FEATURE SELECTION AND TUMOR IDENTIFICATION IN BRAIN MAGNETIC RESONANCE IMAGING USING HYBRID SWARM INTELLIGENCE AND MACHINE LEARNING

Abstract

Automatic classification of Brain MR Images is one of hotspots in the field of Medical Imaging. Feature Selection of Brain MRI is crucial and it affects the classification results; however selecting optimal Brain MRI features is difficult. Particle Swarm Optimization (PSO) is an evolutionary computing technique which was developed observing the social behavior of bird flocks. PSO is an intelligent technique containing properties like adaptation and self-organizing, these intelligent properties make PSO one of the best techniques for searching the optimal solution for optimization problems. The difficulty mentioned above is solved using the Discrete Binary PSO technique which proved it self quite successful. Population of Particles is generated and the search strategy is applied so that an efficient feature selection process for Brain MRI is proposed. Classification of normal and tumor Brain MRI is carried out using PSO and experimental results show that it obviously reduces the number of features and at the same time it achieves high accuracy level. Three different classifiers, Support Vector Machine (SVM), Naïve Bayes Classifier (NB) and K-Nearest Neighbor Classifier (KNN) are utilized to serve as the ‘fitness function’ for PSO. A comparison of achieved accuracy with minimum number of features for different classifiers serving PSO is also provided. PSO-SVM and PSO-KNN are observed to achieve better accuracy level using minimum number of selected features than PSONB.