Classification of breast mass in two-view mammograms via deep learning

Breast cancer is the second deadliest cancer among women. Mammography is an important method for physicians to diagnose breast cancer. The main purpose of this study is to use deep learning to automatically classify breast masses in mammograms into benign and malignant. This study proposes a two-view mammograms classification model consisting of convolutional neural network (CNN) and recurrent neural network (RNN), which is used to classify benign and malignant breast masses. The model is composed of two branch networks, and two modified ResNet are used to extract breast-mass features of mammograms from craniocaudal (CC) view and mediolateral oblique (MLO) view, respectively. In order to effectively utilise the spatial relationship of the two-view mammograms, gate recurrent unit (GRU) structures of RNN is used to fuse the features of the breast mass from the two-view. The digital database for screening mammography (DDSM) be used for training and testing our model. The experimental results show that the classification accuracy, recall and area under curve (AUC) of our method reach 0.947, 0.941 and 0.968, respectively. Compared with previous studies, our method has significantly improved the performance of benign and malignant classification.

Automated summary

This study utilizes deep learning to automatically classify breast masses in mammograms into benign and malignant. By combining convolutional neural network (CNN) and recurrent neural network (RNN), the model proposed in the study significantly improves the performance of benign and malignant breast mass classification.