Classification of Benign and Malignant Breast Mass in Digital Mammograms with Convolutional Neural Networks
Xin Tong Zhao, Xianheng Wang, Hongkai Wang · 2018
Breast cancer is one of the most common cancers affecting women lives worldwide. It is usually quite difficult for radiologists to accurately distinguish between malignant and benign tumor in digital mammograms. An intelligent classifier based on conventional machine learning algorithms can help radiologists classifying abnormal breast mass and diagnosing breast cancer. Recently, deep learning has attracted much research attention in medical image analysis due to its higher classifying accuracy and the capability of learning features from annotated imaging data automatically. Therefore, we proposed a deep neural network model to classify benign and malignant tumors in digital mammograms. Our model is an improved version of the AlexNet, which is a Convolutional Neural Networks (CNN) model of deep learning. Totally 115 regions of interest (ROIs) were extracted from Mammographic Images Analysis Society (MIAS) database and finally augmented to 4600 images used as the training and testing dataset. In order to compare our proposed model with conventional learning models, an SVM-based classifier was implemented based on the same dataset. Experimental results showed that our model has more significant classification capability with the accuracy of 97.57%, while the SVM-based model has only 86.08% accuracy.