A Feature Selection Method Base on GA for CBIR Mammography CAD
Yi Chen, Yihua Lan, Haozheng Ren · 2012
Feature selection is a very important step for almost all of the feature-based mammography computer-aided detection and diagnosis (CAD) system. The purpose of this study was to develop and evaluate a feature selection method for content-based image retrieval (CBIR) CAD system. After examine the problems in tradition genetic algorithm (GA), it is found that there usually are different feature subsets when running genetic algorithm (GA) for feature selection in different time, the reason of it is that the initial values for genes in GA are always generated randomly. Well then, which feature subset could be selected as the optimal one? Motivated by this, we proposed a method for feature selection which called F-GA (Frequency-GA). In the proposed method, GA was run m times for m different feature sub-sets. Then emergence frequency of each feature was counted. At last, those features which have highest frequency (i.e., %p, p is a threshold) were selected to form the ultimate feature sub-set. To test and evaluated the performance of the proposed method, experiments on a public available data set were carried out. The experimental results demonstrated the effect of the proposed method.