<title>Microcalcification classification in mammograms using multiwavelet features</title>

Abstract

In this paper, a multiwavelet based feature extraction method is presented and applied to classification of microcalcification clusters in mammograms. Multiwavelet is a natural generalization to scalar wavelet in which more than one scaling function and wavelet are used to further the design degrees of freedom. We extract energy and entropy features from different channels of multiwavelet. Using a real-valued genetic algorithm (GA), the best sets of features along with their optimal weights are found. The optimal weight vector is found such that within-class scatter is minimized and between-class scatter is maximized. For evaluating the individuals in GA, we use the area under Receiver Operating Characteristic (ROC) curve criterion such that the fittest individual has the largest value of area under ROC curve and the worst has the lowest value. To obtain the ROC curve, we use KNN classifier. Several multiwavelets with different features are employed. An area of 0.91 is obtained for Chui and Lian multiwavelet. A comparative study is conducted to show that the performance of multiwavelet is generally better than the packet wavelet in the present application.© (1999) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.