Published Paper Details:

ABSTRACT Mammogram is one of the methods in understanding the presence of Microcalcification (MC) in the breast of women. MC is an initial form of breast cancer. In this research work, the mammogram image is decomposed using wavelet filters. Daubechi wavelet has been used to decompose the given mammogram image into 5 levels. In each level of decomposition, the approximation is further decomposed. Statistical features are calculated from the approximations of each image from each level of decomposition. These statistical features form training patterns and testing patterns of the proposed ANN algorithms. The patterns are used as inputs for the Back-Propagation Algorithm (BPA), echo state neural network algorithm, Fuzzy logic algorithm and Firefly algorithms. Ten mammogram images are used from Mammographic Image Analysis Society (MIAS) database. Two thousand unique patterns are obtained from the 10 images. From the 2000 patterns, 1000 patterns are used for training the BPA/ESNN/FL/FF. The remaining 1000 patterns are used for testing the BPA / ESNN/FL/FF algorithms. In the training patterns, 500 are with MC and remaining 500 are without MC. In the testing patterns, 500 are with MC and remaining 500 are without MC. Performance of the algorithms is assessed by receiver operating characteristic curve, sensitivity, specificity and accuracy.