Physics Based Contrast Marking and Inpainting Based Local Texture Comparison for Clustered Microcalcification Detection

As important early signs of breast cancers, microcalcifications (MCs) are still very difficult to be reliably detected by either radiologists or computer-aided diagnosis systems. In general, global, regional, and local properties of the mammogram should all be considered in the analysis process. In our effort, we incorporate the physical nature of the imaging process with the image analysis techniques to detect the clustered microcalcifications based on local contrast marking and self-repaired texture comparison. Suspicious areas are first obtained from a simplified X-ray imaging model where the MC contrast is a nonlinear function of local intensity. Following a removal and repair (R&R) procedure of the suspicious areas from their surrounding background textures, pre- and post- R&R local characteristic features of these areas are extracted and compared. A modified AdaBoost algorithm is then used to train the classifier for detecting individual microcalcification, followed by a clustering process to obtain the clustered MCs. Experiments on the MIAS database have shown promising results.