Multi-Agent LLM Framework for Preliminary Disease Diagnosis

Abstract

Effective healthcare delivery depends on early and precise disease diagnosis. Single-agent and conventional Clinical Decision Support Systems (CDSS) Limited thinking abilities, explainability issues, and hallucinogenic medical responses are common problems with Large Language Model (LLM) techniques. This research suggests a RAG-Enhanced Multi-Agent LLM Framework for Preliminary Disease Diagnosis and Clinical Decision Support to overcome these drawbacks. The suggested framework is made up of several intelligent agents that work together to evaluate patient symptoms and produce diagnostic recommendations. These agents include a symptom analysis agent, specialized diagnosis agents, a consensus agent, and a verification agent. In order to improve the dependability and factual consistency of the generated responses, a Retrieval-Augmented Generation (RAG) module is integrated to retrieve pertinent medical knowledge from reliable healthcare resources. Explainable reasoning is further integrated into the framework by offering evidence-based explanations for every diagnostic recommendation. The accuracy, precision, recall, and F1-score of diagnostic performance are evaluated experimentally using benchmark medical datasets. In comparison to traditional single-agent LLM systems, the results show that the suggested multi-agent design improves diagnostic consistency, transparency, and decision-making quality. While preserving the crucial role of medical professionals in final clinical judgment, the suggested framework shows great promise as an intelligent clinical decision support tool for preliminary disease identification.