Radiomics seeks to convert medical images into quantitative biomarkers capable of capturing tumor phenotype, microenvironment, and underlying biology. Over the past fifteen years, the field has expanded beyond handcrafted radiomic features toward deep radiomics, multi-site radiomics, and multi-omics integration, while the need for interpretability has become increasingly central. The aim of this article was to define and clarify these major methodological and conceptual evolutions, to summarize current innovations in deep, explainable, multi-site, and multi-omics radiomics, and to identify the remaining challenges that must be addressed before clinical translation. We first outline how deep learning architectures, including convolutional neural networks, autoencoders, vision transformers, and mask image modeling, enable the extraction of high-level, data-driven imaging representations that theoretically surpass the descriptive power of classical handcrafted radiomic features. Because tumors often display heterogeneous behavior across metastatic sites, we then describe the transition from single-site radiomics to patient-level multi-site approaches integrating all lesions, using aggregation methods, radiomic distance metrics, or attention-based multi-instance learning. Next, we highlight the efforts to harmonize imaging acquisition, preprocessing, and feature extraction across centers, and the growing role of multi-omics frameworks that integrate radiomics with genomic, transcriptomic, immunologic, and clinical data to provide a more complete picture of tumor biology. As model complexity increases, explainable artificial intelligence methods ( e.g ., class activation maps, permutation importance, and Shapley values), structured reporting frameworks, and intrinsically interpretable model architectures should be viewed as complementary rather than competing approaches to ensure transparency, interpretability, and clinical trust. Despite major progress, key challenges persist, including confounding factors, limited public datasets, multicenter variability, inconsistent reporting, and the absence of prospective radiomics-driven clinical trials. Ultimately, radiomics will reach clinical maturity only through the joint advancement of methodological rigor, harmonization, interpretability, and multi-omics integration.