A SVM-based Approach to Microwave Breast Cancer Detection

Early breast cancer detection is of crucial importance: this form of cancer is the second most common cause of death among women due to malignant tumors, whereas early detection leads to longest survival or even full recovery. Conventional X-ray mammography possesses a range of shortcomings and new techniques must be developed. Features of microwave breast imaging make it an attractive alternative. The aim of the present work is to propose a 3-D approach based on support vector machine classifier whose output is transformed to a posteriori probability of tumor presence. Like confocal microwave imaging introduced by S.C. Hagness et al., the present approach is aimed at detecting tumor locations directly, avoiding solving computationally extensive inverse scattering problem. Microwave data have been generated using finite element method with impedance boundary conditions. Noisy environments have been considered as well. The obtained probability maps demonstrate that the region around the tumor location usually clearly stands out against the background of overall probability values.