Unmasking the immune microecology of ductal carcinoma in situ with deep learning

Abstract Despite increasing evidence supporting the clinical relevance of tumour infiltrating lymphocytes (TILs) in invasive breast cancer, TIL spatial variability within ductal carcinoma situ (DCIS) samples and its association with progression are not well understood. To characterise tissue architecture microenvironment DCIS, we designed validated a new deep learning pipeline, UNMaSk. Following automated detection individual DCIS ducts using method IM-Net, applied tessellation to create virtual boundaries for each duct. study local infiltration duct, DRDIN was developed mapping distribution TILs. In dataset comprising grade 2–3 pure adjacent cancer (adjacent DCIS), found that cases had more TILs compared DCIS. However, colocalisation significantly lower which may suggest inflamed ecology cases. Our demonstrates technological developments convolutional neural networks digital pathology can enable an morphological microenvironmental analysis providing way differential immune identify markers progression.