Abstract: A new approach to microcalcification clusters detection is proposed, based on active learning. The proposed algorithm first enhances the microcalcification region with a directional difference filter bank which effectively realizes the feature extraction and meanwhile suppresses the blood vessels and mammary duts. Then the active sample selecting method based on Bootstrap is employed to select the training set and train the Baysian classifier. Finally the obtained classifier can be used to detect microcalcification clusters in the mammogram. Experimental results show that the proposed algorithm achieves good performance. Compared with the traditional passive learning methods, the new algorithm reduce the false positive rate 4.7％ by keeping the same sensitivity.

Key words: directional difference filter bank, active learning, classifiers, microcalcification clusters, feature extraction