Fig. 1: Studying intra-tumor heterogeneity of hypoxia in DCIS using deep learning and digital pathology.
From: Spatial interplay of tissue hypoxia and T-cell regulation in ductal carcinoma in situ
a An illustrative example of a DCIS tumor with high spatial intra-tumor heterogeneity of hypoxia. Shown are images of IHC dual staining with CA9 and FOXP3, cells were classified into five types based on their expression of CA9 and FOXP3 and morphological features. b The deep learning pipeline using convolutional neural networks (CNNs) for single-cell analysis. c Generative adversarial networks (GANs) for semantic segmentation of individual DCIS ducts. d An example of DICS tumor where individual DCIS ducts have been segmented using GANs. Two high-resolution examples show ground truth obtained from annotations by pathologists and output from GANs. Scale bar represents 100 µm.
