GraphDx: A Cost-Aware Knowledge-Enhanced Multi-Agent Framework for Sequential Diagnosis
Quick Answer
GraphDx is a novel framework that enhances sequential diagnosis by integrating LLMs to create Medical Diagnosis Knowledge Graphs (MDKGs).
Quick Take
It improves diagnostic success rates from 50-68% to 79-93% while reducing testing costs by 20-54%, demonstrating a cost-effective solution for automated clinical diagnosis.
Key Points
- GraphDx utilizes to create Medical Diagnosis Knowledge Graphs (MDKGs).
- Success rates improved from 50-68% to 79-93% across three LLM models.
- Testing costs were reduced by 20-54% in clinical diagnosis.
- The framework includes three collaborative agents: Perception, Reasoning, and Decision.
- It addresses the knowledge-reasoning gap in existing LLM approaches.
DeepSignal Analysis
What happened
GraphDx is a framework designed to enhance sequential medical diagnosis by integrating Large Language Models (LLMs) to create Medical Diagnosis Knowledge Graphs (MDKGs). It reportedly improves diagnostic success rates from 50-68% to 79-93% while reducing testing costs by 20-54%. The framework includes three collaborative agents focused on perception, reasoning, and decision-making.
Key evidence
- GraphDx improves diagnostic success rates from 50-68% to 79-93% based on experiments conducted on MedQA and MIMIC-IV datasets.
- The framework reduces testing costs by 20-54%, indicating a significant cost-saving potential in automated clinical diagnosis.
- GraphDx utilizes three collaborative agents: Perception, Reasoning, and Decision, to enhance the diagnostic process through systematic reasoning and cost-aware planning.
Why it matters
The integration of LLMs into medical diagnosis through GraphDx addresses a critical gap in existing models, which often fail to balance accuracy with cost. By improving diagnostic success rates and reducing costs, this framework could lead to more efficient healthcare delivery. The emphasis on cost-awareness is particularly relevant in today's healthcare landscape, where resource allocation is increasingly scrutinized.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Sequential diagnosis requires balancing diagnostic accuracy against resource costs through iterative information gathering. Existing Large Language Model (LLM) approaches exhibit a critical knowledge-reasoning gap: despite encoding extensive medical knowledge, they struggle to reason systematically under cost constraints, often resorting to excessive testing. We propose GraphDx, a knowledge-enhanced framework with two core innovations. First, we design an automated pipeline that leverages LLMs to construct Medical Diagnosis Knowledge Graphs (MDKGs) with quantized typicality, action-centric topology, and dual-objective attributes for both diagnostic relevance and cost-sensitivity. Second, we introduce three collaborative agents (Perception, Reasoning, and Decision) where the Perception and Decision Agents handle language understanding and generation, while the Reasoning Agent performs deterministic evidence scoring and cost-aware planning on the MDKG. Experiments on MedQA and MIMIC-IV across three LLM backbones (DeepSeek-V3, Kimi-k2, Llama-3.3) show that GraphDx improves diagnostic success rates from 50--68% to 79--93% while reducing test costs by 20--54%, providing a robust, economical, and interpretable solution for automated clinical diagnosis.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.15280 [cs.AI] |
| (or arXiv:2607.15280v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.15280 arXiv-issued DOI via DataCite |
Submission history
From: Shaoting Tan [view email]
[v1]
Wed, 8 Apr 2026 05:47:40 UTC (2,282 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.