Computer-aided Diagnosis of Breast Cancer by Hybrid Fusion of Ultrasound and Mammogram Features

Ultrasound images are increasingly being used as an important adjunct to X-ray mammograms for diagnosis of breast cancer. In this paper, a computer-aided diagnosis system that utilizes a hybrid fusion strategy based on canonical correlation analysis (CCA) is proposed for discriminating benign and malignant masses. The system combines information from three different sources, i.e., ultrasound and two views of mammogram, namely, mediolateral oblique (MLO) and craniocaudal (CC) views. CCA is employed on ultrasound-MLO and ultrasound-CC feature pairs to explore the hidden correlations between ultrasound and mammographic view. The two pairs of canonical variates are fused at the feature level and given as input to support vector machine (SVM) classifiers. Finally, decisions of the two classifiers are fused. Results show that the proposed system outperforms unimodal systems and state-of-the-art fusion strategies.