A Hybrid Swarm Optimization approach for Feature set reduction in Digital Mammograms

In this paper a CAD (Computer Aided Diagnosis) system is proposed to optimize the feature set using hybrid of Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) technique called Genetical Swarm Optimization (GSO) in Digital Mammogram. Even though PSO is a good optimization technique, it may be trapped in local minima and may prematurely converge. So, the genetic operators are used in PSO to overcome the difficulties. Feature selection plays a major role in diagnosis of mammogram. Gray Level Co-occurance Matrix (GLCM) texture features are extracted from the mammogram. All the extracted features do not help in detection of abnormality in a mammogram, so it is intended to reduce the feature set to improve classification accuracy. In this work, experiments are conducted on MiniMIAS database and Support Vector Machine (SVM) classifies the mammograms into normal and abnormal mammograms. Performance of GSO is compared with GA and PSO by means of Receiver Operating Characteristic (ROC) curve. Results show that, the GSO convergence is better than both PSO and GA; GSO based SVM (GSO-SVM) classifier exhibits superior performance with an accuracy of 94% which is approximately 1% higher than GA based SVM (GA-SVM) and PSO based SVM (PSO-SVM) classification.