Fuzzy system improves the performance of wavelet-based correlation detectors

Abstract

A fuzzy system is designed to classify features in the output of a wavelets-based correlation filter used for enhancing clusters of fine, granular microcalcifications-an early sign of cancer-in digitized mammograms. Each local peak in the correlation filter output is represented by a set of five features describing the shape, size and definition of the peak. These features-prominence, steepness, distinctness, compactness, and departure-are used in linguistic rules such as "IF prominence is high AND distinctness is mid-ranged AND steepness is mid-ranged THEN it might be a calcification." A fuzzy rule-based system with eight rules is trained to distinguish between microcalcifications and normal mammogram texture. Compared to wavelet processing alone, the fuzzy detection system produces an improvement of around 10% in true positive fraction when tested on a public domain mammogram database.