Novel approaches for the microcalcification problem of non-palpable lesions

Abstract

Fusion of hard and soft information using correlation pattern recognition and fuzzy logic in an integrated detection/classification system based on the DDSM database of pleomorphic microcalcifications provides a detection rate of 0.47 false positives per image at 100% sensitivity using classical matched filter methods on wavelet denoised full mammograms. Advanced correlation filtering yields 0.85 and 0.58 false positives per image at 100% and 87% sensitivity. Classification results generalisable to unseen images at 71.4% specificity for 100% and 96% sensitivity are in a range likely to be clinically acceptable, based on human vision performance of 73.3% specificity at 97.6% sensitivity.