Numerical validation and experimental results of a multi-resolution SVM-based classification procedure for breast imaging

X-ray mammography is the principal technique for breast cancer screening in clinical practice. However, X-ray mammography presents different drawbacks showing that alternative technologies are desirable. For example, the difficulty in detecting the breast tumors at the earlier stage, the destructive effects of the ionizing X-rays on the irradiated tissues, low efficiency in dense breasts and with cancer near the chest wall. Other minor but still negative aspects are also the discomfort due to the breast compression and the expensive costs. In this paper, the inversion process is recast as a classification process by integrating a SVM-based classifier with an iterative multi-zooming procedure (IMSA) for early detection of breast cancer. Once the training procedure of the system is completed, the detection of the pathology is real-time estimated by generating a multi-resolution risk-map of malignant tissue presence. The spatial resolution of the probability risk-map is iteratively enhanced at each step of the IMSA methodology only in the Regions of Interest (Rols) where the pathology is supposed to be located. The effectiveness of the approach has been numerically validated considering some test cases reproducing the geometry and the characteristics of a prototypal imaging system used to evaluate the applicability of the proposed methodology also when applied to experimental measurements.