 The study investigates whether deep convolutional neural networks, CNNs, can extract prognosticators directly from hematoxilineacin, he, stained tissue slides of colorectal cancer, CRC, patients. The authors hand-alineated single tissue regions in 86 CRC tissue slides and used these to train a CNN by transfer learning, achieving a nine-class accuracy of over 94% in an independent data set. They then performed automated tissue decomposition on representative multi-tissue. He images from 862 he slides in 500 stage 1 to 4 CRC patients into the cancer genome atlas TCGA cohort and calculated a deep stroma score based on the output neuron activations in the CNN. This score was an independent prognostic factor for overall survival cancer associated fibroblasts CAFs and relapse free survival in both the TCGA and Darmcrabbs, chancinderverhutungdursh screening, DACHS, cohorts. The study suggests that CNNs can assess the human tumor microenvironment and predict prognosis directly from histopathological images, which could potentially be implemented in clinical workflows with prospective validation. We are article.tv, links in the description below.