A Novel MRI Brain Edge Detection Using PSOFCM ... - IEEE Xplore

2014 International Conference on Electronic Systems, Signal Processing and Computing Technologies

A Novel MRI Brain Edge Detection Using PSOFCM Segmentation and Canny Algorithm Romesh Laishram Electronics & Communication Engineering. Manipur Institute of Technology, Imphal, Manipur, India E-Mail: [email protected]

Wahengbam Kanan Kumar, Anshuman Gupta, Khairnar Vinayak Prakash M.Tech (VLSI Design), Dept. of Electronics and Communication Engineering Maharishi Markandeshwar University, Ambala, India Email: [email protected], [email protected], [email protected] Edge detection and clustering are two basic segmentation methods among the various techniques. Edge detection is basically aimed at identifying the image brightness discontinuities especially along the edges where the intensity tends to change sharply. Canny edge detection is a premium technique used for detecting the edge in an image. Clustering is a process whereby a data set is replaced by clusters, which are collections of data points that "belong together". Thus, it is natural to think of image segmentation as image clustering i.e. the representation of an image in terms of clusters of pixels that "belong together". The specific criterion to be used depends on the application. Pixels may belong together because of the same colour or similarity measure. This work is an enhanced form of edge detection methodology that aids to obtain the best results out of MRI brain images. The method uses fuzzy C means (FCM) [2] clustering algorithm for segmenting the image prior to edge detection process. The main purpose of using PSO is to reach the Global minima of the clustering objective function. In [3] the application PSO in image segmentation problem is investigated.

Abstract— Introduction of many Image processing and segmentation tools has undoubtedly presented the procedures of mapping the human brain in a more efficient way. This paper attempts to pull out a new and a practical approach for enhancing the underlying delicate architectures of the human brain images captured by a Magnetic Resonance Imaging (MRI) machine in a much better way. Edge detection is a fundamental tool for the basic study of human brain particularly in the areas of feature detection and feature extraction. The edge detection methodology presented in this paper relies on two basic stages: Firstly, the original MRI image is subjected to image segmentation which is done using Particle Swarm optimization incorporating Fuzzy C Means Clustering (PSOFCM) technique. And secondly, canny edge detection algorithm is used for detecting the fine edges. After implementation it was found that this technique yields better edge detected image of the human brain as compared to other edge detection methods as discussed below Keywords—Edge ,MRI,PSOFCM,Canny,segmentation.

I.

INTRODUCTION

Magnetic Resonance Imaging (MRI) or nuclear magnetic resonance imaging (NMRI) or magnetic resonance tomography (MRT) is a medical imaging technique used in radiology to visualize internal structures of the body in detail. It makes use of the property of nuclear magnetic resonance (NMR) to image nuclei of atoms inside the body. It provides excellent contrast between the different soft tissues of the body, which makes it especially useful in imaging the brain, muscles, etc. The images produced by an MRI scanner are best described as slices through the brain. MRI has an added advantage of producing images, which slices the brain images in both horizontal and vertical planes. For further processing, these MRI images must be categorized and analysed using edge detection.

The “Particle Swarm Optimization” model was brought into light for the first time by Russel Ebenhart and James Kennedy in 1995. As the name suggest the basic operation simulates the behaviour of flocks of birds or the sociological behaviour of a group of people or schooling of fish. Modelled on the mechanism of evolution and natural movement this algorithm provides an alternative tool for locating optimal solutions. Here swarm of several particles are used as the population size to find the best solution which is analogous to that of Genetic Algorithm in which some probable solutions are initialises as the population size. Each particle keeps track of their current positions which correspond to potential solutions of the function to be minimized. Each particle also keeps track of the speed and direction of travel by the particle. Similar to genetic algorithm a fitness value is also associated with each particle which directly depends on the particle’s current position. Additionally each particle also remembers its personal best positions which on comparison with the personal best solutions of other particle are used to determine the overall

Image segmentation is an attempt to partition a digital image into multiple segments which is also known as sets of pixels or super pixels. The goal of segmentation is to simplify or change the image into something that is more meaningful and easier to analyse.

978-1-4799-2102-7/14 $31.00 © 2014 IEEE DOI 10.1109/ICESC.2014.78

398

III. PARTICLE SWARM OPTIMIZATION INCORPORATING FUZZY C MEANS (PSOFCM)

best solution or population. Moreover, this algorithm is also metaheuristic as it makes few or no assumptions about the problem being optimized and can search very large spaces of candidate solutions. To be more specific it does not use gradient of the problem being optimized, which means PSO does not require that the optimization problem being differentiable as is required by classic optimization methods. So it can therefore be used on problems that are partially irregular, noisy, change over time, etc. II.

PSO algorithm works by having a population (called a swarm) of candidate solutions (called particles).These particles are moved around in the search-space according to a few simple formulae. The movements of the particles are guided by their own best known position in the search-space as well as the entire swarm's best known position. When improved positions are being discovered these will then come to guide the movements of the swarm. The process is repeated and by doing so it is hoped, but not guaranteed, that a satisfactory solution will eventually be discovered. The mathematical steps used to formulate the PSOFCM algorithm is discussed below: Consider a D-dimensional search-space S D and a swarm consisting of I particles. The i-th particle is in effect a Ddimensional vector: Xi = (xi1, x i2, x iD)T S, (in our program we initialized the number of particles in the swarm or swarm size as 20) The velocity of this particle is also a D-dimensional vector, Vi = (vi1, v i2,……, v iD)T The best previous position encountered by the i-th particle in S is denoted by: Yi = (yi1, y i2,……, y iD)T These particles’ initial velocities and positions are randomly initialized. Let Y* be the global best position amongst all the particles, and t be the program iterations. During each iteration, the velocity and position of each particle/swarm is updated using (4) and (5) as given below. Further in this algorithm we are trying to minimize a function Yi = f (Xi) using Fuzzy c means algorithm. The objective function of the fuzzy c means clustering algorithm is used for achieving the global minima of the clustering objective. In this way by minimizing the objective function value FCM keeps track of the best particle or best position ever.

FUZZY C MEANS ALGORITHM

Segmentation is greatly being improved by using the FCM algorithm [4]-[6] instead of using K-Means Clustering algorithm. It divides the images into number of homogenous classes effectively. It has some success to detect the noise from an image. The Traditional FCM algorithm is an iterative algorithm that produces optimal C partitions, centers V= {v1, v2,…, vc}. Let unlabelled data set X={x1, x2,…, xn} be the pixel intensities, where n is the number of image pixels to determine their membership. The FCM algorithm tries to partition the dataset X into C clusters. The standard FCM objective function is defined as follows. c n m 2 (1) J m (U, V) (u ik ) d (x k , v i ) i1k 1 2 where d (x k , v i ) represents the square of the Euclidean distance between the pixel intensity value xk and the c centroid value vi along with constraint u ik 1 ,and the i1 degree of fuzzification m≥1.A data point xk belongs to the specific cluster vi that is given by the membership value uik of the data point to that cluster. Local minimization of the objective function J m (U, V) is accomplished by repeatedly adjusting the values of uik and vi according to the following equations. 1 1

c d 2 x , v (m1) k i U ik (2) 2 j 1

d (x k , v j )

Vi(t+1) = W Vi(t)+c1r1(Yi(t) - Xi(t)) + c2r2(Y* - Xi(t))

Where Vi is calculated using the following equation n m (u ik ) x k k 0 Vi (3) n m (u ik ) k 0 As Jm is iteratively minimized, the centroid matrix is more stable. This iteration is terminated when the difference between the maximum of current centroid value, maximum of previous iteration centroid value is less than the 0.0001. The value 0.0001 is predefined termination threshold. Finally, all homogeneous pixels are grouped into the same class to evaluate the Fuzzy C-means algorithm.

and Xi (t+1) = Xi (t) + Vi (t+1)

Where Yi = Yi, if f (Xi) ≥ f (Yi) = Xi, if f (Xi) < f (Yi) Y* 0, Y1, ….. Yi} such that f (Y*) = min (f (Y0), f (Y1),……., f (Yi))

r1 and r2 are elements from two uniform random sequences in the range (0,1).W is called inertia weight matrix representing weights of Vi(t) as a contribution to Vi(t+1).

399

It is quite clear from (7) that Y* is the global best position amongst all the particles. The value of each component of Vi vector is clamped to range [-vmax, vmax] to prevent PSOFCM from leaving the search space. c 1 and c2 are acceleration co-efficient which controls the displacements of a particle in a single iteration. In this work we used W = 0.5 and c 1 = c2 = 2 IV.

should not create false edges. The resulted final images with their best edges details are very much essential and helpful for further brain MRI image processing and analysis. The main idea of this present work is the application of PSO based optimization in FCM problem. However for medical image processing the desired result should be highly objective. The present work may be use as a preprocessing stage for further analysis.

SIMULATION RESULTS

In this section we present the simulation results of the MRI image edge detection problem. The performance of PSOFCM (with canny) based edge detection algorithm is compared with the GAFCM (with canny) for four different MRI images as shown in figure 1-4 respectively. The simulation is performed in MATLAB environment.

REFERENCES [1]

Romesh Laishram, W.Kanan Kumar Singh, N.Ajit Kumar, Robindro.K, S.Jimriff, “MRI Brain Edge Detection Using GAFCM Segmentation and Canny Algorithm”, International Journal of Advances in Electronics Engineering – IJAEE,volume 2 - Issue 3, ISSN:- 2278-215X, pp. 168-171,December 8,2012 [2] Rafael C. Gonzalez, Richard E.Woods, “Digital Image Processing”, Pearson Education, Second Edition, ISBN 81-7758-168-6, 2005. [3] Amit Konar, “Computational Intelligence Principles, Techniques and Applications”, Springer edition, ISBN 3-540-20898-4 Springer Berlin Heidelberg, New York [4] R.Venkateswaran, S.Muthukumar, “Genetic Approach on Medical Image Segmentation by Generalized Spatial Fuzzy C- Means Algorithm”, 2010 IEEE International Conference on Computational Intelligence and Computing Research, pp: 210—213.. [5] Yingjie Wang, “Fuzzy Clustering Analysis Using Genetic Algorithm”, and ICIC International @ 2008 ISSN 1881-803 X, pp: 331—337. [6] A.Halder , S.Pramanik , A.Kar, “Dynamic Image Segmentation using Fuzzy C-Means based Genetic Algorithm”, International Journal of Computer Applications (0975 – 8887) Volume 28– No.6, August 2011, pp: 15 – 20. [7] Huang Ying, Wang Weixing, and Li Weisheng, “Anisotropic Filter Based Modified Canny Algorithm”. Fourth International Conference on Fuzzy Systems and Knowledge Discovery, Vol: 1, pp: 736 -740, Aug 2007. [8] Natarajan.P, “New Fangled MRI Brain Edge Detection Using Enhanced Canny Algorithm”, 2011 IEEE International Conference on Computational Intelligence and Computing Research, pp: 84—87. [9] Davoud Sedighizadeh and Ellips Masehian, “Particle Swarm Optimization Methods, taxonomy and applications”, International Journal of Computer Theory and Engineering (1793-8201) Vol. 1, No. 5, December,2009, pp: 486-502 [10] J. Kennedy and R.C. Eberhart, "Particle swarm optimization" , Proceeding of the 1995 IEEE International Conference on Neural Networks (Perth,Australia), IEEE Service Centre, Piscataway, NI, (1995), Iv: 1942-1948

The comparison result shows that PSOFCM based canny edge detection method yields better result than GAFCM based canny edge detection. It can also be seen that more number of edges are also fetched and the image is also segmented and clustered properly as compared to GAFCM based canny edge detection method; each and every edge is fetched without missing even a single curve. This new algorithm goes even to the most delicate parts of the brain for extracting the fine edges and shows astounding performance. This shows that the novel technique that is being incorporated in this paper (i.e. PSOFCM based canny edge detection) is the best edge detection methodology for MRI Brain image edge detection as compared to GAFCM based canny edge detection technique [1]. V.

CONCLUSION

In this paper we have presented PSO based optimization of FCM algorithm i.e. PSOFCM and its application in human Brain MRI image segmentation for automatic detection of abnormalities. The segmented image is further processed for edge detection using canny edge detector. The result obtained through PSOFCM yields better edge detected image compared to GAFCM segmentation [1]. It may also be noted it was proven that GAFCM based canny edge detection is better than ordinary canny edge detection. The whole comparison is based on canny edge detection’s principles: 1) Good detection: the algorithm should mark as many real edges in the image as possible. 2) Good localization: edges marked should be as close as possible to the edges in the real image. 3) Minimal response: a given edge in the image should only be marked once,and where possible,image noise

400

(a)

(b)

(c)

Figure 1. (a) Original Image (b) PSOFCM with Canny algorithm (c) GAFCM with Canny algorithm

(a)

(b)

(c)

Figure 2. (a) Original Image (b) PSOFCM with Canny Algorithm (c) GAFCM with Canny algorithm

(a)

(b)

(c)


(a)

(b)

(c)


401