A Classification System of Mammograms Based on Convolutional Neural Networks

This paper presents an approach to improve a classification system based on convolutional neural networks for classifying breast cancer X-ray images into three classes of Normal, Benign, and Malignant by increasing the training set. For training convolutional neural network models, we use 15,040 Vietnamese mammograms, which were collected and annotated by the radiologists of the Vietnam National Cancer Hospital. Our experiment was conducted with the Resnet 34. The evaluation of system performance using the testing set achieves a macAUC of 0.84807, an average sensitivity of 0.69058, and an average specificity of 0.84336 which are higher than a macAUC of 0.828247, an average sensitivity of 0.64738, and an average specificity of 0.825670 when we trained the model Resnet 18 with 8,777 mammograms. For the classification of breast cancer X-ray images into malignant class, our network achieves an AUC of 0.89758. On the other hand, our proposed model achieves results which is higher than the results of a classification system applied to the model ResNet 50 for classifying breast cancer X-ray images into BI-RADS 045, BI-RADS 1, BI-RADS 23 categories which achieves a macAUC of 0.754, an average sensitivity of 0.58, and an average specificity of 0.79 when using 7,912 X-Ray images collected by Vietnam Hanoi Medical University Hospital. We observe that the classification performance improves as the number of mammograms increases. With the achieved result, our classification system can help radiologists in Vietnam in the interpretation of breast cancer X-ray images.