The new approaches for the classification of microcalcifications in the visibility-enhanced mammogram images

In this thesis, we proposed a new computer aided diagnosis (CAD) system which has the pixel assignment-based spatial filter to enhance the visibility of microcalcifications in mammograms. The proposed filter used as a part of the CAD system, first sums the absolute values of the differences between the center pixel of interest and its 8-neighbors, and then assigns this summed value to that center pixel of interest. This process was repeated for each pixel of the image. Then, the contrast stretching was applied to the filtered-images in order to ensure the fairness in the comparison of the contribution of the proposed filter to the CAD system. The feature vectors were extracted from the filtered-contrast stretched-images by using the widely used and well-known statistical parameters which are mean, standard deviation, entropy, energy, skewness and kurtosis. The extracted feature vectors were applied to the inputs of support vector machines, linear discriminant analysis and multilayer perceptron neural network classifiers in order to detect the mammograms with microcalcifications as absent/present and to classify the detected microcalcifications as benign and malignant. In order to evaluate the effect of the proposed filter on the success of detection and classification, it was compared to widely used Laplace, Gabor, Top-hat and Gaussian high pass filters. In the implemented experiments, the comparison showed that this filter provided higher contribution to the success of detection and classification than the others, and hence enhanced the visibility of microcalcifications in mammograms. Finally, it can be concluded that the proposed filter can contribute to the development of the state-of-art methodologies and can be used with a CAD system as a diagnostic decision support mechanism in the analysis of mammograms.