Feature Selection for the Classification of Microcalcifications in Digital Mammograms Using Genetic Algorithms, Sequential Search and Class Separability
Book in Scopus
- Additional Document Info
- View All
The presence of certain class of microcalcification clusters in digital mammograms is a primary indicator of early stages of malignant types of breast cancer and its detection is important to prevent the disease. One of the approaches we are following for the detection of such microcalcification clusters is the use of Artificial Neural Networks for the recognition of individual microcalcification signals in the mammogram images, and for the classification of groups of these signals into malignant or benign clusters. Since the accuracy and effectiveness of these classifiers strongly depend on the features used to describe the training data for the classifiers and the classification inputs, this chapter presents and compares three different methods for the selection of subsets of features to use: one that selects features based on a class separability criteria, a second one that does the selections implementing a forward sequential search, and a third one that uses Genetic Algorithms (GAs). We found that the use of GAs for selecting the features from microcalcifications and microcalcification clusters results mainly in improvements in the overall accuracy, sensitivity and specificity of the classification. © 2011 John Wiley & Sons, Ltd.