TUMOR DETECTION IN MRI BRAIN IMAGE SEGMENTATION USING PHASE CONGRUENCY MODIFIED FUZZY C MEAN ALGORITHM

Transcription

1 TUMOR DETECTION IN MRI BRAIN IMAGE SEGMENTATION USING PHASE CONGRUENCY MODIFIED FUZZY C MEAN ALGORITHM M. Murugeswari 1, M.Gayathri 2 1 Assoiate Professor, 2 PG Sholar 1,2 K.L.N College of Information Tehnology, India Abstrat - Image segmentation is an essential proedure in many appliations of image proessing. Image segmentation an be lassified to boundary representation and regional representation. Magneti Resonane Image (MRI) is one of the best tehnologies urrently being used for diagnosing Brain Tumor in advaned stages. MRI is a form of medial imaging using nulear magneti resonane of protons in the body. Segmentation proess to extrat suspiious region from omplex medial images is very important. Brain image segmentation is a omplex and hallenging part in the Medial Image Proessing. This projet deals with new approah for MRI Brain image segmentation. The Improved FCM algorithm attempts to partition a finite olletion of elements into a olletion of C Fuzzy Clusters with respet to some given riterion. The proposed algorithm inorporate phase ongrueny features of the neighborhood pixels with FCM lustering. The proposed algorithm is effiiently segmented the MRI brain image. Keywords: Improved Fuzzy C Mean, Segmentation, Phase ongrueny feature, Tumor Detetion. I. INTRODUCTION IMAGE segmentation is one of the key tehniques in image understanding and omputer vision. The task of image segmentation is to divide an image into a number of non-overlapping regions, whih have same harateristis suh as gray level, olor, tone, texture, et. A lot of lustering based methods have been proposed for image segmentation. Among the lustering methods, one of the most popular methods for image segmentation is fuzzy lustering, whih an retain more image information than hard lustering in some ases. Fuzzy -means (FCM) algorithm is one of the most widely used fuzzy lustering algorithms in image segmentation. FCM algorithm was first introdued by Dunn and later extended by Bezdek. Although the onventional FCM algorithm works well on most noise-free images, it fails to segment images orrupted by noise, outliers and other imaging artifats. Its non-robust results are mainly beause of ignoring spatial ontextual information in image and the use of non-robust Eulidean distane. Digital Image Proessing onsists of several steps. The first step is image aquisition-that is, to aquire a digital image. After a digital image has been obtained, the next step deals with preproessing that image. The key funtion of preproessing is to improve the image in ways that inrease the hanes for suess of the other proesses. The next stage deals with image segmentation. Image segmentation partitions an input image into its onstituent parts or objets. The next step is representation and desription. Representation is the transformation of raw data into a desriptive form suitable for omputer proessing. Desription deals with extrating features that result in some quantitative information of interest. Suh desriptions are neessarily task speifi. The last step is reognition and interpretation. Reognition is the proess that assigns a label to an objet based on the information of the objet. Interpretation assigns meaning to reognized objets. Image segmentation is an essential proedure in many appliations of image proessing. Image segmentation an be lassified to boundary representation and regional representation. Eah representation is identifiation of homogeneous regions or ontours of loal inhomogeneity, respetively. Segmentation algorithms for monohrome images generally are based on one of two basi properties of gray-level values: disontinuity and similarity. In the first ategory, the approah is to partition an image based on abrupt hanges in gray level. The prinipal areas of interest within this ategory are detetion of isolated points and detetion of lines and edges in an image. The prinipal approahes in the first ategory are based on edge detetion, and boundary detetion. Basially, the idea underlying most edge-detetion tehniques is the omputation of a loal derivative operator. The first derivative of the gray-level profile is positive at the leading edge of a transition, negative at the trailing edge, and zero in areas of onstant gray level. Hene the magnitude of the first derivative an be used to detet the presene of an edge in an image. If many lines pass through the point, but they all satisfy a line equation for varying values of slope and 190

2 interept. In the parameter (slope and interept) spae, when the line intersets were found, those interset points mean that they are on the same line. Therefore, the edges and boundaries are found by this tehnique. The prinipal approahes in the seond ategory are based on thresholding and label region algorithm. The onept of segmenting an image is based on disontinuity or similarity of the gray-level values of its pixels. II. RELATED WORK The Data Base is used to store the tissue lass. The MRI Brain Image is automatially segmented as erebrospinal fluid (sf), gray matter (gm), white matter (wm), and mixtures of sf and gray matter, aording to[1].this will helpful in automati detetion of brain tumor in MRI Brain Image. The integration of membership funtion in to spatial information of input image ompensate the effet of noise aording to[2]. The trade-off weighted fuzzy fator is introdued in the improved fuzzy means lustering as[3]. The fuzzy membership of pixels has influened in the prior probability of an image pixel in its immediate neighborhood as[4]. The segmentation proess is used to partition an image into different regions with respet to feature extration [5]. The segmentation of MRI image using fuzzy with some modifiation give better improvement in [6]. The possibility to impliitly segment Tumor-bearing brain images by atlas-based registration is offered as [7]. A loal intensity lustering property of the image intensities is derived, and a loal lustering riterion funtion is defined for the image intensities in a neighborhood of eah point. The loal lustering riterion funtion integrated into the neighborhood enter to give a global riterion of image segmentation [8]. The auray of segmentation is obtained by the ratio between sum of the orretly lassified pixels to the total number of pixel [9]. The segmentation method is deided by the neighboring pixel and loations [10]. The effetiveness of spatial onstraints ontributes exploitation of spatial ontextual information [11]. III. WORK MODULE The detetion of brain tumor in MRI Brain Image is done with the help of data base whih already have the information about MRI tissue. The Training Input has some noise. The input is preproessed to remove the noise. The median filter is used to remove the noise and it help full to spatial detetion. Most of our training input is segmented with spatial feature. The median filter gives the mean value among the neighborhood pixels. It Give better result than other filtering method. For better result the ombined form median filter is omposed of a median filter and a seond median filter that filters the error signal. The proposed segmentation method gives the additional iteration and lustering and maps the best mathing. This will gives improvement in the segmentation using Improved Fuzzy C Means lustering tehnique. The segmented image using proposed algorithm then ompared with the data base system. By using the morphologial operation the data base system is used for deteting the tumor in the segmented output. Input Preproessing Spatial Congrueny Feature added with FCM Segmented output Figure 1Blok Diagram of Proposed Method IV. PREPROCESSING The key funtion of pre-proessing is to improve the image in ways that inrease the hanes for suess of the other proesses. The next stage deals with image segmentation. The double median filter is omposed of a median filter and a seond median filter that filters the error signal. The error signal is the differene between the input signal and the filtered signal after the first median filter. The traditional median filter is used. The median filter is a nonlinear filter often used to remove noise. Suh noise redution is a typial preproessing step to improve the results of later proessing. It preserves edges while removing noise. V. SEGMENTATION Image segmentation is an essential proedure in many appliations of image proessing. Brain image segmentation is a omplex and hallenging part in the Medial Image Proessing. The new approah for deteting tumor in MRI Brain done with modified FCM iterative proess. A new level of region growing algorithm has ome up that IFCM Iterate and map the luster with IFCM Segmented output 191

3 overomes severe limitations of older approahes. This will extend those ideas by putting them in a general framework, whih allows for the ombination of a variety of segmentation algorithms, resulting in great potential to optimize all aspets of the segmentation proess. Segmentation is the proess of partitioning an image into different segments. In medial imaging, these segments often orrespond to different tissue lasses, organs, pathologies, or other biologially relevant strutures. Medial image segmentation is made diffiult by low ontrast, noise, and other imaging ambiguities. Although there are many omputer vision tehniques for image segmentation, some have been adapted speifially for medial image omputing. This projet is foused on MRI Brain images segmentation and analyse about the brain tumor. These MRI Brain images are segmented and the results of segmentation are used for deteting the brain tumor. SEGMENTATION USING IMPROVED FUZZY C MEANS CLUSTERING: Clustering is a proess for lassifying objets or patterns in suh a way that the samples of the same luster are more similar to one another than samples belonging to different lusters. Fuzzy lustering is a soft segmentation method. Fuzzy - means (FCM) algorithm is the most popular method used in image segmentation beause it has robust harateristis for ambiguity and an retain muh more information than hard segmentation methods. Step1. 1) Set the number of the luster prototypes hange from 2 to max. 2) Initialize randomly those prototypes and set ε>0 to a very small value. Step2. Compute the loal similarity measures sij for all neighbor windows over the image using s s ij s g ij, j 0 s ij = 0, j = 0 Step3. Compute linearly-weighted summed image ξ in terms of ξ i. ξ i = s jεni ij xj jεni s ij Step4. Update the partition matrix using u il = (ξ i v i ) 2 j=1 (ξ i v j ) 2 Step5. Update the prototypes using v i = q l=1 γ l u m il ξ l q m l=1 γ l u il Repeat Steps 4-5 until the following termination riterion is satisfied: Vnew Vold <ε Where V= [v1, v2 v] are the vetors of luster prototypes. SEGMENTATION USING MAPPING THE ITERATIVE OF IFCM CLUSTERING (proposed method1 PM1): Conatenate the IFCM lustering output. Initialize the luster pixel values. Obtained the repmatrix with respet to the output of IFCM Clustering. Now get Conatenation of rep matrix Find the distane between the Conatenation of input and the Conatenation luster pixel. Update the distane with respet to the rep matrix. Get the partition matrix with respet to the updating of distane. Update the luster pixel value with respet to the partition matrix. Find out the temporary matrix from the partition matrix. Map the values of temporary matrix into the output of IFCM lustering. Repeat this still max (temporary matrix) < SEGMENTATION USING PHASE CONGRUENCY FEATURES WITH FCM (proposed 2): Step1: Set the number of lusters(c), degree of fuzziness,stop riterion and neighborhood size. Step2: Calulate phase ongrueny features and define the neighborhood onfiguration for eah pixel. Step3: Initialize the enter of the lusters v i I =1,2,..C. Step4: Calulate the new similarity measure using, D ij = w ij x j v i 2 =(1-αs ij ) x j v i 2 Step5: Calulate the membership value, Step6: u ij = [ ( D ij ) 1 k=1 ] 1 D kj Calulate the new membership values, 192

4 Step7: u ij new = β u ij M ij β u kj M ij k=1 PARAMETERS IFCM PM1 PM2 ENTROPY Update v i using u ij new v i = N j=1 u ij m x j N u m j=1 ij Repeat the steps from 4 to 7 until it reahes the stopping riteria, max i [1,] v i l v i l+1 <. MEAN SQUARED ERROR PEAK SIGNAL TO NOISE RATIO PE -PROJECTION ERROR e e e VI. RESULTS VII. Table 1 Comparison of Parameters CONCLUSION Figure 2 segmented output (IFCM) Thus the image enhanement is done using the median filtering whih is often used to remove noise. Median filtering is very widely used in digital image proessing beause, under ertain onditions, it preserves edges while removing noise. The median filtered image is well suited for MRI brain image segmentation using lustering. The MRI brain image segmentation is done by improved fuzzy means lustering. And the parameters Entropy, Mean square error, Peak signal to noise ratio, Re-projetion error have been evaluated. For better performane of image segmentation the Entropy should be small and the other parameters should be improved. The Improved FCM algorithm attempts to partition a finite olletion of Fuzzy C lusters with respet to some given riterion. The outome will be used to further analysis of MRI image more aurately. Referenes: Figure 3 segmented output (proposed method 1) Figure 4 segmented output (proposed method 2) [1] Colm Elliott, Douglas L. Arnold, D. Louis Collins, and Tal Arbel, Temporally Consistent Probabilisti Detetion of New Multiple Slerosis Lesions in Brain MRI, IEEE Transations On medial imaging, Vol. 32, No. 8, August [2] Ivana Despotović, Student Member, IEEE, Ewout Vansteenkiste, and Wilfried Philips, Member, IEEE, Spatially Coherent Fuzzy Clustering for Aurate and Noise-Robust Image Segmentation, IEEE Transations On Image Proessing, Vol. 22, No. 2, APRIL [3] Maoguo Gong, Member, IEEE, Yan Liang, Jiao Shi, Wenping Ma, and Jingjing Ma, Fuzzy C-Means Clustering With Loal Information and Kernel Metri for Image Segmentation, IEEE Transations On Image Proessing, Vol. 22, No. 2, February [4] Hui Zhang, Member, IEEE, Q. M. JonathanWu, Senior Member, IEEE, and Thanh 193

5 Minh Nguyen, A Robust Fuzzy Algorithm Based on Student s t-distribution and Mean Template for Image Segmentation Appliation, IEEE Signal Proessing Letters, FEBRUARY [5] D. Salas-Gonzalez a,n, J.M.Go rriz a, J.Ramı rez a, M.Shloegl b, E.W.Lang b, A.Ortiz, Parameterizationofthedistributionofwhiteandgrey matterinmri using the a-stable distribution, Elsevier JAN [6] Zexuan Ji, Yong Xia, Member, IEEE, Quansen Sun, Qiang Chen, Member, IEEE, Deshen Xia,and David Dagan Feng, Fellow, IEEE, Fuzzy Loal Gaussian Mixture Model for Brain MR Image Segmentation IEEE Trans, May [7] Stefan Bauer*, Student Member, IEEE, Christian May, Dimitra Dionysiou, Georgios Stamatakos, Member, IEEE, Philippe B uhler, and Mauriio Reyes, Member, IEEE Multisale Modeling for Image Analysis of Brain Tumor Studies IEEE Trans, JAN [8] Chunming Li, Rui Huang, Zhaohua Ding, J. Chris Gatenby, Dimitris N. Metaxas, Member, IEEE, and John C. Gore, A Level Set Method for Image Segmentation in the Presene of Intensity Inhomogeneities With Appliation to MRI IEEE Trans, July [9] Stelios Krinidis and Vassilios Chatzis, A Robust Fuzzy Loal Information C-Means Clustering Algorithm IEEE Transations On Image Proessing, Vol. 19, No. 5, May [10] Shan Shen, William Sandham, Malolm Granat, and Annette Sterr, MRI Fuzzy Segmentation of Brain Tissue Using Neighborhood Attration with Neural-Network Optimization IEEE Transations On Information Tehnology In Biomediine, Vol. 9, No. 3, September [11] Songan Chen & Daoqiang Zhang, Robust Image Segmentation Using FCM With Spatial Constraints Based on New Kernel-Indued Distane Measure, IEEE Transations On Systems, Man, And Cybernetis Part B: Cybernetis, Vol. 34, No. 4, August