Preface -- Acknowledgments -- Introduction -- Experimental Setup and Databases of Mammograms -- Multidirectional Gabor Filtering -- Landmarking Algorithms -- Computer-aided Detection of Bilateral Asymmetry -- Design of Contour-independent Features for Classification of Masses -- Integrated CADe/CADx of Mammographic Lesions -- Concluding Remarks -- References -- Authors' Biographies.

The identification and interpretation of the signs of breast cancer in mammographic images from screening programs can be very difficult due to the subtle and diversified appearance of breast disease. This book presents new image processing and pattern recognition techniques for computer-aided detection and diagnosis of breast cancer in its various forms. The main goals are: (1) the identification of bilateral asymmetry as an early sign of breast disease which is not detectable by other existing approaches; and (2) the detection and classification of masses and regions of architectural distortion, as benign lesions or malignant tumors, in a unified framework that does not require accurate extraction of the contours of the lesions. The innovative aspects of the work include the design and validation of landmarking algorithms, automatic Tabár masking procedures, and various feature descriptors for quantification of similarity and for contour independent classification of mammographic lesions. Characterization of breast tissue patterns is achieved by means of multidirectional Gabor filters. For the classification tasks, pattern recognition strategies, including Fisher linear discriminant analysis, Bayesian classifiers, support vector machines, and neural networks are applied using automatic selection of features and cross-validation techniques. Computer-aided detection of bilateral asymmetry resulted in accuracy up to 0.94, with sensitivity and specificity of 1 and 0.88, respectively. Computer-aided diagnosis of automatically detected lesions provided sensitivity of detection of malignant tumors in the range of [0.70, 0.81] at a range of falsely detected tumors of [0.82, 3.47] per image. The techniques presented in this work are effective in detecting and characterizing various mammographic signs of breast disease.