Large language models driven reliable clinical decision-making: Framework and application

Abstract

With the proliferation of data and increased complexity of clinical decision-making in the medical field, powerful computational tools are needed to assist physicians in making precise and reliable decisions. While the Large Language Models (LLMs) with billions of parameters in model size have obtained a series of achievements in a broad range of biomedical and healthcare applications, the issues in terms of reliability and stability are still needed to be addressed. To this end, we propose the framework of MedRad, a system that combines LLMs, knowledge engineering, Chain of Thought (CoT) reasoning, Retrieval-Augmented Generation (RAG) techniques, and intelligent agents (Agents) to improve clinical decision-making reliability. Based on fine-tuned LLMs and existing studies in the biomedical and healthcare domain, we further concentrate on how these techniques could be utilized to achieve highly reliable clinical decision-making in scenarios with varying complexity, such as medical knowledge QA and clinical diagnosis recommendations. Experimental results demonstrate that MedRad has the ability to provide high-quality decision paths in the above scenarios, and the potential to extend to more biomedical and healthcare scenarios through its loosely coupled design. • MedRad combines LLMs, knowledge engineering, CoT reasoning, RAG and Agent to improve clinical decision-making reliability. • The combination of GPT-4 model and MedRad has the potential to achieve the best performance in different clinical scenarios. • MedRad integrates advanced technology into clinical practice, promoting self-diagnosis and combining AI with human expertise.

Ähnliche Arbeiten