An optimized SVM kernel for texture classification and its application in microcalcification detection

Mammograms, commonly used to diagnose breast cancer, are difficult medical images to interpret. Computer aided diagnosis (CAD) systems have the potential to assist radiologists by locating suspicious regions in the mammograms for more detailed examination. One approach is for CAD systems to detect microcalcification. This approach uses classification of texture features and has applications for the detection of breast cancer as well as other abnormalties in medical images. The Support Vector Machine (SVM) has been shown to be effective in texture classification. SVM performs well in high dimensional space such as the space spanned by texture images. The kernel function in SVM algorithm implicitly performs feature extraction. Since SVM is basically suited for two-class classification problems, it is potentially a good choice for several different medical imaging which deal with abnormality detection. The main contribution of this thesis in the sense of texture classification is proposing a new texture classification algorithm by effectively employing external features within SVM kernel and introducing a new feature extraction method for texture classification.