Breast tumor classification based on deep convolutional neural networks

This paper presents a novel deep learning approach focused on the classification of tumors in mammograms as malignant or benign. It is a modern machine learning method which promises to create models that learn from large dataset and make accurate predictions. In this study, we propose a discriminative objective for supervised feature learning by training a Convolutional Neural Network (CNN). Choosing CNN involves input image with a fixed-length and as a consequence, we equip our networks with a scaling process based on Gaussian pyramids for obtaining regions of interest with normalized size. The dataset used in this research is augmented with applying the geometric transformation techniques in order to prevent overfitting and create a robust deep learning model. We perform classification with Softmax layer. It is used to train CNN for classification. We evaluate our methodology on both of the publicly available dataset DDSM and BCDR. In comparison with the current state-of-the-art methods, the experiments show that our proposed system provides good results, achieving high accuracy of 97.28% that will assist radiologists in making diagnostic decisions without increasing false negatives.