A New Ensemble Learning Approach for Microcalcification Clusters Detection

Abstract

A new microcalcification clusters (MCs) detection method in mammograms is proposed, which is based on a new ensemble learning method. In this paper, , we propose a bagging with adaptive cost adjustment ensemble algorithm; and a new ensemble strategy, called boosting with relevance feedback, by embedding the relevance feedback technique into the heterogenous base learner training, and meanwhile carefully design an effectively systematical feedback scheme, which promise the preventing of overfitting. The ground truth of MCs is assumed to be known as a priori. In our algorithm, each MCs is enhanced by a well designed high-pass filter. Then the 116 dimentional image features are extracted by the feature extractor and fed to the ensemble decision model. In image feature domain, the MCs detection procedure is formulated as a supervised learning and classification problem, and the trained ensemble model is used as a classifier to decide the presence of MCs or not. Case study on microcalcification clusters detection for breast cancer diagnosis illustrates that the proposed algorithm is not only effective but also efficient.