Radiomics: Current Applications and Future Directions

Radiomics enables high-throughput extraction of quantitative imaging features to decode tumor phenotypes and biological behaviors, representing a transformative noninvasive tool for precision oncology. In recent years, radiomics has rapidly evolved from static feature analysis to dynamic multi-dimensional assessment, and it has been widely explored in various solid tumors, yet its pan-cancer generalization, biological interpretability, and clinical translation still face prominent bottlenecks. Cancer remains the leading cause of global mortality, and solid tumors account for more than 90% of adult malignant cases, while conventional medical imaging and invasive biopsies have inherent limitations in reflecting tumor heterogeneity and dynamic evolution. This review outlines the unified technical pipeline of radiomics across solid tumors, highlights cancer-specific imaging considerations, and summarizes standardization strategies for multi-center, multi-scanner, and multi-cancer heterogeneity. We systematically review pan-cancer clinical applications covering early detection, molecular characterization, treatment response prediction, and prognostic stratification, with lung cancer as a paradigmatic example while integrating evidence from breast, colorectal, liver, glioma, and prostate cancers. We also discuss multi-omics integration, biological interpretability, and translational bottlenecks including domain shift and reproducibility crisis. Finally, we prospect cutting-edge directions including foundation models, causal inference, and federated learning to advance generalizable and clinically actionable radiomics toward routine clinical practice.