AN AUTOMATIC MODEL FOR BRAIN TUMOR DETECTION USING MACHINE LEARNING TECHNIQUES
Machine learning is a technique which is used to detect the pattern based on the given training data. In this paper, we have proposed a method for detecting tumor area from the given brain MRI images. The Brain is a major organ in our body, it controls the entire body system of a human. There is an uncontrollable and abnormal growth of cells in tissues creates cancer in the body. Our brain is enclosed by a rigid skull, there is any unwanted growth in that portions creates a problem. Two types of tumor are Benign and malignant, the Benign tumor is noncancerous, the malignant tumor is cancerous also life-threatening disease. Glioma is one of the brain tumors which is originated from the glial cells of the brain or spine. Based on the survey 80 percent of the tumor is malignant and very short life expectancy. Magnetic Resonance Imaging is a technique to produce the MRI images of the brain, this has to produce more information about the brain. Based on the MRI image information the treatment for tumor patients has been planned. Brain tumor image segmentation and detection of the tumor area are not the simple processes by manual. In this paper, we have proposed an automatic and efficient method for detecting tumor from the MRI images using a machine learning technique. An automatic method to isolate the brain tumor area from given MRI based on Support Vector Machine(SVM) and it is achieved 96.2% accuracy,94.1%specificity, and 97%sensitivity, the results shows that the better efficiency.
In recent time, Machine learning model has been developed for various applications like images, agriculture,chemical descriptions, textiles,social networks and financial applications etc. Medical image analysis is important for the current scenario. Our goal is to develop an efficient automatic model for brain tumor detection from the given MRI images. The technology, communications,and e-health care systems were assisted the patient for better treatment. This study represents the problems of segmentation of
normal and abnormal tissues from the MRI images using feature extraction methods and classifier of SVM1,2. Brain tumor is the reason for high mortality of the people.There is a treatment for brain tumor are chemotherapy,surgeroy and radiological treatment or combined of these treatment methods.MRI is essential tool in the clinical and surgical environment due to superior soft tissue differentiation, high spatial resolution, contrast and it does not use any harmful ionizing radiation which may affect patients. In MRI technique, brain is imaged on the basis of density of water in soft tissue which is higher compared to other tissues such as bone 3. Through the MRI images, the radiologist can see the brain anatomy without performing surgery. However, this process is still done manually and could lead to misdiagnose. In addition, the different complex characteristics of brain tumor make diagnosis more difficult 4. Therefore, system of Computer Aided Diagnostic (CAD) is indispensable that will help radiologist in identifying and classifying brain tumors. ShanShen et al proposed fuzzy c means clustering (IFCM). The proposed algorithm is based on neighborhood attraction. It is considered that it exist between neighboring pixels. This neighborhood attraction depends on the pixel intensities, the spatial position of the neighbor pixels and on neighborhood pixel structure. The classification of tumor is done with the help of artificial neural network (ANN) based on the similarity between feature vectors 5.In recent years, researchers have proposed a lot of approaches for this goal, which fall into two categories. One category is supervised classification, including support vector machine (SVM) 6 and k nearest neighbors (k-NN) 7. The other category is unsupervised classification 8, including self-organization feature map (SOFM) 6 and fuzzy c-means 9. While all these methods achieved good results, and yet the supervised classifier performs better than unsupervised classifier in terms of classification accuracy (success classification rate). However, the classification accuracies of most existing methods were lower than 95%, so the goal of this paper is to find a more accurate method. Among supervised classification methods, the SVMs are state-of the-art classification methods based on machine learning theory 10-12. Compared with other methods such as artificial neural network, decision tree, and Bayesian network, SVMs have significant advantages of high accuracy, elegant mathematical tractability, and direct geometric interpretation. Besides, it does not need a large number of training samples to avoid over fitting 13.A new incorporation of SVM and ANN for tumor classification is introduced in this research. Brain MRI images with normal and abnormal behavior is firstly enhanced through some filter and preprocessing steps. Therefore, for the detection and classification of the tumor in the brain there be incorporated segmentation processes namely temper based K means and modified Fuzzy C-means (TKFCM) clustering algorithm 14 is used.
In this paper, an automatic method is designed using machine learning technique and it is used to detect and isolate tumor from normal brain MRI images with SVM as the classifier. Segmented images are enhanced using enhancement techniques such as contrast improvement, and mid-range stretch. The experimental results achieved 96.2% accuracy, 94.1% specificity, and 97% sensitivity, demonstrating the effectiveness of the proposed technique for identifying normal and abnormal tissues from brain MR images
Feature Extraction and Selection using IBF,GLCM and PCA
FBB SegementationSVM Classifier
Classify Tumor and Non-Tumor images
2.1 Proposed model block diagram
Image data collection
The first step for detecting tumor from the brain MRI is the collecting images images.The images were taken from the the internet public resources.The repository of MRI images contains normal and tumor affected images like low grade glioma and high grade gliomas13-15.There are 50 images were taken for the classification of tumor detection.
Image pre- processing
The next step for classifying tumor is preprocessing the images that is retrieved from the public repository of internet sources.Preprocessing step,which is helpful for enhancing the parameters improve visual appearance of MR images.The method signal to noise ratio improves the clarity of the raw image.Removing the irrelevant noise and unwanted portions in the backround of the image. Smoothening is the technique which is used to smoothing the inner part of the region and it is preserving edges of the images. Proposed method for this paper is anisotropic diffusion24 is a method for reducing noise without removing necessary contents like edges,lines,etc to be used for analyze the image. Anisotropic diffusion look likes the process that creates a scale space, where an image generates a parameterized family of successively more and more blurred images based on a diffusion process. Each of the resulting images in this family are given as a convolution between the image and a 2D isotropic Gaussian filter, where the width of the filter increases with the parameter. This diffusion process is a linear and space-invariant transformation of the original image.
Skull stripping 25is a technique which is for removing non-brain tissues like scalp,fat,eyes,neck,etc from the given input MR images.This method is helpful for increase the speed and accuracy of detecting and predicting medical images.The proposed technique is double thresholding segmentation technique to convert our input image into binary form of the image which does not contains the unwanted tissues details are not available.In this double thresholding method,upper and lower thresholds were considered The first step is erosion for removing skull portion for that we are taking 3 as the radius for removing unwanted parts.Next,region filling algorithm for filling the holes in the MR images.Finally,the related backround pixels are converted into foreground pixels so the holes were removed in brain MR images.
The erosion of a binary image f by a structuring element s (denoted f s) produces a new binary image g = f s with ones in all locations (x,y) of a structuring element’s origin at which that structuring element s fits the input image f, i.e. g(x,y) = 1 is s fits f and 0 otherwise, repeating for all pixel coordinates (x,y).
The dilation of an image f by a structuring element s (denoted f s) produces a new binary image
g = f s with ones in all locations (x,y) of a structuring element’s orogin at which that structuring element s hits the the input image f, i.e. g(x,y) = 1 if s hits f and 0 otherwise, repeating for all pixel coordinates (x,y). Dilation has the opposite effect to erosion — it adds a layer of pixels to both the inner and outer boundaries of regions.
There are two types of features are extracted in this model, the primary first order statistic feature is used to find tumor cells area and that exact position of the tumor in the brain MRI image. The next second order region based feature is to classify the type of the tumor affected. In this model we are using Gray level co-occurrence Matrix(GLCM) statistical based feature extraction is two dimensional matrix which contains row and columns numbers are equal to the number of gray levels .The image features extracted in the GLCM are homogeneity, contrast, correlation, energy, etc. The Gray Level Coocurrence Matrix (GLCM) procedure is a way of pull out second order statistical texture features 18.GLCM matrix represents the relationship between the pixels inside the region.It models the relationships among pixels inside the region by building Gray Level Co-occurrence Matrix. The GLCM is based on an inference of the second-order mutual limited probability density functions p(i, j | d, ?) for a diversity of direction ? = 0, 45, 90,135°, etc., and unlike distances, d = 1, 2, 3, 4, and 5. The function p(i, j | d, ?) is the probability that 2 pixels, which are placed with an intersample distance d and a direction ?, have a gray level i and j. The spatial relationship is defined in terms of distance d and angle ?. If the texture is coarse, and distance d is small, the pair of pixels at distance d should have same gray values. On the other hand, for a fine texture, the couple of pixels at distance d should often be quite unlike, so that the value in the GLCM should be stretching out moderately uniformly 22.
Intensity based features (IBF) extracted 5 features namely mean, standard deviation, coarseness, skewness,and kurtosis. Total 10 features are extracted by the original and segmented image.Area of the extracted tumor and dice coef?cient similarity index is also used by the classi?er to the classify the tumor types. Through the analysis, it is found that the performance parameters of the classi?er such as accuracy, speci?city, and sensitivity are improved by considering the area of the tumor and dice coef?cient as a part of the features.
The features are calculated based on these formulas
Energy=i,j=0N-1?( Pij )2Entropy=i,j=0N-1-ln?( Pij ) Pij Contrast=i,j=0N-1 Pij (i-j)2Correlation=i,j=0N-1 Pij i-?(j-?)?2Homogeneity=i,j=0N-1Pij1+(i-j)2Mean(?i)=i,j=0N-1 iPij Mean(?j)=i,j=0N-1 jPij
Variance(?i2)=i,j=0N-1 Pij (i-?)2Variance(?j2)=i,j=0N-1 Pij (j-?)2 Standard Deviation(?i2)=2?i Standard Deviation (?j2)=2?j Dissimilarity=i,j=0N-1 Pij i-jThe features extracted from the method of GLCM, SFTA,and IBF are now to optimized using feature selection, so that only relevant features are employed for the classi?er to classify tumor type. The principal component analysis (PCA) is used to select only relevant features for the optimization of the complexity of the calculation by the classi?er. If feature selection is not done, then it will increase mathematical computations on the classi?er, so to reduce the complexity of the classi?er, relevant features must be extracted before it is processed by the classi?er. The objective of the PCA is described as follows:
1) To identify only relevant features from large set of feature vectors and eliminate remaining
2) To discover new features depending on the relevance
3) To emphasize variations in similar features and bring out only strong patters based on largest
variance4) Optimize selected features, i.e. decide the relevance according to the importance of the features.
Image segmentation is important step for brain tumor detection model,it divides the set of pixels into group based on homogeneous properties such as intensity, depth, color or texture.The output of image segmentation step is to label for identify the homogeneous region or a set of contours to describe the boundaries of region.In the segmentation step output is to divide the image elements based on the tissue types such as white matter(WM),gray matter(GM), and cerebrospinal fluid(CSF).The results of segmentation step is also used for analyzing anatomical structures, detect the pathological regions, treatment of surgical planning and for visualizing. There are numerous segmentation methods are available, they are thresholding based, region growing, water sheds and contours based methods. In these existing methods problems were rectified by the proposed technique. At the preprocessing level, to extract the information regarding tumor from the input MRI image then extra portions of skull information were removed, after that using anisotropic diffusion filter is applied to remove noise.To apply Fast Bounding Box algorithm(FBB)20 into that image and it retrieves the tumor area bounded by the bounding box and the central part in that box sample points for training of SVM classifer.
Fast bounding box20 works in two progressive stages. Initially, the idea set of 2D MR slices is treated separately, for catching axis-parallel rectangles. Afterward, all the bounding boxes are collected to recognize the truly mounted tumor or edema. These two stages stay defined in the following subdivisions. Now in this sector ornate the basic principle ahead FBB: an alteration discovery code, wherever a region of change (D) is noticed proceeding a test image (Im), once linked by a reference image (Rm). In FBB, later ruling the alignment of symmetry taking place an axial MR segment, the left (or the right) partial aids by way of the test image Im, and the right (or the left) half deliveries as the reference image R. The region of change D here stays limited near an axis parallel rectangle that basically purposes to define irregularities. This technique remains unlike by means of utmost of the alteration finding approaches planned to time in that we understands this alteration as a region-based total conversion that be altered as most techniques, which view the change as a local pixel-to-pixel deviations. At this point tumor is reflected as the change area in the test image and all addition al intracranial tissues excluding tumor are reflected as the no change area. We use a new score function that can recognize the region of change D with two precise rapid examinations single beside upright track of the image and the additional along the horizontal track.
The rectangle region is defined as
D=LX ,Ly × Ly, Uy
D denotes tumor containing region.T(I) is the top sub rectangle and B(I)is the bottom sub rectangle of the image.I is the midpoint of the image.Score function is used to find the Ly and Uy values in a vertical sweep.
E(I)=BC(PI T(I), PR T(I)) –BC(PI B(I), PR B(I))
Where PI T(I) denotes normalized intensity histogram of the region T(I) in the test image. PR T(I) , PI B(I), PR B(I) are defined accordingly.E(I) is the score function.
Bhattarcharya coefficient is defined as
Where a(i) and b(i) are the two normalized histograms.Detected tumor part is segmented using active contour model21.
To detect the boundary of the image pixels within the image, it provides the detail characteristics of image for segmentation. This is may consider as a vector contains magnitude of the gradient and direction of an edge. Feature extraction has to reduce the amount of data in an image and the essential structural properties were used for remaining image processing. In this proposed approach filter the region of interest(ROI) also identify the infected surrounded portions in the brain image. This method improves and enhance the processing time because the feature processing algorithm going to apply for identified areas not into the whole image frame. In this edge detection step first apply the vector subtraction algorithm next ROI is identified by find the similar adjacent parts in the resultant portions in the resultant image from the vector subtraction21.
7.ClassificationThe major function of classification step is to group the similar features of input brain MRI image pixels based on the retrieved statistical features in the feature extraction step of the image. In machine learning there are two types classification, one is supervised classification another is unsupervised classification. Supervised classification techniques to train by the features of training data, based on these trained features the test data has been predicted and evaluate the model. But in unsupervised learning technique the model has not been required training set based on the features the pattern has been predicted.
There are several categories of Supervised classification, they are
• Artificial/logical techniques: Decision trees.
• Perceptron based techniques: Single layered perceptron, Multilayered Perceptron, Radial Basis
• Statistical techniques: Naive Bayes, Bayesian networks,
• k-Nearest neighbor classifiers.
• Support Vector Machines.
In this paper, We have proposed the supervised classification technique is Support Vector Machine to classify the MRI image pixels are normal or abnormal.
8. SUPPORT VECTOR MACHINE:
SVM is efficient classification algorithm, it is used for various tasks such as simple and linear classification tasks, in addition to it more complex of nonlinear classification problems are solved. In SVM hyperplane to separate the different class boundaries of data points, this is the key role of SVM classification. In linear and nonlinear cases handled all problems of separable and no separable by SVM. The proposed classification problem is to separate the tumor and normal image from the given MRI image so SVM hyper plane to separate the normal and abnormal brain images based on the given data and features.SVM starts a classification initially by discriminative features then add less discriminative features. GLCM and IBF methods were extracted the optimal features, feature selection done by PCA method and it is given the input of SVM22 classifier. Compare with conventional classifier SVM manages the large feature spaces and better generalization properties. Larger feature sets were given as the input of SVM classifier so structural misclassification risk has been reduced at the time of training phase. Here we are used binary classification because the image were classified based on two classes they are normal and abnormal pixels. We are proposed SVM classifier kernel function is radial basis function(RBF)kernel. From the set of N samples some of them as training and remaining are testing. To improve the accuracy of the model, same methods were repeated until get the better accuracy.
III.RESULTS AND DISCUSSION
The database consists of brain MRI images, an automatic model for detecting tumor to test the various patient image features has been extracted, analyzed and classified whether it has affected or not. The system has been implemented in MATLAB R2017a and SVM as the classifier.
The model result has been evaluated based on the following parameters sensitivity, specificity and accuracy.
Patient: Tumor positive
Healthy: Tumor negative
True positive (TP) = No of cases identified Tumor correctly
False positive (FP) = No of cases wrongly identified of tumor patients
True negative (TN) =No of cases identify the healthy images correctly
False negative (FN) = No number of cases incorrectly identified as normal
The sensitivity of a model is its capability to find out the patient cases correctly. To calculate it, we should calculate the fraction of TP in patient cases. Mathematically, this can be represented as:
Sensitivity=TPTP+FNThe specificity of a model is its capability to find the normal cases correctly. To calculate it, we should calculate the fraction of true negative in normal cases. Mathematically, this can be represented as:
Specificity= TPTN+FPThe accuracy of a model is its ability to categorize the tumor and normal cases correctly. To calculate the accuracy of a test, we should calculate the proportion of true positive and true negative in all evaluated cases. Mathematically, this can be represented as:
Accuracy=TP+TNTP+TN+FP+FN After running the model, Our system achieved an accuracy of 96.2%,Sensitivity of 97%,Specificity of 94.1%.
We had developed a model for automatic detection tumor from the human MRI images using machine learning technique. This model classify and discriminate between normal and abnormal MRIs of the brain using SVM classifier. From the pre-processed images,features were extracted using GLCM,IBF and features are selected by the method of PCA.FBB segmentation method and SVM classifier used to classify and segment the tumor from the input MRI images.The method obtained 96.51% classification accuracy on the brain MRI image datasets. Further research could be progressed by enlarging the experimental patients to reach a more-accurate, stronger statistical significance result.
L. Guo, L. Zhao, Y. Wu, Y. Li, G. Xu, and Q. Yan, “Tumor detection in MR images using one-class immune feature weighted SVMs,” IEEE Transactions on Magnetics, vol. 47, no. 10, pp. 3849–3852, 2011. View at Publisher · View at Google Scholar · View at Scopus
R. Kumari, “SVM classification an approach on detecting abnormality in brain MRI images,” International Journal of Engineering Research and Applications, vol. 3, pp. 1686–1690, 2013. View at Google Scholar
R. Rana, H. S. Bhadauria, and A. Singh, “Study of various methods for brain tumor segmentation from MRI images”, International Journal of Emerging Technology and Advanced Engineering, vol. 3, no. 2, pp.338-342, 2013.
J. Joshi, et. al., “Feature Extraction and Texture Classification in MRI” In Special Issue of IJCCT, pp. 130-136, 2010.
Shan Shen, William Sandham , 2005 , ” MRI Fuzzy Segmentation of Brain Tissue Using Neighborhood
Attraction With Neural-Network Optimization “, IEEE transactions on information technology in
biomedicine Volume 9,issue no.3,pp. 459 -467.
Chaplot, S., L. M. Patnaik, and N. R. Jagannathan, “Classification of magnetic resonance brain images using wavelets as input to support vector machine and neural network,” Biomedical Signal Processing and Control, Vol. 1, No. 1, 86–92, 2006.
Cocosco, C. A., A. P. Zijdenbos, and A. C. Evans, “A fully automatic and robust brain MRI tissue classification method,” Medical Image Analysis, Vol. 7, No. 4, 513–527, 2003.
Zhang, Y. and L. Wu, “Weights optimization of neural network via improved BCO approach,” Progress In Electromagnetics Research, Vol. 83, 185–198, 2008.
Yeh, J.-Y. and J. C. Fu, “A hierarchical genetic algorithm for segmentation of multi-spectral human-brain MRI,” Expert Systems with Applications, Vol. 34, No. 2, 1285–1295, 2008.
Patil, N. S., et al., “Regression models using pattern search assisted least square support vector machines,” Chemical Engineering Research and Design, Vol. 83, No. 8, 1030–1037, 2005.
Wang, F.-F. and Y.-R. Zhang, “The support vector machine for dielectric target detection through a wall,” Progress In Electromagnetics Research Letters, Vol. 23, 119–128, 2011.
Xu, Y., Y. Guo, L. Xia, and Y. Wu, “An support vector regression based nonlinear modeling method for Sic mesfet,” Progress In Electromagnetics Research Letters, Vol. 2, 103–114, 2008.
Li, D., W. Yang, and S. Wang, “Classification of foreign fibers in cotton lint using machine vision and multi-class support vector machine,” Computers and Electronics in Agriculture, Vol. 4, No. 2, 274–279, 2010.
R. Ahmmed and M. F. Hossain, “Tumor stages detection in brain MRI image using Temperbased K-means and Fuzzy C-means Clustering Algorithm”, Proceeding of 11th Global Engineering, Science and Technology Conference, pp. 1-10, 18 – 19 December, 2015, BIAM Foundation, Dhaka.
O.P. Verma, M. Hammandlu, S. Susan, M. Kulkami and P.K. Jain, “A simple single seeded region growing algorithm for color image segmentation using adaptive thresholding,” 2011 International Conference on Communication Systems and Network Technologies,©2011 IEEE.
M. Lugina, N. D. Retno, and R. Rita, “Brain Tumor Detection and Classification in Magnetic Resonance Imaging (MRI) using Region Growing, Fuzzy Symmetric Measure, and Artificial Neural Network Back propagation”, International Journal on ICT, vol. 1, pp. 20-28,December, 2015.
Baidya Nath Saha ,Nilanjan Ray,Russel Greiner ,Albert Murtha,Hong Zhang “Quick detection of brain tumors and edemas: A bounding box method using symmetry”, comput Med Imag.and Graph , Vol. 36, No.2, pp. 95 – 107, 2012.
T. F Chan and L. Vese, “An active contour model without edges”, C pp. 141 – 151,1999.
Haralick R.M., Shanmugam K. and Dinstein I., “Textural Features for Image Classification,” IEEE Trans. on Systems, Man and Cybernetics
3(6),1973, pp 610- 621.
Saha, B.; Ray, N.; Greiner, R.; Murtha, A.; Zhang, H. Quick detection of brain tumors and
edemas: A bounding box method using symmetry. Comput. Med. Imaging Graph. 2012, 36, 95–107.
21.Zhang KH, Zhang L, Song HH, Zhou WG (2010) Active contours with selective local or global segmentation: a new formulation and level set method. Image and Vision Computing 28: 668–676.
S. C. Chen and D. Q. Zhang, ?Robust image segmentation using FCM with spatial constraints based on new Kernel -induced distance measure,? IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 34, issue 4, pp. 1907-1916, 2004.
Y. Zhang and L Wu, AN MR BRAIN IMAGES CLASSIFIERVIA PRINCIPAL COMPONENT ANALYSIS AND KERNELSUPPORT VECTOR MACHINE, Progress InElectromagnetics Research, VoL 130, 369,388, 2012.
Haralick, R.M., Shanmugam, K., Dinstein, Its’Hak: Textural Features for Image Classication. IEEE Transactions on Systems, Man and Cybernetics SMC-3 (1973) 610-621
Jaffar, Arfan ; Zia, Sultan ; Latif, Ghazanfar ; Mirza, Anwar ; Mehmood, Irfan ; Ejaz, Naveed ; Baik, Sung. (2012). Anisotropic Diffusion based Brain MRI Segmentation and 3D Reconstruction. International Journal of Computational Intelligence Systems. 5.10.1080/18756891.2012.696913.
Roy, S., ; Maji, P. (2015). A simple skull stripping algorithm for brain MRI. 2015 Eighth International Conference on Advances in Pattern Recognition (ICAPR).