Using Transfer Learning to Detect Breast Cancer without Network Training
Hao Pang, Wenjie Lin, Cong Wang, Chen Zhao · 2018
In recent years, the use of convolutional neural networks has made great success in the analysis of digital pathological images. However, due to the slower running speed of the model and the large amount of data in the single image, the model based on full sampling runs very slowly. It is of great significance to optimize the speed of the model. This paper proposes a method to complete the breast cancer detection by incomplete sampling of the features of the transfer learning output without network training. This method verified on Camelyon16 dataset. The experimental results show that while ensuring the accuracy of the model, it can greatly reduce the time for model construction and use.