Computer aided detection of microcalcifications in digital mammograms adopting a wavelet decomposition

Over recent years breast cancer prevention campaigns have resulted in widespread screening. Until now, mammography has been one of the most reliable methods for the early detection of this disease. The high correlation between the appearance of microcalcification clusters and the presence of cancer shows that Computer Aided Detection systems of microcalcifications are extremely useful and helpful in an early detection of breast cancer. Several techniques can be adopted to accomplish this task. In this paper an efficient tool for a fully automatic microcalcification cluster detection/localization is presented. Adopting this procedure, all suspect microcalcifications are preserved and background noise is reduced by thresholding mammograms through a wavelet filter, according to image statistical parameters (i.e. mean gray level pixel value and standard deviation). Moreover, in order to localize singularity points, the reconstructed image is decomposed adopting another wavelet and each decomposition level is processed using a hard threshold technique. To reduce false positive detections in microcalcification recognition, the results obtained in each level are combined with a suitable procedure. The Mammographic Image Analysis Society database is used to test the procedure. The performance obtained highlights the validity of the method; indeed, the evaluated sensitivity parameter (true positive rate) is about 98% at an average rate of 1 false positive per image.