Classification of mass and normal breast tissue: Feature selection using a genetic algorithm

We investigated a genetic algorithm (GA) based method for feature selection in computer-aided diagnosis. The GA-based feature selection method was guided by the probabilities of survival which were determined by the fitness of the combinations of features. The fitness of a combination of features was measured by the classification performance when the features were used jointly for classification. We studied the application of GA-based feature selection to the problem of differentiation of regions of interest (ROIs) on mammograms as either mass or normal breast tissue. The classifier used for measuring the fitness of chromosomes was a linear discriminant classifier, and the fitness measure was based on the area (A(2)) under the receiver operating characteristic (ROC) curve. We studied the effect of GA parameters on classification accuracy, and compared the classification results to other feature selection techniques.