Skrede OJ, De Raedt S, Kleppe A, et al. - This study was conducted to generate a biomarker of patient outcome after primary colorectal cancer resection by directly examining scanned conventional haematoxylin and eosin-stained sections using deep learning. Researchers used more than 12,000,000 image tiles from individuals with a distinctly good or poor disease result from four cohorts to train a total of ten convolutional neural networks, purpose-built for classifying supersized heterogeneous images. From four cohort, 828 individuals had a distinct outcome and were accepted as a training cohort to obtain clear ground truth and 1,645 individuals were a non-distinct outcome and were practised for tuning. The study developed a clinically useful prognostic marker using deep learning allied to digital scanning of conventional haematoxylin and eosin-stained tumour tissue sections. The assay has been extensively estimated in large, independent individual populations, associates with and outperforms stabilised molecular and morphological prognostic markers, and provides consistent outcomes across tumour and nodal stage. The biomarker stratified stage II and III individuals into adequately distinct prognostic groups that potentially could be applied to guide the selection of adjuvant treatment by avoiding therapy in very low-risk groups and distinguishing individuals who would help from more intensive treatment regimes.