we proposes a dual input semantic segmentation network for WSI based on attention
the paper's address :https://arxiv.org/abs/1907.06358
Abstract—Due to the high resolution of pathological images, the automated semantic segmentation in the medical pathological images has shown greater challenges than that in natural images. Sliding Window method has shown its effect on solving problem caused by the high resolution of whole slide images (WSI). However, owing to its localization, Sliding Window method also suffers from lack of global information. In this paper, a dual input semantic segmentation network based on attention is proposed, in which, one input provides small-scale fine information, the other input provides large-scale coarse information. Compared with single input methods, our method based on dual inputs and attention: DA-RefineNet exhibits a dramatic performance
The original dataset is from ICIAR2018 challenge. The dataset is composed of Hematoxylin and eosin (H&E), stained breast histology microscopy, and whole-slide images. we only use the whole slide image. This part contains 10 whole slide images. Whole-slide images are high resolution images containing the entire sampled tissue. In this sense, microscopy images just served as details of the whole-slide images. Because of that, each whole-slide image could have multiple normal, benign, in situ carcinoma and invasive carcinoma regions.
The whole slide image is very big, so we need to use sliding windows. Firstly, we use the sliding window to cut the whole slide image into 50005000 pixel images which provide large-scale coarse contextual information. Secondly, we use the sliding window with a size of 10001000 and a step size of 500 pixels to cut the 50005000 pixel images imto 10001000 pixel images which provide small-scale fine localited information. Thirdly, we resize both images to 500*500 pixels. Finally, we use data screening to balance the four types of data to a certain extent.
You can find the example of image with 50005000 pixels in folder data/splited_svs, its corresponding mask can be found in folder data/splited_xml. You can find the example of image with 10001000 pixels in folder data/splited_svs_little, its corresponding "RGB" mask and "P" mask can be found in folder data/splited_xml_little and data/splited_xml_little_P respectively. You can find the example of image with 500500 pixels resized by image with 50005000 pixels in folder data/splited_svs_resize. The final lists of train and test images can be found in folder data/train.txt and data/test.txt.
We use this dual input segmentation network based on attention for our work. Other details will be displayed in the paper.
We use the SGD optimizer to train our model. Since imagenet pre-training exists in the encoding stage, we set different hyperparameters for the encoder and decoder. The encoder parameter suffix is ENC and the decoder suffix is DEC. At the same time, we divide the training into three steps, and each step have 25 epochs. The initial learning rate of each step are LRENC = [5e-4; 2.5e-4; 1e-4];LRDEC = [5e-3; 2.5e-3; 1e-3], the Momentum is set to 0.9 and WD is 1e-5. Batch size is 12.All code is written by Pytorch. And we use four GTX1080Ti for our training.
You can use three resnet for this assignment by changing ICIAR.sh
We first thank the data provider, the organizer of the ICIAR2018 Grand Challenge. Secondly, we would like to thank the university of Adelaide and Australian Centre for Robotic Vision (ACRV) for the code of Refinenet, the project address: https://github.com/DrSleep/refinenet-pytorch.