Two and multiple categorization of breast pathological images by transfer learning
Jiaxin Yan, Bei Wang · 2021
Breast cancer is the most common cancer in women worldwide. By using artificial intelligence technology to assist doctors in pathological diagnosis can effectively improve the efficiency of cancer diagnosis. However, the computer-aided diagnosis (CAD) has the problems of long training time for large-resolution pathological pictures and insufficient data that can be marked for training. In this paper, a transfer learning model is developed for the pathological diagnosis of breast cancer to overcome those problems. Four common deep learning models (VGGnet, Resnet, Densenet, Mobilenet) were adopted to train breast pathology images under four different resolutions (40X, 100X, 200X, 400X). A transfer learning framework was established to distinguish benign and malignant breast pathology and their subtypes. The accuracy of the two-classification model can reach 91.24% at the best magnification (200X), and the average accuracy is 89.31%. At the same time, the multi-classification model for the eight subtypes of pathological sections also achieved quite satisfied results. It is indicated that the presented transfer learning framework has great potential for exploring the CAD of breast cancer.