A generalized multiple classifier system for improving computer-aided classification of breast masses in mammography

PurposeThe objective of this paper is to present a generalized multiple classifier system for improved classification of mammographic masses in Computer-aided detection (CAD).MethodsTo encourage different base (component) classifiers to learn different parts of an object instant space, we develop a novel base classifier generation algorithm which combines data resampling underpinning AdaBoost with the use of different feature representations. In addition, our proposed multiple classifier system can be generalized beyond the limitation of weak classifiers in conventional AdaBoost learning. To this end, our multiple classifier system has an effective and efficient mechanism for tuning the level of weakness of base classifiers.ResultsExtensive experiments have been performed using benchmark mammogram data set to test the proposed method on classification between mammographic masses and normal tissues. In addition, to assess classification performance, we used the area under the receiver operating characteristic (AUC) and the normalized partial area under the curve (pAUC). Results show that our method considerably outperforms (in terms of both AUC and pAUC) the most commonly used single neural network (NN) and support vector machine (SVM) based classification approaches. In particular, the effectiveness of our method in terms of correct classification is much more significant over difficult mammogram cases with dense tissues that have higher risk of cancer incidences and cause higher false-positive (FP) detections.ConclusionsOur multiple classifier system shows quite promising results in terms of improving classification performances on the FP reduction application using classification between masses and normal tissues in mammography CAD systems.