Next-Generation Radiomics: Foundation AI Models for HighDimensional Imaging Biomarker Discovery

Abstract

Radiomics has emerged as a transformative discipline in precision oncology by enabling the extraction of highdimensional quantitative features from medical imaging that reflect tumor phenotype, biological heterogeneity, and therapeutic response. Traditional radiomics relies on handcrafted imaging features and conventional machine learning algorithms, which often exhibit limited reproducibility and generalizability across imaging platforms and patient populations. Recent advances in foundation artificial intelligence (AI) models have fundamentally redefined radiomics through self-supervised learning, transformer architectures, multimodal representation learning, graph neural networks, and generative AI capable of learning generalized imaging representations from massive collections of radiological data. Integration of radiological imaging with digital pathology, genomics, transcriptomics, proteomics, metabolomics, laboratory biomarkers, electronic health records, and longitudinal clinical outcomes has expanded radiomics into comprehensive multimodal computational ecosystems supporting precision diagnosis, biomarker discovery, therapeutic prediction, adaptive treatment planning, digital twins, and intelligent clinical decision support. Emerging technologies including large language models, retrieval-augmented generation, federated learning, reinforcement learning, explainable artificial intelligence, and agentic AI further enhance the translational potential of imaging-based computational oncology. Despite remarkable advances, significant challenges remain regarding imaging standardization, data harmonization, computational scalability, interpretability, interoperability, regulatory validation, and ethical implementation. This review provides a comprehensive overview of nextgeneration radiomics, emphasizing foundation AI models for high-dimensional imaging biomarker discovery, clinical applications, implementation challenges, and future perspectives for precision oncology.[1]