Breast tumor classification in ultrasound images using deep convolutional neural network

Abstract

Recently, deep learning has shown great success in many computer vision applications. The ability to learn image features and use these features for object localization, classification and segmentation has paved the way for new medical image studies, improving the performance of automated computer-aided detection (CADe) systems. In this paper, a new approach is proposed for the classification of breast tumors in ultrasound (US) images, based on convolutional neural networks (CNN). The database consists of 641 images, histopathologically classified in two categories (413 benign and 228 malignant lesions). To have a better estimate of the model’s classification performance, the data were split to perform 5-fold cross-validation. For each fold, 80% of data was used for training, and 20% for the evaluation. Different evaluation metrics were used as performance measurements. With the proposed network architecture, we achieved an overall accuracy of 85.98% for tumor classification and the area under the ROC curve (AUC) equal to 0.94. After applying image augmentation and regularization, the accuracy and the AUC increased to 92.05% and 0.97, respectively. The obtained results surpassed other machine learning methods based on manual feature selection, demonstrating the effectiveness of the proposed method for the classification of tumors in US imaging.