Fig. 1: Study design.

a Digital haematoxylin and eosin (H&E)-stained slides of clear-cell renal cell carcinoma (ccRCC) patients were collected from The Cancer Genome Atlas (TCGA). Six texture subtypes as well as lymphocytes were detected and quantified with convolutional neural networks. Imaging data were integrated with clinical, genomic, transcriptomic and transcriptome-based immune profiling data. b Flow chart of the patient number included in this study. c Image tile examples annotated by texture subtypes and d lymphocytes. e Classification accuracy of the computational texture mapping (CTM) and lymphocyte classifiers. f Heatmap of median proportion of texture subtypes by participating TCGA clinical site. g Cancer tissue proportion by participating TCGA clinical site. The box plots indicate the interquartile ranges and median values.