Abstract
Training a deep convolutional neural network from scratch requires massive amount of data and significant computational power. However, to collect a large amount of data in medical field is costly and difficult, but this can be solved by some clever tricks such as mirroring, rotating and fine tuning pre-trained neural networks. In this paper, we fine tune a deep convolutional neural network (ALEXNET) by changing and inserting input layer convolutional layers and fully connected layer. Experimental results show that our method achieves a patch and image-wise accuracy of 75.73% and 81.25% respectively on the validation set and image-wise accuracy of 57% on the ICIAR-2018 breast cancer challenge hidden test set.