A Neural Network Method for Mammogram Analysis Based on Statistical Features

AlxtmcfIn this paper, wc prcscr a novel approach to the problem of computer-aided analysis of digital mammograms for breast cancer detection. The algorithm devcloped herc classifics mammograms into norma; & abnormal. First, thc structures in mammograms produced by normal glandular tissue of varying dcnsity are eliminated using a Wavelet Transform (WT) based local averagc subtraction. Then the linear markings formed by the normal connective tissue are identified and removed. Any abnormality that inay exist in the mammogram is thercforc cnhanced i n tlic rcsidual image, which makes the decision rcgardiilg thc normality of the mammogram much easicr. Statistical descriptors based on high-order statistics derived from the residual image are applied to a Probabilistic Neural Network (PNN) for classification. Using thc mammographic data fiom the Mammographic Image Analysis Society (MIAS) databasc a recognition scorc of 7 1%) was achicvcd.