Fuzzy system improves the performance of wavelet-based correlationdetectors

摘要

A fuzzy system is designed to classify features in the output of awavelets-based correlation filter used for enhancing clusters of fine,granular microcalcifications-an early sign of cancer-in digitizedmammograms. Each local peak in the correlation filter output isrepresented by a set of five features describing the shape, size anddefinition of the peak. These features-prominence, steepness,distinctness, compactness, and departure-are used in linguistic rulessuch as “IF prominence is high AND distinctness is mid-ranged ANDsteepness is mid-ranged THEN it might be a calcification.” A fuzzyrule-based system with eight rules is trained to distinguish betweenmicrocalcifications and normal mammogram texture. Compared to waveletprocessing alone, the fuzzy detection system produces an improvement ofaround 10% in true positive fraction when tested on a public domainmammogram database