A. Vithyavallipriya 1 , B. Sankaragomathi 2 , T. Ramakrishnan 3 I , 2 , 3 Dept. of Electonics and Instumentation Engineering 1 , 2 , 3 National Engineering College I , 2 , 3 Kovilpatti, India I vithyapr56@gmail.com Abstact- This paper presents a modern semi supervised scheme for the detection and segmentation of abnormalities present in the brain MR images. The high degree of automation can be attained by using semi supervised learning, because it does not require any pathology modeling. If the dimensionality of the data is large then the estimation of the probability density function is not possible. To overcome this every image is handled as a network of locally coherent image partitions. Median flter is used for preserving edges while removing noise. Contrast enhancement automatically adjusts the intensity values of the image to achieve a better quality. The block wise separation is carried out by calculating the parameter like principal component analysis (PCA), Eigen value, Eigen vector, maximum likelihood function. The maximum likelihood function which estimating the abnormality for each partition is formulated. The likelihood function consists of a model and a data term and is formulated as a quadratic programming problem. Keywords-distibuted estimaton; mamum lkelhood function; semi supervised learing I. INTRODUCTION Brain tumor is an abnormal growth of tissue in brain. Brain tumor can be spread by local extension and rarely metastasize (spread) outside of the brain. A benevolent brain tumor consists of non-cancerous cells and does not metastasize beyond the part of the brain where it originates. One of the existing methods for the detection and segmentation of brain tumor is supervised learing. The supervised learing uses the pathology model. The pathology model consists of the parameters for the both healthy and unhealthy brain by comparing the given image with the pathology model the tumor can be detected. The supervised method includes several patter recognition techniques. Many patter recogmtlOn methods are assuming particular distibutions of the features, and are called parametic methods. For example, the maximum likelihood (ML) method commonly assumes multivariate Gaussian distibutions [6]. The means and covariant matices for each of the tissues are estimated fom a user supplied pathology model, usually initiated by drawing regions of interest (ROT) on the images. The remaining pixels are then classifed by calculating the likelihood of each block, and picking the block intensity with the highest probability. Parametic methods are only usefl when the feature distibutions for the different classes are well known, which the case for MR images is not necessarily. The semi supervised learing method is applied for automatically segmenting pathologies in brain images, such as lesions, brain infarction, or dysplasia. The paper is organized as follows. Section IT is devoted to methodology. The results are presented in Section ITI followed by some concluding remarks in Section IV. IT. METHODOLOGY In this proposed method a new semi supervised scheme is intoduced for detection and segmentation of tumor. The input MRI image is obtained and it is converted into gray scale image. The image will consist of noise, so the initial step is to remove the noise and enhance the image by preprocessing. Here, the image is de-noised by using the median flter. Median fltering is very widely used here because, under certain conditions, it preserves edges while removing noise. Then the quality of the image can be adjusted by using the contrast enhancement technique. The contast ehancement is used to adjust image intensity values of the given image. Contast ehancement increases the total contast of an image by making light colors lighter and dark colors darker at the same time. Fig. 1 shows the Flow chart of proposed methodology. Afer preprocessing the image, the block wise separation is carried out. The image is separated into blocks. Because of the different size of the given image, the image is resized to a fxed size before preprocessing. The statistical parameters are calculated afer this step, because the statistical parameters are thee linear parameters which are used to show the signifcance of the abnormality in the given MRI image. The statistical parameters are Principal component analysis, Eigen vector, Eigen value and Maximum likelihood for each block. The parameters are classifed into two major categories they are public variables and private variables. The public variables are the variables which has the common value in the whole brain that is they are the parameters of the healthy brain. Another one is the private variables which have the parameters which are different fom the public variables and they are the parameters of the unhealthy brain. 978-1-4673-5301-4/13/$31.00 2013 IEE Based on the calculated parameters the tumor is detected. Then the tumor is segmented by using the likelihood fnction. By the segmentation, the abnormality in the brain is located. M TLAB is the tool used for the detection segmentation and location of abnormality in the given MR image. Input MB Brain Image Retrieving the data Pre-proce s s ing of the data Block wis e Sepaation Caculating paaeters for each block paraeters like PCJ Eigen value, Eigen Vector, Entopy Maimum likelihood function Tumor Detecton Using Distibution function Tumor Segentation Fig. I. Flow chart of proposed methodology. A. Retrieving the data Initially the image is taken and resized into 255 X 255 because the obtained image is in different dimension. The resized image is converted into gray scale image for frther processing. RGB image is converted into gray scale by eliminating the hue and saturation information while retaining the luminance. B. Preprocessing Median fltering and contast ehancement is used in the pre-processing to improve image data quality. The median flter is a digital nonlinear fltering technique, used to remove noise in the image. I image processing, it is ofen desirable to be able to perform some kind of noise reduction on an image. Here the median flter is used for fltering the given image. The median flter is a nonlinear digital fltering technique, ofen used to remove noise. Such noise reduction is a typical pre-processing step to improve the results of later processing. Median fltering is very broadly used in digital image processing because, under certain conditions, it conserves edges while removing noise. Contast enhancement is used for enhancing the image. Automatically brightens images that appear dark or unclear, and applies suitable tone alteration to deliver improved quality and clarity. Here by adjusting the intensity values of the image the better quality is achieved. C. Block wise separation The given image is separated into blocks for calculating the parameter in the block. lf the given image (A) consist of 'i' rows and 'j' columns then the block (B (i, j)) will consist of rows and columns such as i = 2: (size of rows in the given image)/x j = 2: (size of columns in the given image)/x Where x = a number for selecting the size of the block The statistical parameter for each block can be calculated by increment the separated block size. For example, B (i, j) = A (((i-I )*x) + I: i*x, (((-I )*x) + I: j*x) Then the parameters are calculated for each block. D. Parameter estimation The statistical parameter of each block is calculated. Based on the statistical parameter the abnormality which is present in the image can be determined. The parameters are entopy and PCA. 1) Entrop y : Entopy is a statistical measure of randomness that can be used to characterize the texture of the input image. It is used to fnd the variation in the image intensities. 2) Principal Component Anal y sis (PCA): A PCA is a way of identifing patter in data, and convey the data in such a way as to highlight their similarities and differences. Principal component analysis is a quantitatively rigorous method for achieving this simplifcation. The method creates a novel set of variables, kown as principal components. Principal components are known as the linear combination of the original variables. Each and every principal component is orthogonal to each other, so there is no unnecessary information. E. Semisupervised learning Semi supervised anomaly detection offers a solution to this problem by modelling normal data and then using a distance measure and tresholding to determine abnormality. Without using the pathology model the abnormality present in the given image can be calculated. So the high degree of automation can be obtained. The semi supervised learing can be obtained by calculating the statistical parameters such as entopy, PCA, Eigen value. PCA is the simplest of the tue eigenvector based multivariate investigation. Ofen, its operation can be thought of as revealing the interal stucture of the data in a way that best explains the variance in the data. If a multivariate dataset is viewed as a set of coordinates in a towering-dimensional data space, PCA can deliver the user with a lower-dimensional picture, a "shadow" of this object when viewed fom its most usefl viewpoint. This is completed by using only the frst few principal components so that the dimensionality of the transformed data is reduced. Entopy is a statistical measure of randomness that can be used to characterize the texture of the input image. F Distributed estimation The distibuted estimation [6] is nothing but the calculation of statistical parameter for each block in an image. The distibution algorithm will calculate the private and public parameters for each block. The public variables are the common variables in the image they have the parameter of the healthy brain which are common troughout the brain. The private variables are the parameters which are different fom the public variable and they are the parameters of the unhealthy brain. The image is divided into fve clusters for calculating the likelihood fnction. The clusters are based on the intensities. They are white matter, gray matter, skull, CSF and tumor. The likelihood fnction is used for fnding the difference between the pixel intensities. Log likelihood fnction is estimated for the process. The logaritm is a monotonically increasing task, the logarithm of a task achieves its maximum value at the same points as the fnction itself and thus the log-likelihood be able to used in place of the likelihood in maximum likelihood estimation. By calculating the distibution fnction for the parameters the tumor can be detected. G. Tumor segmentation and location The color map is used to segment the tumor in the given brain MR image. Finally the tumor in the MR image can be located by using the appropriate mask. m. RESULTS AND DISCUSSION Fig. 2 shows the location of tumor in the MR image. The Fig. 2(a) shows the input image for the process which is resized and converted into gray scale. The preprocessing of the image is shown in the Fig. 2(b). The median flter and contrast ehancement is used for pre-processing. The Fig. 2( c) shows the segmentation of tumor in the image. The Fig. 2( d) shows the tumor location. (a)The resized image (b) The pre processed image ( c) The segmented tumor (d) The location of tumor Fig. 2. The Location of Tumor in the MR image IV. CONCLUSION Image segmentation is the most challenging and active research area in the feld of image processing for the last decade. Tn this project, semi-supervised learing and distibuted estimation are used for the detection, segmentation and location of abnormality (tumor) in the given brain image which in tm leads to high degree of automation. Prominent features can be estimated and the type of tumor can be classifed for making improvement in brain segmentation methods. Because of the ongoing research in biological world, increasing new knowledge about the relationship between different disorders with anatomical deviation is coming up. So, brain segmentation is gaining importance in using as the frst stage in tools for detection and analyzing anatomical deviation. ACKNOWLEDGMENT The authors thak the management of National Engineering College for their support and motivation. REFERENCES [ I]. 1. Sled. A. Zijdenbos, and A. Evans, "A nonparametric method for automatic correction of intensity nonuniformity in MRI data," IEEE Trans. Med Imag., vol. 17, no. 1, pp. 87-97, Feb. 1998. [2]. K.Van Leemput, F. Maes, D.Vandermeulen, A. Colchester, and P. Suetens, "Automated segmentation of multiple sclerosis lesions by model outlier detection," IEEE Trans. Med Imag., vol. 20, no. 8, pp. 677-688, Aug. 2001. [3]. S. Shen, A. 1. Szameitat, and A. Sterr, "Detection of infarct lesions fom single MRI modality using inconsistency between voxel intensity and spatial location-A 3-D automatic approach," IEEE Trans. Ir. Technol. Biomed, vol. 12, no. 4, pp. 532-540, Jul. 2008. [4]. N. Batmanghelich, X. Wu, E. 1. Zacharaki, C. E. Markowitz, C. Davatzikos, and R. Verma, "Multiparametric tissue abnormality characterization using manifld regularization," in Proc. SPJE Med Imag.: Comput.-Aided Diagnosis, vol. 6915, no. I, article 16, pp. 1-6, 2008. [5]. 1. Theiler and L. Prasad, "Overlapping image segmentation fr context dependent anomaly detection, " Proc. SPJE, vol. 8048, 2011. [6]. S. Samar, S. Boyd, and D. Gorinevsky, "Distributed estimation via dual decomposition," in Proc. Eur. Control Cor., pp. 1511-1516, 2007. A.VithyavaJlipriya was born in Tirunelveli, India in 1990. He received his Bachelors of Engineering degree fom the depmiment of Electronics and Communication fom Anna University, Tirunelveli, India in 2011 and currently pursuing Masters Degree in Control and Instrumentation fom National Engineering College, Anna University, Tirunelveli. Her research interests include Image processing applications in various felds and control systems. Dr. B.Sankaragomathi was born on December 26 th 1963 in Sankarankovil. Tirunelveli District. Tamil Nadu. India. Currently she has been working as Dean (Academic) in National Engineering College. K.R.Nagar. Kovilpatti since 1987. She has received her B.Sc degree in Applied Sciences fom Thiyagarajar College of Engineering. Madurai. Madurai Kamaraj University in April 1984. B.Tech degree in Electronics Engineering fom Madras Institute of Technology. Chromepet. Anna University Chennai in May 1987 and M. Tech degree in Instrumentation and Control Systems fom Calicut Regional Engineering College. Calicut University in December 1992. She has obtained her Ph.D degree in the area of Image Processing fom MS University in June 2010. She has published eight papers in International Journals. Her research interests include Biomedical Signal/Image Processing. Process control. VLSI etc. She is the recognized supervisor fr Ph.D. programme at Ana University, Chennai. Presently she is guiding ten Ph.D. Scholars in different disciplines. She has been a reviewer of Inder Science Journals, World Scientifc Journals and Advancement of Modelling and Simulation Techniques in Enterprises (AMSE-France) Journals. T.Ramakrishnan was bor on July 23, 1978 in Tirunelveli, Tamilnadu. India. He received B.E in Electronics & Instrumentation engineering fom Manonmanium Sundaranar University, Tirunelveli, Tamilnadu,India in 1999 and M. Tech in Biomedical Engineering ,fom IITM, Chennai,Tamilnadu.India in 2008 and doing his research work in Anna University, Thiruchirappalli. India. He is currently pursuing Ph.D Degree in medical image processing at IIT Madras. He is working as Assistant professor (Senior Grade) in , Departmentof Electronics & Instrumentation Engineering, National Engineering College K.R Nagar,Kovilpatti, TamilNadu .India. He had published 3 papers in conferences. His research interests include Medical Image Processing, Measurements&Instrumentation.