Key points are not available for this paper at this time.
Artificial intelligence (AI) is rapidly reshaping clinical oncology, as cancer care increasingly relies on integrating heterogeneous data streams spanning radiology, digital pathology, genomics, and longitudinal electronic health records. However, the sheer complexity and fragmentation of these multimodal inputs remain a major bottleneck for achieving truly personalized cancer management. Recent advances in AI, including foundation models, synthetic data generation, large language models, and agents, are enabling more robust representation learning, cross-modal reasoning, and clinically actionable decision support beyond what traditional single-modality systems can provide. AI-powered platforms are now accelerating molecular subtyping, refining risk stratification, and supporting individualized therapeutic recommendations by jointly modeling imaging, tissue architecture, and molecular landscapes. Moreover, emerging virtual cell and mechanistic foundation frameworks introduce a new computational paradigm for simulating cellular responses and drug-tumor interactions, offering predictive insights for treatment design and drug discovery. Despite these breakthroughs, critical challenges persist, including limited generalizability across patient populations and centers, insufficient prospective validation, regulatory uncertainty, scalability constraints, and ethical concerns surrounding fairness, transparency, and privacy. In this review, we synthesize the latest progress in multimodal oncology AI through a translational lens, emphasizing methodological trade-offs, validation readiness, and responsible deployment frameworks. We highlight how AI is moving from performance-driven benchmarking toward clinically trustworthy precision cancer care, with transformative implications for early detection, diagnosis, therapy optimization, drug development, and clinical trial design.
Lin et al. (Wed,) studied this question.