Information Fusion from Mammogram and Ultrasound Images for Better Classification of Breast Mass

Various medical modalities are used in all phases of cancer detection. Information extracted from these modalities reveals morphological, metabolic and functional information of tissues. Integrating this information in a meaningful way assists in clinical decision making. Sometimes using multimodal techniques supply complementary information for improved therapy planning. Proposed investigation is on classification of breast mass as benign or malignant for early detection of breast cancer using mammograms and Ultrasound modalities. The proposed approach is based on the fusion of information from two modalities at image feature level with different normalization techniques to improve the performance of breast mass classification. Gabor filters are used to retrieve texture features from mammograms, shape and structural features are retrieved from ultrasound images. Training of classifier is done using Support vector machine (SVM) classifiers to classify masses. Receiver operating characteristic curves (ROC) are used to evaluate the performance. Our method was validated on 20 set of images. Where each set consists of one mammogram and one ultrasound image of a same person out of which 9 sets were malignant and 11 were benign. SVM classifiers achieved 95.6% sensitivity in classifying the masses using the features retrieved from two modalities.