Methods for clustered microcalcifications detection in digital mammograms

This paper presents the comparison of three methods for clustered microcalcifications (MCCs) detection, which are associated with a high probability of malignancy. The proposed methods are morphological approach, fractal analysis and high-order statistics (HOS) test. We apply these methods on two sets of digital mammograms (MIAS database and McGill University database) to test their efficiency, accuracy and reliability in MCCs detection. Statistical analysis using receiver operating characteristics (ROC) curves and mean areas under the ROC curves are used in evaluating the performance of detection methods. It shows that the HOS test proved to be the most efficient and accurate and give reliable results for every mammogram tested.