Precision oncology: from large language models to multi-agent systems

Area of Science:

• Artificial Intelligence in Oncology
• Precision Medicine
• Clinical Decision Support Systems

Background:

• Growing volumes of electronic health records, medical imaging, and omics data necessitate advanced computational approaches in precision oncology.
• Large language models (LLMs) and multimodal AI show promise for integrating complex health data and supporting clinical decisions.
• Current single-model AI approaches face limitations in clinical reasoning, traceability, and workflow integration.

Purpose of the Study:

• To review the applications of LLMs and multimodal AI across the precision oncology continuum, from screening to documentation.
• To explore the emerging role of AI agents and multi-agent systems (MAS) in addressing limitations of single-model approaches.
• To propose a task-architecture alignment framework for selecting appropriate AI systems in precision oncology.

Main Methods:

• Structured narrative review of current AI technologies in precision oncology.
• Analysis of LLM and multimodal AI applications in cancer screening, diagnosis, staging, treatment recommendation, and documentation.
• Exploration of AI agents and MAS for advanced clinical decision support and workflow integration.

Main Results:

• LLMs and multimodal AI are increasingly applied across the precision oncology care pathway, demonstrating significant potential.
• Single-model AI systems exhibit limitations in complex clinical reasoning and integration.
• AI agents and MAS represent a promising direction for more sophisticated and integrated precision oncology solutions.

Conclusions:

• A task-architecture alignment framework is proposed to guide the selection of foundation models, single-agent, and multi-agent systems based on clinical task complexity and risk.
• This framework aims to facilitate the design, evaluation, and clinical translation of AI systems in precision oncology.
• The judicious application of AI, guided by task-specific needs, is crucial for advancing precision oncology.