Histopathological Tissue Segmentation of Lung Cancer with Bilinear CNN and Soft Attention

Figure 2

An overview of the proposed multitissue segmentation algorithm. Stage 1: a classification network is pretrained on the colorectal cancer dataset, and transfer learning is used to train the classification network with the training set of the lung cancer dataset. The independent image dataset is used to evaluate the classification accuracy of the network. Stage 2: H&E-WSI image (20x magnification) is segmented through stitching the classification results tile by tile. H&E: hematoxylin and eosin; WSI: whole-slide image; TUM: tumour epithelium; LYM: tumour-infiltrating lymphocytes; STR: stroma; NOR: normal; VES: vessel; BRO: bronchus; NEC: necrosis; APC: areas polluted by carbon dust; BAC: background; OTH: others.