A hybrid feature selection method using both filter and wrapper in mammography CAD

Feature selection methods are critical in mammography computer-aided diagnosis and clinical decision support systems. However, searching for an optimal or near optimal feature subset is still a difficult task. After examining the problems with both filter and wrapper methods in feature selection, we propose a hybrid feature selection method using both Filter and Wrapper by taking advantage of both approaches in a content-based image retrieval computer-aided diagnosis. At first, we used step-by-step linear discriminative analysis (SLDA) algorithm, which belongs to filter approach, to remove irrelevant features, and then we used genetic algorithm (GA, wrapper approach) to remove useless features and achieve the ultimate feature subset. To test and evaluate the proposed method, we compared our method with using either GA or SLDA algorithm singly; the result is encouraging.