Computerized Medical Imaging and Graphics 37 (2013) 522–537 Contents lists available at ScienceDirect Computerized Medical Imaging and Graphics journa l h o me pag e: w ww.elsevier.com/locate/compmedimag Level set method with automatic selective local statistics for brain tumor segmentation in MR images Kiran Thapaliya a , Jae-Young Pyun a , Chun-Su Park b , Goo-Rak Kwon a,∗ a Department of Information and Communication Engineering, Chosun University, 375 Seosuk-dong, Dong-gu, Gwangju 501-759, South Korea b Department of Information & Telecommunication Engineering, Sangmyung University, South Korea a r t i c l e i n f o Article history: Received 27 December 2012 Received in revised form 21 May 2013 Accepted 22 May 2013 Keywords: Level set method Active contours Geodesic active contours Chan–Vese model Image segmentation MR images a b s t r a c t The level set approach is a powerful tool for segmenting images. This paper proposes a method for seg- menting brain tumor images from MR images. A new signed pressure function (SPF) that can efficiently stop the contours at weak or blurred edges is introduced. The local statistics of the different objects present in the MR images were calculated. Using local statistics, the tumor objects were identified among different objects. In this level set method, the calculation of the parameters is a challenging task. The cal- culations of different parameters for different types of images were automatic. The basic thresholding value was updated and adjusted automatically for different MR images. This thresholding value was used to calculate the different parameters in the proposed algorithm. The proposed algorithm was tested on the magnetic resonance images of the brain for tumor segmentation and its performance was evalu- ated visually and quantitatively. Numerical experiments on some brain tumor images highlighted the efficiency and robustness of this method. Crown Copyright © 2013 Published by Elsevier Ltd. All rights reserved. 1. Introduction Segmentation of brain tumors from MR images is a difficult task that involves a range of disciplines covering pathology, MRI physics, radiologist’s perception, and image analysis based on the inten- sity, shape and size. Several issues and challenges affect the proper segmentation of brain tumors. According to the World Health Orga- nization (WHO), more than 400,000 people undergo treatment for brain tumors every year. The tumors differ in shape, size and location, and they may appear at different places with different intensities. Therefore, it is very difficult to find the precise tumor in the brain. The accurate segmentation of brain tumors is of great interest. Brain tumors can be classified as primary benign tumors that do not spread elsewhere and secondary or malignant brain tumors that spread from the other location of the body to the brain. Patients suspected of having tumors undergo many diagnostic CT- scans and MRI in hospitals. Although, the radiologist performs these diagnoses, it is very difficult to identify a tumor in the brain due to the involvement of various abnormalities, noise and intensities. Several approaches have been proposed in the field of tumor segmentation. Lee et al. [1] examined the use of the support vector machine (SVM) classification method and Markov random fields (MRFs) for brain tumor segmentation and claimed the superiority of SVM-based approach. Fuzzy-correctness is a useful method that ∗ Corresponding author. Tel.: +82 62 230 6268. E-mail addresses: grkwon@chosun.ac.kr, grkwon72@gmail.com (G.-R. Kwon). has been used to measure the tumor volume in MR images [2,3]. Markov random fields (MRFs) are also popular models for many types of medical image processing used mostly in segmentation [4,5]. Recently, Corso et al. [6] applied the extended graph-shifts algorithm for image segmentation. Active Contour is one of the most powerful methods that have been applied to the image segmentation. The basic idea is to evolve a curve around the object to be detected, and the curve moves toward its interior normal and stops on the true boundary of the object based on the minimization energy. The active contour method can be classified as the snake [7] and level set methods pro- posed by Osher and Sethian [8]. Snake is a semi-automatic method based on an energy minimizing spline guided by the external con- straint forces and pulled by image forces toward the contours of the targets. The main drawbacks of the snake method are its sensitiv- ity to the initial conditions and the difficulties associated with the topological changes for the merging and splitting of the evolving curve. In recent years, the level set method has become popu- lar because it can handle the complex geometries and topological changes. Compared to the snake model, level set methods depend on full-domain energy minimization to implicitly represent the evolution curves and utilize the concepts of dynamics to guide the evolving curve. The level set is in fact a shape driven tool, which using a properly defined speed function, can grow or shrink to take the shape of any complex object of interest. Unlike the traditional deformable model, the level set method does not depend on the parameterization of the surface [9]. This makes it quite attractive and flexible in shape modeling and image segmentation. Another 0895-6111/$ – see front matter. Crown Copyright © 2013 Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.compmedimag.2013.05.003