Generative Artificial Intelligence and Large Language Models in Clinical Oncology.

Area of Science:

• Oncology
• Artificial Intelligence
• Biomedical Data Science

Background:

• Cancer is a significant global health issue, with growing multimodal data offering potential for precision oncology.
• Generative AI, especially large language models (LLMs), can integrate diverse data like EHRs, imaging, genomics, and clinical text.
• Current research on generative AI in oncology is fragmented, lacking a comprehensive overview of its potential across the entire patient journey.

Purpose of the Study:

• To provide a comprehensive review of generative AI applications in clinical oncology.
• To synthesize methodological foundations and applications of various generative AI models (LLMs, GANs, diffusion, multimodal foundation models).
• To discuss challenges and future directions for implementing safe and trustworthy AI in oncology.

Main Methods:

• Review of recent advancements in generative AI, including LLMs, generative adversarial networks (GANs), diffusion models, and multimodal foundation models.
• Analysis of applications in cancer diagnosis, prognosis, treatment planning, patient management, and clinical trial optimization.
• Focus on multimodal data integration, synthetic data generation, clinical reasoning, and decision support.

Main Results:

• Generative AI models show promise in integrating heterogeneous data for enhanced cancer care.
• Applications span the entire oncology continuum, from diagnosis to patient management and clinical trials.
• Key challenges include interpretability, reliability, data privacy, regulatory hurdles, and real-world implementation.

Conclusions:

• Generative AI offers transformative potential for precision oncology by integrating complex data.
• Addressing challenges in trust, privacy, and regulation is crucial for clinical adoption.
• Future directions include agent-based architectures and human-AI collaboration for advanced oncology systems.