gray scale values, but MRI images consist of primary colors (RGB) content. A ‘Gr color is one in which the red, green and blue components all have equal intensity in RGB space and so it is only necessary to specify a single intensity value for each pixel, as opposed to the three intensity values needed to be specified for each pixelChapter 2:Principle 7 in a full color image. When MRI images are viewed, they look like black and white but they contain some primary colors(RGB), So, for further processing of MRI brain image, iv must be converted to perfect gray scale image. FILTERING: processing operations implemented with filtering include smoothening, sharpening, Filtering is a technique for modifying or enhancing an image. Image and edge enhancement. Filtering is a neighborhood operation, in which the value of any given pixel in the output image is determined by applying some algorithms to the values of the pixels in the neighborhood of the corresponding input pixel. MEDIAN FILTER:- In image processing, it is very important to perform some kind of noise reduction method in order to extract maximum informa: tion from the image. In pre-processing module we have proposed median filter. Median filter is very widely used in digital image processing because under cer- tain condition, it preserves edges while removing noise. Median filter is used for noise reduction in general and removal of salt and pepper noise in specific. Usually MRI images contain salt and pepper noise present due to motion arti- facts (movement of patient during scan) and it is desirable to use median filter, It is done for smoothening of MRI brain image. Here we are using median fil- ter to climinate salt and pepper noise. The median filter works by moving through the image pixel by pixel, replacing each value with the median value of the neighboring pixels, The pattern of neighbors is called the “window”, which slides pixel by pixel, over the entire image. In MATLAB we have used “imfilter” command to remove the noise from the image.Chapter 2:Prineiple 8 3,3.3,4.@,5.5,5,10 Ficure 2.2: Median Filter a)Shows pixel intensity value before filtering b)Shows pixel intensity value after filtering, GAUSSIAN FILTER:- In image processing, a Gaussian blur (also known as Gaussian smoothing) is the result of blurring an image by a Gaussian fune- tion. It is a widely used effect in graphics software, typically to reduce image noise and reduce detail. The visual effect of this blurring technique is a smooth blur resembling that of viewing the image through a translucent screen. Math- ematically, applying a Gaussian blur to an image is the same as convolving the image with a Gaussian function. This is also known as a two dimensional Weierstrass transform. Since the Fourier transform of a Gaussian is another Gaussian, applying a Gaussian blur has the effect of reducing the image's high frequency components; a Gaussian blur is thus a low pass filter. The Gaussian distribution in 2-D has the form: 1 _atag? G(z,y) = aot a Where is the standard deviation of the distribution and we also assume that the distribution has a mean of zero, mean(0,0).Chapter 2:Principle 9 FIGURE 2.3: Gaussian Filter Example SOBEL GRADIENT: The Sobel operator is a derivate mask and is used for edge detection. It is used to detect two kinds of edges in an image: Ver- tical Direction and Horizontal Direction. In sobel operator the coefficients of mask are not fixed and they can be adjusted to our requirement unless they do not violate any property of derivative masks. The mask will prominent the horizontal and vertical edges in an image. It works on the principle of mask and calculates difference among the pixel intensities of a particular edge. IMAGE ENHANCEMENT:- Image enhancement provides the details that are obscured and highlight certain features in an image. The fundamental en- hancement needed in the MRI images is the contrast enhancement. Here we have performed image enhancement by removing noise and blur from the im- age. By enhancing the image we will find the actual clear suspicious portion of the tumor. IMAGE SEGMENTATION:- The third stage is image segmentation. Tmage segmentation is the process of partitioning a digital image into multiple segments. Image segmentation will be typically used to locate objects and boundaries in images and it helps to make the analysis easier. It is the process of assigning a label to every pixel in an image. Such pixels with the same label share certain visual characteristics. There are several algorithms for segmentation. However the selection of segmentation method depends on the type of the features to be preserved and extracted. The goal of segmentation is to simplify and change the representation ofChapter 2:Principle 10 an image into something that is more meaningful and easier to analyze. The result of image segmentation is a set of regions that collectively cover the entire image, or a set of contours extracted from the image. Here we have implemented Thresholding method for image segmentation. For intensity based segmentation it is better to use thresholding rather than color based segmentation technique. In thresholding based. segmentation each an every pixel is compared with the threshold value. THRESHOLD SEGMENTATION:- Thresholding technique can be used for MRI brain tumor segmentation. According to intensity/brightness is a simple technique for images which contain solid objects on a background of different, tmt uniform, brightness. Each pixel is compared to the threshold: if its value is higher than the threshold, the pixel is considered to be “fore- ground” and is set to white, and if it is less than or equal to the threshold it is considered “background” and set to black, Varios versions of thresh- olding technique have been introduced that segments MR images by using the information based on local intensities and connectivity. Most of the ex- isting thresholding methods are bi-level, which use two levels to categorize the image into background and object segments. However, MR images have many different parts which make these methods non-applicable, Thus, the loss of information from the image may occur and diagnosis system may mislead physicians in their clinical task. Therefore, multi-level thresholding algorithms have been developed to ensure that all important information from MR images are retained, but they become computationally expensive, because a large no. of iterations would be required for compu global thresholding method is the most suitable image segmentation method to segment a brain tumor from a Magnetic Resonance Image. It selects that gray level value as threshold for which between-class variance is maximised. In general, thresholding algorithms do not use spatial information of an image and they usually fail to segment objects with low contrast or noisy images with varying background. The simplest method of image segmentation is called the thresholding method. This method is based on a clip-level (or a threshold value) to turn a gray-scale image into a binary image. The key of this method is to select the threshold value (or values when multiple levels are selected) ig the optimum threshold. Otsu's A zero matrix of same size of original image matrix is considered. Each pixel value of the image matrix is compared with the threshold point. If the valueChapter 2:Principle uu of pixel is greater than threshold, coordinate of ¢ matrix is assigned a value 255 otherwise 0 is assigned to that. This process is repeated till all the pixel values are compared to threshold point. A grayscale image is turned into a binary (black and white) image by first choosing a grey level T’ in the original image, and then turning every pixel black or white according to whether its grey value is greater than or less than T: A pixel becomes, white if its grey level is T, black if its grey level is T, Matrix is then transformed into image using ‘imshow’ command. FEATURE EXTRACTION:- The fifth stage is the feature extraction. The Im- age Feature Extraction stage is very important in our work where we will use al- gorithm and techniques to detect and isolate various desired portions or shapes of an image. After the segmentation is performed the features can be obtained from it and the diagnosis rule can be designed to exactly detect the tumor. Feature extrac- tion is the transformation of input data into a set of features. It is a key stage of performing the task, which identifies set of features that describe the visual texture of an image. MR Image Segmentation is based on a set of measurable features which are extracted or computed from the image. It is a process of extracting quantita- tive information from an image such as color features, texture, sape and contrast. Here, we have used discrete wavelet transform (DWT) for extracting wavelet coef- cients and gray-level co-oceurrence matrix (GLCM) for statistical feature extraction, DISCRETE WAVELET TRANSFORM:- The features extracted from MRI images by Wavelets are considered to be the best method to extract most empha- sizing pixels present in images to improve results. To decompose data into different frequency components Wavelets mathematical functions will be used and then each component: will be studied having resolutions matched to its degree. For analysis of complex datasets Wavelet is a new powerful mathematical tool. Wavelets pro- vides time-frequency information which is localized in space and perfect tool for pattern recognition task. Discrete Wavelet Transform is an implementation method. for wavelet obeying some defined rules and discrete set of wavelet translations. It is necessary for practical computation to make wavelet discrete. In other words, this transform decomposes the image into mutually orthogonal set of wavelets, whichChapter 2:Principle 12 is the main difference from the continuous wavelet transform(CW'), or its imple mentation for the discrete time series sometimes called Discrete-Time Continuous Wavelet Transform(DT-CWT). The wavelet can be constructed from a sealing func- tion which describes its scaling properties. The restrictions that the scaling functions must be orthogonal to its discrete translations implies some mathematical conditions on them are mentioned everywhere. DWT will be used to extract the wavelet coef- ficient from MRI at different directions and scales. Wavelet decomposition provicles four sub bands. @ Approximated (LL) © Horizontal(LH) © Vertical(HL) © Diagonal(HH) Approximated sub band(LL) is normally further decomposed at different scales while three sub bands (LH.HL,HH) include characteristic of an image. DWT overcomes the drawback of Fourier transform by recognizing point discontinuities in image Discrete Wavelet transform gives local frequency information. We have to extract different features like Contrast Homogeneity, Correlation, Encrgy ete. These features are used as an input to classifier for tumor detection. Figure 2.4: Discrete Wavelet DecompositionChapter 2:Principle 1B GLCM(Gray Level Co-Occurence Matrix):- Texture analysis differentiates nor- mal and abnormal tissues easily for human visual perception and machine learning. It also provides variation between malignant and normal tissues, which may not be visible to human eye. It improves the accuracy by choosing effective quantitative features for early diagnosis. The statistical features were extracted using gray-level co-occurrence matrix (GLCM), also known gray-level spatial dependence matrix (GLSDM), It is an approach that describes the spatial relation between pixels of various gray-level values. In this method, gray-level co-occurrence matrix was ini- tiated and the textural features such as contrast, correlation, energy, homogeneity, entropy and variance were obtained from LL and HL subbands of rst four levels of wavelet decomposition. The GLCM functions characterize the texture of an image by calculating how often pairs of pixel with specific values and in a specified spatial relationship occur in an image, creating a GLCM, and then extracting statistical measures from the matrix. After we create the GLCMs’ using “graycomatrix”, we can derive several statistics from them them using “graycoprops”. These statistics provide information about the texture of an image. WORKING OF GLCM: i.The GLCM described here is used for a series of “second order” texture calculation. Second order means they cousider the relationship between groups of 2 pixels in the original image. ii.It considers the relation between two pixels at a time, called the reference and the neighbor pixel where the neighbor pixel is chosen to be in the east of the reference pixel. One pixel in the x-direction and the other in the y-direction, iii. The gray value relationship of the image are transformed into the co-occurrence matrix space by a given kernel mask such as 3*3,5*5 and so on. iv.Each clement (i,j) in GLCM specifies the mumber of times that the pixel with value i occurred horizontally adjacent to a pixel with value j. v.Once after the GLCM is computed, texture features of the image are being ex- tracted successively. vi.Some Textural features that we have extracted are Contrast, Homogeniety, Energy ete.Chapter 2:Principle “4 Figure 2.5: Working of GLCM Some of the features that are extracted are listed below: Contrast(CONT): Measurement of pixel intensities and its neighbours given by the equation: CONT = > > («=») f(y) 0 0 Energy(ENG): Energy defines the quantitative amount of repetitive pixel pairs. It is the measurement of affinity in an image, given by equation: r= = ENG = ,\|3 > 0.9) p= G=0 Correlation(COR): The measurement of spatial features dependencies be- tween the pixels, given by the equation:Chapter 2:Principle 15 o (P-gp. 4) ~ Mya, Fp q Homogeneity(HOM): Measurement of local uniformity in an image. It is also known as inverse difference moment and contains a single or more range of values to distinguish between textured and non textured, given by the equa- tion: it id I HOM = SF /i(p, inet Entropy(ENT);: It calculates the designated interference of the textural im- age. It is given as: iy Jt ENT = — 3° Sf (p.¢) logsf(p,4) pao gaa Gray-level co-occurrence matrix (GLCM) is 2D histogram in which (p,q)th element is the frequency of event p occurs with q. CLASSIFICATION:- The last stage is the Classication stage. The classication of MRI images is the most challenging task for automatic detection of brain tu- mor from the images. Classication might provide the answer whether the image contain brain tumor or not. For classication purposes, many classiers will be used. Some commonly used classication methods are Articial Neural Networks (ANN), Bayesian classication, K-nearest neighbor classiers, support vector machine(SVM), and random forest(RE). Supervised machine learning approaches are used for the classication of MRI images. Undoubtedly, each approach has its advantages and it's inconvenient. However the selection of the classication algorithm depends on giving uus the more accuracy value which we will implement in our system.Chapter 2:Prineiple 16 In our project we have implemented Artificial Neural Network method for classifi- cation. ARTIFICIAL NEURAL NETWORK::- An artificial neural network{ANN) generally called neural network (NN).is a mathematical model or computa- tional model that is inspired by the structure and/or functional aspects of bio- logical neural networks. A neural network contains of an interconnected group: of artificial neurons(processing element), working in unison to solve specific problems. ANNs, like people, learn by example. The neuron has two modes of operations: The training/learning mode and the using/testing mode. In mainly cases an ANN is an adaptive system that converts its structure based on external or internal information that flows through the network in the learning phase. Recent neural networks are non-linear statistical data modeling tools. They are generally used to model complex relationships between inputs and outputs or to find patterns in data. Multi-layer perceptron learning algorithm is a supervised learning algorithm. It is one of the most important develop- ments in neural networks. ‘This learning algorithm is applied to multilayer feed-forward networks consisting of processing elements (neurons) with con- tinuous differentiable activation functions (Tan-sigmoid and log-sigmoid). For a given set of training input-output pair, this algorithm provides a procedure for changing the weights in a BPN to classify an input correctly. The concept for this weight update algorithm is basically the gradient descent method as used in case of simple perceptron networks with differentiable units. This is a way where the error is propagated back to hidden unit. The aim of the neural network is to train the net to achieve a balance between the net’s ability to re- spond (memorization) and its ability to give reasonable responses to the input that is similar but not identical to one of that is used in training. The algorithm uses a multi-layer perceptron neural network the schematic representation of neural network " inputs, “n” hidden units and one output unit. The extracted features are considered as input to the neural classifier. A neural network is a set of connected input/output units in which each connection has a weight associated with it, The neural network trained by adjusting the weights so as to be able to predict the correct class. The desired output was specified as 0 for non-cancerous anid 1 for cancerous. The classification process is divided into the training phase and the testing phase. During training, theChapter 2:Principle 7 features are extracted from the images in which the diagnosis is known. After training is over, the trained networks are stored to be used in the algorithm. ‘Whenever an image is taken as input in the algorithm, it is simulated with the trained net-works and goes for testing the data. Input Layer Hidden Layer ‘Output Layer FIGURE 2.6: Architecture of Neural Network MLP networks are feed forward networks with one or more layers of units between the input layer and the output layer. The network is termed feed forward since network signals flow only in forward direction. Starting at the input layer forward to hidden inputs and then still outputs are forwarded to output layers. The inputs to the ith unit are weighted sum of output from the previous layers. We have to assign some random biases, after that we can calculate the output which is the output we want to predict. We have to calcu- late error which is the actual output minus the predicted output. We will use errors this error to correct the weights. This is called backward propagation. We will calculate till we get the approximate value of the output. How Artificial Neural Network works?Chapter 2:Principle 18 Ficure 2.7: Artificial neuron and Biological neuron The basic idea is that we have three layers of “nodes”. The “nodes” are intended to be analogous to neurons in a neural network of the brain, but the similarity is only metaphorical. The nodes have values of 0.0 to 1.0, where 0 represents fully inactive “off and 1 represents fully active “on” with many values in between. The three layers are an input layer, and output layer, and a “hidden” layer in the middle. The nodes are linked by connections which have a “weight” that are analogous to synapses in the brain. Signal values propagate from the inputs, through the connection weights to the hidden nodes, and then onward through more connection weights to the output nodes [8]. DESICIONS:- All classification result could have an error rate and on occasion will either fail to identify an abnormality, or identify an abnormality which is not present. It is common to describe this errar rate by the terms true and false positive and true and false negative as follows: * True Positive (TP) — counts of all samples which are correctly called by the algorithm as being cancer. © False Positive (FP) ~ counts of all samples which are incorrectly called by the algorithm as being cancer while they are normal.Chapter 2:Principle 19 # True Negative (TN) ~ counts of all samples which are correctly called by the algorithm as being normal. ‘@ False Negative (FN) ~ count of all samples which are incorrectly called by the algorithm as being normal while they are cancer.Chapter 3 EXPERIMENTAL RESULTS 3.1 Input MRI Image Step 1: First we have browsed one training dataset image ie. Laat file from the database in the GUL. WRI Image processing for Brain Tumor Using GLCM for Feature Extraction — aren ten tneces ot mds Ps La i » ws 20Chapter 3:Experimental Results 21 3.2. Preproc sing using Median Filter,Gaussian Filter and Sobel Edge Detection Step 2: Here we have taken the browsed image and preprocessed the image which is the second step of image processing. 3.3 Segmentation and Feature Extraction Step 3: After preprocessing we have segmented the image using threshold segmen tation and after that we have extracted the features using GLCM Feature Extraction Algorithm .Chapter 3:Experimental Results NR & (WR nage processing for Brain Tamar Using LCM for Feature Extraction 3.4 Classification using Artificial Neural Network Step 4: For classification of tumor and non tumor image we have used ANN classi- fier. Here we have used “nntool” in MATLAB for classification. We have given the extracted features as input to the classifier where we are taking 10 hidden layers and two output class. For training we have taken nat dataset and we have compared with the reference data of the matlab. The standard network is used for pattern recognition is a two-layer feedforward network, with sigmoid tranfer functions in both the hidden layer and ontput layer. Here we got the following graphs of ANN Algorithm. This is the BE VALIDATION PERFORMANCE GRAPH where there is 22 epoch. For this image at epoch 16 we get the better validation performance, where training and validation set somewhat becomes equalChapter 3:Experimental Results 23 SF Neurai Network Training Performance (plotperform), Epoch 22, Validatio. = — o x File Edit View Insert Tools Desktop Window Help ~ Best Validation Performance is 0.031246 at epoch 16 10° } Cross-Entropy (crossentropy) o 2 4 6 B 1 12 14 16 18 2 2 22 Epochs Training Set: This dataset is used to adjust the weights on the neural. Validation Set: This dataset is used to minimize over fitting. We are not adjusting the weights of the network with this dataset, we are just verifying that any increase in accuracy over the training dataset actually yields an increase in accuracy over a dataset that has not been shown to the network before, or at least the network hasn’t trained on it. If the accuracy over the validation data set stays the same or decreases, then we are over fitting our neural network. Testing Set: This dataset is used only for testing the final solution in order to confirm the actual predictive power of the network. 3.5 Gradient and Validation Check Graph Step 5: We have found out the gradient which is used to calculate the backpropa- gation using logarithmic scales. Validation check is used to stop training with the trained dataset and non trained dataset when the accuracy of the training dataset increases. Here valdation fails atChapter 3:Experimental Results rey value 6 at epoch 22. FH Neurat Network Training Training State (plottrainstate). Epoch 22, Vala — o x File Edit View Insert Tools Desktop Window Help ~ Gradient = 0.028739, at epoch 22 10° 3.6 Error Histogram Graph Step 6: Here we have calculated the errors which is the difference between Targets and the Outputs. Bins are the number of vertical bars we are observing on the graph. The total error from neural network ranges from -0.2222(leftmost bin) to 0.198 (rightmost bin). This error range is divided into 20 bins. Each vertical bar represents the number of samples from our dataset, which lies in a particular bin. For Example at the mid of our graph, we have a bin of -0.04783 and the height of that bin for training dataset: is 450. It means 450 samples from our validation dataset have an error lies in that range. Similarly for other dataset and other bins ‘we can interprete our results. Zero error is the number of outliers datapoints that doesnot belong to neither of the set.Chapter 3:Experimental Results 5 F Neurai Netwone training Error Histogram (ploterrseD. Epoch 22. Valdas — = File Edit View Insert Tools Desktop Window Heip <4 Error Histogram with 20 Bins 700 3.7 Confusion Matrix Step 7: Confusion matrix basically gives us an idea about how well our classifier has performed, with respect to performance on individual on individual classes. So typically a confusion matrix is filled up based on test set whose true labels is known. The test data is passed through the classifier and predictions are noted. A table of predicted labels vs true label is then filled out. Green color boxes correctly predict the class and red color values are those that incorrectly predict the class diagonally. ‘The lower right blue squares illustrate the overall accuracies.Chapter 3:Experimental Results 26 16 pees eters arene Comba pera Ean Wane = , ee = i rerio : Test Contusion mate Farget Ctaee 3.8 ROC Curve Step 8: A Receiver Operating, Characteristic Curve, or ROC Curve, is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrim- ination threshold is verified. The ROC Curve is created by plotting the true positive rate(TPR) against the false positive rate (FPR) at various threshold settings. The true positive rate is also known as sensitivity, recall or probability of detection. The false positive rate is also known as the fall-out or probability of false alarm and can be calentated as (I- specificity), The colored lines in each axis represent the ROC Curves. A perfect test would show points in the upper-left corner, with 100% sensitivity and 100% specificity. For this problem, the network performs very well.Chapter 3:Experimental Results 27 F tesa NetaorTainng Raves Operating Shaves iach Fancn 3H Whanon fae fam View tment Foc Demme Winton Het oa? oa eo os + ‘os aa ‘Pawo Pounive Rate 7] oa ‘Fates Poative Hate ata Positive 3.9 Accuracy We have Obtained from ANN Al- gorithm: METHOD | ACCURACY| SENSITIVITY | SPECIFICITY| DATASET IMAGE ANN (ARTIFICIAL NEURAL 0.765 0.975 0.987 300 MRI NETWORK) IMAGES FIGURE 3.1: Accuracy Table Sensitivity: It is the measure of true positive rate(TPR) which is given by the formula- ‘TPR= TP/P = TP/TP+FN =1-FNR [Where TP is true positive, FN is false neg- ative and FNR is false negative rate value].Chapter 3:Experimental Results 28 Specificity: It is the measure of true negative rate(TNR) which is given by the formula- TNR = TN/N = TN/TN+FP = 1-FPR [Where TN is true negative, N is negative FP is false positive and FPR is false positive rate value].Chapter 4 COMPARATIVE STUDY 4.1 Comparative Study of Different Brain Tumor Detection and Classification Techniques Us- ing MRI images TABLE 1: COMPARISON OF BRAIN TUMOR DETECTION AND CLASSIFICATION Proposed Datavet ‘Technig ‘svateaeaes =| aT ae a RT ne 29Chapter 4 TABLE 2: COMPARISON TABLE OF ACCURACY Author Accuracy Benefits Limitations Parveen, Amaripalsingh | 91.66% ‘Ttcombines clustering | Brain tumortype can't (2018) and classification be classified Difficult to algorithmEfficient —_| choose SVM kernel method function ‘Astina Mina, Prof. $9.90%and 74.00% | Minimize the error, Less | Ttean maximize the (Chandrakant Mahebiya time consuming ‘margin withrespect to (2017) features thathave already been selected. Mulambila?,5.Uma [90 12%and6235% | Increased Accuracy and | Potentialof RaniK. (2017) ‘Robust modeling imisidentification of what is supposed to be categorized Rasel Ammed, Tae Tnereased Accursry | Difficultyin selecting Anirban Sen Swaksbar, Classifybrain tumor | optimal features to ‘Md. Foisal Hossain, ‘with braintumor distinguish different Md. AbdurRafig affected stages classes Time (2017) Consuming KSudaasDeT. [95.00% Gimpleandflexibleto | Largescarch problemto Sarma. Dr. K. implement, Handle | find nearest neighbor Rasad(2015) multiclass cases ‘Storage of dataConclusion es are hest suitable for brain tumor detection. In this study Digital Im- age Processing ‘Techniques are important for brain tumor detection by MRI images. ‘The preprocessing techniques include different methods like Filtering, Contrast en: hancement, Edge detection is used for image smoothing. The preprocessed images are used for post, processing operation like ¢hreshold and then features are extracted using GLCM and classification is done by Artificial Neural Network where we have obtained 76.5% of accuracy when run on a dataset of 300 images in classifying tumor and non tumor images. This work help in detection of tumor which in turn save th precious time of doctor and pathologist to diagnose the tumor automatically in short span of time. 31Future Work In our project we have successfully implemented all the steps of image processing that is pre-processing, segmentation, feature extraction and classification using various algorithm as we have mentioned above. As we have received 76.5% of accuracy in our project, for future we will try to implement Support Vector Machine (SVM) Algorithm using more MRI images so that we can get better accuracy result,Bibliography IJASindhul , S.Meera2 “A Survey on Detecting Brain Tumor in mri Images Us- ing Image Processing Techniques "International Journal of Innovative Research in Coriputer and Communication EnginceringVol. 3, Issue 1, January 2015. 2]Parveen,Amritpal Singh “Detection of Brain Tumor in MRI Images, using Com- ination of Fuzzy ¢-means and SVM" in 2nd International Conference on Signal Processing and Integrated Networks (SPIN) 2015. 3] Astina Minz, Prof. Chanddrakant Mahobiya “MR Image classification using Adaboost for brain tumor type” in IEEE 7th International Advance Computing ‘onference(IACC) 2017. Q 4] Mukambika P. S., Uma Rani K. “Segmentation and Classification of MRI Brain Tumor” in International Research Journal of Engineering and Technology (IRJET) Volume: 04 Issue: 07 July -2017 . [5] Rasel Ahmmed, Anirban Sen Swakshar, Md. Foisal Hossain, and Md.Abdur Rafiq*Classification of Tumors and It Stages in Brain MRI Using Support Vec- tor Machine and Artificial Neural Network” in International Conference on Electri- cal, Computer and Communication Engineering (ECCE), Cox's Bazar, Bangladesh, February 16-18, 2017.Bibliography 34 [6] K. SUDHARANI, Dr. T. C. SARMA, Dr. K. SATYA RASAD “Intelligent Brain Tumor Lesion Classification and Identification from MRI Images Using k-NN Tech- nique” in International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT) 2015. [7] Ruchi D., Deshmukh et al “Study of Different Brain Tumor MRI Image Segmen- tation Techniques” in International Journal of Computer Science Engineering and ‘Technology( IJCSET) Vol 4, Issue 4133-136 April 2014. [S]Kumar, GV., Raju, GV, (2001). Biological early brain cancerdetection using ar- tificial neural network, International Journal on Computer Science and Engineering , 22721-2725. DATASET LINK:- L Brain MRI Dataset (BRATS 2015)- https://in.mathwork.com/matlabcentral /answer /343191-brain mri-dataset-brats-2015. IL, Normal brain MRI Dataset link: http://brainweb.bic.mni.mcgill.ca/cgi/brainweb]