An SVM Based Approach for the Analysis Of Mammography Images
Xiangchao Gan, L. Kapsokalivas, A. Skaliotis, Kathleen K. Steinhöfel, Sabina Tangaro, Theodore E. Simos, George Psihoyios, Ch. Tsitouras · AIP conference proceedings · 2007
Mammography is among the most popular imaging techniques used in the diagnosis of breast cancer. Nevertheless distinguishing between healthy and ill images is hard even for an experienced radiologist, because a single image usually includes several regions of interest (ROIs). The hardness of this classification problem along with the substantial amount of data, gathered from patients' medical history, motivates the use of a machine learning approach as part of a CAD (Computer Aided Detection) tool, aiming to assist radiologists in the characterization of mammography images. Specifically, our approach involves: i) the ROI extraction, ii) the Feature Vector extraction, iii) the Support Vector Machine (SVM) classification of ROIs and iv) the characterization of the whole image. We evaluate the performance of our approach in terms of the SVM's training and testing error and in terms of ROI specificity—sensitivity. The results show a relation between the number of features used and the SVM's performance.