Fig. 3

Radiomics and computational pathology in tumor immunity exploration. Radiological and pathological image-derived omics data enable the investigation of the tumor immune microenvironment (TIME) and response to immunotherapy. For a raw radiology image, regions of interest, generally representing the tumor lesion area, are segmented, while a pathological image is divided into numerous sub-images. Two methods can be applied to analyze these high-dimensional data. First, features, including but not limited to statistical features, tumor volume features, and texture features, are extracted and analyzed by professional clinicians. Alternatively, images are input into a convolutional neural network (CNN). After a complicated deep learning process, robust models are output. These radiomics or digital pathologic models could finally be established to evaluate or predict the immune index, which can be divided into three aspects. First, TIME dissection encompasses distinct immune cell subset classification and TIME spatial architecture characterization, resembling single-cell technologies and CODEX, respectively. Second, immune-related biomarkers, such as the tumor mutation burden (TMB), could be predicted. Third, response to immunotherapy and clinical outcomes could be predicted. ROI regions of interest, TIME tumor immune microenvironment, TMB tumor mutation burden