Microcalcification clusters in mammograms may be an early sign of breast cancer. Therefore, their forward detection is of primary importance for treatment effectiveness and, consequently, for reduction of breast cancer
mortality. Unfortunately, mammogram analysis is a complex and hard task which requires specialized radiologists. In fact, the intrinsic difficulty in cancer sign detection makes the image study particularly tiring, especially in mass screening situations. The difficulty in microcalcification detection is due both to their small size and to low contrast between microcalcifications and surrounding tissues. The high correlation between microcalcification cluster presence and disease appearance shows that computer-aided detection (CAD) systems for automated microcalcification detection are very useful and helpful for breast cancer control. In this paper, a new method for computer-aided detection and localization of microcalcifications which combines wavelet transform and neural network is introduced. The implemented procedure is fully automatic, without any necessity of manual selection of image region of interest or parameter calibration. For background noise removal, microcalcification enhancement and recognition of true
microcalcifications, two different wavelet filters are adopted according to image statistical parameters. To minimize false microcalcification identifications, a special procedure is applied to the image representations which take into account singularity points. These suspected zones provide inputs for the neural network which is able to localize microcalcification clusters. Sensitivity and specificity parameters are considered for the procedure of performance
evaluation. Adopting the Mammographic Image Analysis Society (MIAS) database for procedure testing, the proposed method is able to achieve high performance in microcalcification cluster detection and localization, as is shown through the plotted free-response operating characteristic (FROC) curve.