A New Approach for the Diagnosis of Mammographic Masses Based on BI-RADS Features and Opposition-Based Classification
Subject Areas : مهندسی برق و کامپیوترF. Saki 1 , A. Tahmasbi 2 , Shahriar Baradaran Shokouhi 3
1 - University of Science and Technology
2 - University of Science and Technology
3 -
Keywords: BI-RADS CADx system feature extraction mammography images opposition-based classifier,
Abstract :
Fast and accurate classification of benign and malignant patterns in digital mammograms is of significant importance in the diagnosis of breast cancers. In this paper, we develop a new Computer-aided Diagnosis (CADx) system using a novel Opposition-based classifier to enhance the accuracy and shorten the training time of the classification of breast masses. We extract a group of Breast Imaging-Reporting and Data System (BI-RADS) features from preprocessed mammography images and feed them to a Multi-Layer Perceptron (MLP). The MLP is then trained using a new learning rule which we will refer to as the Opposite Weighted Back Propagation (OWBP) algorithm. We evaluate the performance of the system, in terms of classification accuracy, using a Receiver Operational Characteristics (ROC) curve. The proposed system yields an area under ROC curve (Az) of 0.924 and an accuracy of 92.86 %. Furthermore, the speed analysis results suggest that, with the same network topology, the convergence rate of the proposed OWBP algorithm is almost 4 times faster than that of the traditional Back Propagation (BP) algorithm.