Cancer metastasis fast location based on coarse-to-fine network

The Whole Slide Images(WSIs) play a very important role in breast cancer diagnosis and the pathologist need to locate the lymph node metastasis on the such the gigapixel pathology image. Recently, the deep convolutional neural network has show the promise of metastases localization, it is still a challenge to fast locate the metastases. Either divide WSIs into small patches and perform classification or scan the bigger image block on WSIs in inference. In either way, the algorithm needs to be performed at the finest magnification, which greatly limits inference time. In this paper, we propose a cascade coarse to fine network to expedite the speed of to locate metastases in WSIs, which contain the coarse network to handle the low magnification to find the rough metastases speedily and the fine network efficiently reclassifies the positive responses at high magnification. The experiment is performed on the Camelyon16 dataset demonstrated that the proposed method compared to the previous method is the fastest and also can achieve the localization average FROC score of 81.0 on the test set.