Improving the Diagnosis of Breast Cancer by Combining Visual and Semantic Feature Descriptors

Abstract

AbstractMammography still remains the foremost effective procedure for early diagnosis of breast cancer. To compensate with the increasing number of readings that radiologists have to undertake, Computer-aided Diagnosis (CAD) systems have become a significant assistance tool, that are used to help identify abnormal and normal regions of interest in mammograms faster and more effectively than human readers. In this paper, we propose a new approach for breast cancer identification of all type of lesions in digital mammograms by combining low- and high-level mammogram descriptors in a compact quaternionic form. Our proposed method has two major stages. Initially, a feature extraction process is proposed to extract low-level visual descriptors of Region of Suspicion (ROS), that utilizes two dimensional discrete transforms based on Angular Radial Transform (ART), Shapelets as well as textural representations based on the Haralick features. To further improve our method’s performance, the semantic information of the mammogram given by the radiologists is encoded in a 16-bit length word high-level feature vector. The features are stored in a quaternion and fused using the L2 norm prior to their presentation to the classification module. For the classification task, each ROS is recognized using a trained radial basis function neural network. The proposed method is evaluated on regions taken from the DDSM database. The proposed feature set achieves a mean accuracy of 94.22% (±1.594) and AUC = 0.9578 that is 33% higher than using only visual descriptors, showing that semantic information can improve the diagnosis when combined with standard visual features.KeywordsShapeletsARTHaralickQuaternionRBF-NNBreast cancerComputer-Aided Diagnosis (CAD)Semantic information