Study on the Generation of Tumor Individualized Treatment Schemes Based on Large Language Models and Literature Retrieval

Abstract

To address the challenge of formulating individualized treatment schemes caused by tumor heterogeneity, this study adopts a method combining large language models and retrieval-augmented generation (RAG) technology to construct a "clinical data-literature knowledge-model decision-making" collaborative framework, and designs and implements a tumor individualized treatment scheme generation system. The system generates personalized schemes conforming to clinical guidelines through multimodal patient data preprocessing, structured medical literature database construction, model medical fine-tuning, and retrieval-augmented adaptation. Experimental verification shows that the average accuracy score of the system's scheme generation is 85.6, the RAG technology increases the citation rate of the latest literature in the scheme by 42.3%, and the average score of clinical physician evaluation is 8.7. The research indicates that integrating large language models and literature retrieval technology can effectively improve the accuracy and evidence-based nature of tumor individualized treatment schemes, providing strong support for clinical decision-making.