Using Transfer Learning With Convolutional Neural Networks To Diagnose Breast Cancer From Histopathological Images

Diagnosis from histopathological images is the gold standard in diagnosing breast cancer. This paper investigates using transfer learning with convolutional neural networks to automatically diagnose breast cancer from patches of histopathological images. We compare the performance of using transfer learning with an off-the-shelf deep convolutional neural network architecture, VGGNet, and a shallower custom architecture. Our proposed final ensemble model, which contains three custom convolutional neural network classifiers trained using transfer learning, achieves a significantly higher image classification accuracy on the large public benchmark dataset than the current best results, for all image resolution levels.