[A Three‒Tiered AI Framework Toward Precision Oncology-Structuring, Reclassification, and Integration].

Abstract

Artificial intelligence(AI)is transforming cancer medicine across three key domains. First, AI enables the conversion of unstructured visual data-such as pathology slides and radiological images-into structured, quantifiable formats. This improves diagnostic reproducibility and allows for automated tumor detection, classification, and prognostication with accuracy comparable to or exceeding that of specialists. Second, AI facilitates high‒dimensional analysis of omics data, including RNA sequencing and DNA methylation profiles. Machine learning models can uncover latent molecular patterns, predict splicing abnormalities, and identify dependency genes, enabling more refined molecular classification and novel therapeutic target identification. Third, multimodal AI integrates heterogeneous data types-images, genomics, and clinical text-into unified analytical frameworks. This allows for non‒invasive prediction of molecular alterations, treatment responsiveness assessment, and outcome prediction. Integration with large language models(LLMs)further enhances interpretability and enables cross‒modal reasoning in clinical decision‒making contexts. Together, these 3 layers of AI application-image structuring, omics analysis, and multimodal integration-form the foundation for a next‒generation approach to cancer care. AI does not merely automate existing tasks but offers new pathways to understanding cancer's origins, molecular essence, and progression trajectories‒bringing the vision of true precision oncology closer to clinical reality.