Deep Transfer Learning-Based Intelligent Diagnosis of Malignant Tumors on Mammography

Abstract

In this paper, we propose a deep transfer learning-based intelligent diagnosis approach for malignant tumors on mammography. An image segmentation algorithm is developed to remove the background, noise, and other redundancy in the image, for improving the learning efficiency. Based on the GoogleNet after training, we apply the transfer learning technique to the processed image. In this way, the accuracy of the classification model is improved. The experiment results show that the accuracy of our image segmentation algorithm is 100%, using only one-third of the data in training; the accuracy of our training approach is with the highest and average accuracy of 83% and 70%, respectively, by 2 × 104 iterations; and the area under the receiver operating characteristic curve is 0.77. These results are superior to those obtained by the existing methods.