MedAide: Information Fusion and Anatomy of Medical Intents via LLM-based Agent Collaboration

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yang, Dingkang, Wei, Jinjie, Li, Mingcheng, Liu, Jiyao, Liu, Lihao, Hu, Ming, He, Junjun, Ju, Yakun, Zhou, Wei, Liu, Yang, Zhang, Lihua
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913923281715200
author Yang, Dingkang
Wei, Jinjie
Li, Mingcheng
Liu, Jiyao
Liu, Lihao
Hu, Ming
He, Junjun
Ju, Yakun
Zhou, Wei
Liu, Yang
Zhang, Lihua
author_facet Yang, Dingkang
Wei, Jinjie
Li, Mingcheng
Liu, Jiyao
Liu, Lihao
Hu, Ming
He, Junjun
Ju, Yakun
Zhou, Wei
Liu, Yang
Zhang, Lihua
contents In healthcare intelligence, the ability to fuse heterogeneous, multi-intent information from diverse clinical sources is fundamental to building reliable decision-making systems. Large Language Model (LLM)-driven information interaction systems currently showing potential promise in the healthcare domain. Nevertheless, they often suffer from information redundancy and coupling when dealing with complex medical intents, leading to severe hallucinations and performance bottlenecks. To this end, we propose MedAide, an LLM-based medical multi-agent collaboration framework designed to enable intent-aware information fusion and coordinated reasoning across specialized healthcare domains. Specifically, we introduce a regularization-guided module that combines syntactic constraints with retrieval augmented generation to decompose complex queries into structured representations, facilitating fine-grained clinical information fusion and intent resolution. Additionally, a dynamic intent prototype matching module is proposed to utilize dynamic prototype representation with a semantic similarity matching mechanism to achieve adaptive recognition and updating of the agent's intent in multi-round healthcare dialogues. Ultimately, we design a rotation agent collaboration mechanism that introduces dynamic role rotation and decision-level information fusion across specialized medical agents. Extensive experiments are conducted on four medical benchmarks with composite intents. Experimental results from automated metrics and expert doctor evaluations show that MedAide outperforms current LLMs and improves their medical proficiency and strategic reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2410_12532
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MedAide: Information Fusion and Anatomy of Medical Intents via LLM-based Agent Collaboration
Yang, Dingkang
Wei, Jinjie
Li, Mingcheng
Liu, Jiyao
Liu, Lihao
Hu, Ming
He, Junjun
Ju, Yakun
Zhou, Wei
Liu, Yang
Zhang, Lihua
Computation and Language
In healthcare intelligence, the ability to fuse heterogeneous, multi-intent information from diverse clinical sources is fundamental to building reliable decision-making systems. Large Language Model (LLM)-driven information interaction systems currently showing potential promise in the healthcare domain. Nevertheless, they often suffer from information redundancy and coupling when dealing with complex medical intents, leading to severe hallucinations and performance bottlenecks. To this end, we propose MedAide, an LLM-based medical multi-agent collaboration framework designed to enable intent-aware information fusion and coordinated reasoning across specialized healthcare domains. Specifically, we introduce a regularization-guided module that combines syntactic constraints with retrieval augmented generation to decompose complex queries into structured representations, facilitating fine-grained clinical information fusion and intent resolution. Additionally, a dynamic intent prototype matching module is proposed to utilize dynamic prototype representation with a semantic similarity matching mechanism to achieve adaptive recognition and updating of the agent's intent in multi-round healthcare dialogues. Ultimately, we design a rotation agent collaboration mechanism that introduces dynamic role rotation and decision-level information fusion across specialized medical agents. Extensive experiments are conducted on four medical benchmarks with composite intents. Experimental results from automated metrics and expert doctor evaluations show that MedAide outperforms current LLMs and improves their medical proficiency and strategic reasoning.
title MedAide: Information Fusion and Anatomy of Medical Intents via LLM-based Agent Collaboration
topic Computation and Language
url https://arxiv.org/abs/2410.12532