Microcalcifications detection in digital mammograms

A method is presented for detecting minimally sized microcalcifications on ma mammograms to add extra security to the radiologist's classification. The method imitates the normal procedure followed by the specialist, and is easily implemented on low-cost PCs. As input, it accepts the usual digital mammograms. Tested against one of the most extensive databases - the DDSM of the University of South Florida - it gave a 100% success rate. For any suspicious regions (the so-called regions-of-interest or ROI) a separate image of suitable size is generated and displayed. The system also allows feature vectors to be generated for use in an automatic classifying system - such as a neural network (NN) - to determine the malignancy of the ROIs that were detected.