Brain MRI segmentation using initial contour KPCM and optimal speed function for improved level set method

Abstract

Brain tumors are most aggressive kind of diseases, if left untreated may lead to very short life expectancy. Assessment of these tumors is usually done by Magnetic Resonance Imaging (MRI), but MRI produces large amount of data which rule out the manual segmentation in the stipulated time, also restricts the use of accurate quantitative evaluation in the clinical practice. So a reliable and automatic segmentation approach is required. Tumor segmentation in MRI brain image is a basic task in many computer vision problems. A typical methodology is to utilize Fuzzy iterative clustering algorithms that segregates the pixels into a given number of clusters. Nevertheless, the majority of these computations pose a few disadvantages which includes, they are time consuming, sensitive to noise and requires intialization. To overcome these problems, a novel segmentation method based on Particle Swarm Optimization (PSO) and rejection of outliers combined with level set method is developed. So as to enhance the segmentation result obtained from the previous research an Optimized Kernel Possibilistic C-Means (OKPCM) algorithm is proposed. Generally, in KPCM algorithm, an initial value of cluster centers is chosen randomly, but in our proposed method we change the existing KPCM algorithm by taking the cluster center initialization in to account. With the assistance of PSO method the cluster centers are picked ideally and the subsequent fuzzy clustering is utilized to specify an initial level set counter in the proposed improved level set based segmentation. The new improved level set based method has new speed function which efficiently removes boundary (contour) leakage problem is designed. The experiment is carried out on BRATS 2015 database and results are analyzed. The experimental results show that the proposed approach achieved segmentation accuracy of 96.83%.