ABSTRACT
Radiomics and radiogenomics have demonstrated enormous potential for precision medicine and brought diagnostic, prognostic, and predictive value to a variety of clinical applications. In this chapter, we first review some emerging paradigms and challenges in this field. To address regional variations in genotypes and phenotypes of tumors, new radiomics approaches analyzing image features and interactions among intratumoral subregions are being developed for better characterization of the spatial heterogeneity. In radiogenomics, research is moving beyond simple association between imaging and genomic data and starting to leverage the complementary values of different types of data to further improve prediction. We then highlight some potential avenues for future investigations in radiomics and radiogenomics toward the ultimate goal of precision medicine. Issues important to meet the unmet demands of radiomics and radiogenomics, such as data collection, data sharing, quantitative imaging, optimization of imaging protocol, and adaptation of new imaging techniques, are discussed. Technically, as an alternative to feature based prediction models, deep neural networks are emerging for substantially improved prediction by seamlessly integration of optimal feature extraction and unified prediction into an end-to-end mapping. Applications of deep convolutional neural network techniques in the context of radiomics, particularly in classification, image segmentation and image reconstruction, are illustrated with some specific examples. With the support of the exciting deep learning technologies, significant performance improvements can be expected in radiomics and radiogenomics.
