Dynamic Graph Neural Network for Belief Fusion in Multi-Agent Collaborative Diagnosis
Keywords:
Multi-Agent Systems, Graph Neural Networks (GNNs), Belief Fusion, Clinical Decision Support, Dynamic Graph LearningAbstract
Multi-agent diagnostic systems often rely on static voting or rule-based mechanisms to resolve conflicts among specialist agents, yet these approaches fail to capture the dynamic and context-dependent relationships that arise in clinical reasoning. We propose a Dynamic Graph Neural Network for Belief Fusion (DGNN-BF) that reformulates conflict resolution as a learnable, time-varying graph process. In our framework, each specialist agent corresponds to a node in a graph whose edges are constructed on-the-fly from patient-specific clinical features and the agents’ evolving belief states. An adaptive adjacency matrix is generated through a multi-scale observation alignment mechanism, which computes pairwise dependencies and applies top-k thresholding to ensure sparsity. To handle the asynchronous nature of agent outputs, we introduce a temporal alignment gate that exponentially decays messages from stale beliefs, thereby preventing outdated information from distorting the consensus. A spatial-temporal graph attention layer then performs attention-weighted message passing, dynamically weighting each agent’s contribution based on both its current reliability and temporal relevance. Domain-specific priors on disease progression pathways are incorporated as a structure-informed regularization loss, guiding the learned graph to align with established medical knowledge while still permitting patient-specific deviations. After multiple layers of propagation, the final node representations are concatenated and decoded into a fused belief vector, which replaces conventional aggregation methods in the diagnosis synthesis engine. The entire module is trained end-to-end using cross-entropy loss jointly with the structure-informed loss. Our approach offers three key contributions: a dynamic graph construction that adapts to each patient case, a temporal asynchrony handling mechanism that respects real-world computational delays, and a principled integration of domain priors into the learning process. This work is significant because it transforms conflict resolution from a static, hand-crafted procedure into a data-driven, context-aware process that can improve diagnostic accuracy in complex, multi-agent clinical environments.
References
J Spieser, A Balapour, J Meller, KC Patra, et al. A review of multi-agent ai systems for biological and clinical data analysis. Methods and Protocols, 2026.
S Young and L Xu. Xrayclaw: Cooperative-competitive multi-agent alignment for trustworthy chest x-ray diagnosis. Technical report, arXiv preprint arXiv:2604.02695, 2026.
Z Wang, Y Zhu, H Zhao, X Zheng, D Sui, et al. Colacare: Enhancing electronic health record modeling through large language model-driven multi-agent collaboration. In Proceedings of the ACM on Web Conference 2025, 2025.
A Hunter and M Thimm. Probabilistic reasoning with abstract argumentation frameworks. Journal of Artificial Intelligence Research, 2017.
Z Diao, X Wang, D Zhang, Y Liu, K Xie, and S He. Dynamic spatial-temporal graph convolutional neural networks for traffic forecasting. In Proceedings of the AAAI Conference on Artificial Intelligence, 2019.
S Yan, Y Xiong, and D Lin. Spatial temporal graph convolutional networks for skeleton-based action recognition. In Proceedings of the AAAI Conference on Artificial Intelligence, 2018.
T Yang, L Hu, C Shi, H Ji, X Li, and L Nie. Hgat: Heterogeneous graph attention networks for semi-supervised short text classification. ACM Transactions on Information Systems, 2021.
Y Li and C Tan. A survey of the consensus for multi-agent systems. Systems Science & Control Engineering, 2019.
P Jiang, CD Xiao, A Cross, and J Sun. Graphcare: Enhancing healthcare predictions with personalized knowledge graphs. In International Conference on Learning Representations, 2024.
K Cao, R Li, M Stražar, EM Brown, PNU Nguyen, et al. A structure-informed deep learning framework for modeling tcr-peptide-hla interactions. Bioinformatics, 2026.
S Han, U Topcu, and GJ Pappas. Differentially private distributed constrained optimization. IEEE Transactions on Automatic Control, 2016.
X Zhang, X Nie, Y Liu, and Z Wang. Collaborative medical intelligence: A state-machine-based multi-agent architecture for multimodal diagnostic reasoning. Mathematics, 2026.
Y Feng, J Wang, L Zhou, Z Lei, et al. Doctoragent-rl: A multi-agent collaborative reinforcement learning system for multi-turn clinical dialogue. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2026.
P Xia, J Wang, Y Peng, K Zeng, Z Dong, X Wu, et al. Mmedagent-rl: Optimizing multi-agent collaboration for multimodal medical reasoning. Technical report, arXiv preprint arXiv:2506.00555, 2025.
P Zhan, B Cai, J Zhang, H Lv, Y Tan, et al. Care-agent: Multi-agent collaboration with conflict-aware routing mechanism for diagnosis prediction. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2026.
Y Chen, X Zhou, X Qiao, Y Lian, et al. Kgdagents: A knowledge graph enhanced multi-agent framework for clinical diagnosis. In 2026 29th International Conference on Computer Supported Cooperative Work in Design (CSCWD), 2026.
F Zhang, K Zhao, and X Liu. A hierarchical multi-agent framework for improved clinical diagnosis. 2025 11th International Conference on Computer and Communications (ICCC), 2025.
W Yu, G Ma, H Lv, Z Lin, X Chen, et al. Dmap: A deep-model driven multi-agent framework for reliable diagnosis prediction. In 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2025.
S Lan, Y Ma, W Huang, W Wang, et al. Dstagnn: Dynamic spatial-temporal aware graph neural network for traffic flow forecasting. In Proceedings of the VLDB Endowment, 2022.
T Zhao, M Yue, and J Wang. Structure-informed graph learning of networked dependencies for online prediction of power system transient dynamics. IEEE Transactions on Power Systems, 2022.
M Zhang, S Wu, X Yu, Q Liu, et al. Dynamic graph neural networks for sequential recommendation. IEEE Transactions on Knowledge and Data Engineering, 2022.
A Yekeen. Using graph neural networks to fuse heterogeneous object detections across sensors for multi-agent awareness. Available at SSRN 5436521, 2025.
X Sun. Chimera: A multi-agent graph-temporal framework for interpretable medical insurance fraud detection. In Proceedings of, 2026.
A Raza, S Maqsood, et al. Hybrid transformer-graph neural network framework with reinforcement learning for explainable multi-agent decision making in dynamic environments. Research Consortium Archive, 2026.
Alejandro Barredo Arrieta, Natalia Díaz-Rodríguez, Javier Del Ser, Adrien Bennetot, Siham Tabik, Alberto Barbado, Salvador Garcia, Sergio Gil-Lopez, Daniel Molina, Richard Benjamins, Raja Chatila, and Francisco Herrera. Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai. Information Fusion, 58:82–115, 2019.
L Xin, Y Zheng, K Cheng, C Jiang, Z Zhang, et al. Hytrec: A hybrid temporal-aware attention architecture for long behavior sequential recommendation. Technical report, arXiv preprint arXiv:2602.18283, 2026.
P Veličković, G Cucurull, A Casanova, et al. Graph attention networks. In International Conference on Learning Representations, 2018.
C Huang, C Hong, T Xu, D Zhao, Z Wu, et al. Risk factors for pulmonary embolism in icu patients: a retrospective cohort study from the mimic-iii database. Clinical and Applied Thrombosis/Hemostasis, 2022.
M Shafiq and Z Gu. Deep residual learning for image recognition: A survey. Applied sciences, 2022.
J Devlin, MW Chang, K Lee, et al. Bert: Pre-training of deep bidirectional transformers for language understanding. North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2019.
J Irvin, P Rajpurkar, M Ko, Y Yu, S Ciurea-Ilcus, et al. Chexpert: a large chest radiograph dataset with uncertainty labels and expert comparison. arxiv. Technical report, arXiv preprint arXiv ..., 2019, 2019.
S Ardabili, A Mosavi, and AR Várkonyi-Kóczy. Advances in machine learning modeling reviewing hybrid and ensemble methods. In International Conference on Global Research and Education, 2019.
JA Hoeting, D Madigan, AE Raftery, et al. Bayesian model averaging: a tutorial (with comments by m. clyde, david draper and ei george, and a rejoinder by the authors. Statistical Science, 1999.
M Azad, TH Nehal, and M Moshkov. A novel ensemble learning method using majority based voting of multiple selective decision trees. Technical report, repository.kaust.edu.sa, 2024.
C Grimsley, E Mayfield, and JRS Bursten. Why attention is not explanation: Surgical intervention and causal reasoning about neural models. In International Conference on Language Resources and Evaluation, 2020.
F Cheng, Y Ming, and H Qu. Dece: Decision explorer with counterfactual explanations for machine learning models. IEEE Transactions on Visualization and Computer Graphics, 2020.
B Munyampirwa, V Lakshman, and B Coleman. Down with the hierarchy: The ’h’in hnsw stands for "hubs". Lecture Notes in Computer Science, 2026.
R Child, S Gray, A Radford, and I Sutskever. Generating long sequences with sparse transformers. preprint. Technical report, arxiv, 1904, 1904.
C Xu, C Siebenbrunner, and L Bindschaedler. Grappa: Gradient-only communication for scalable graph neural network training. Technical report, arXiv preprint arXiv:2602.01872, 2026.
T Chen, B Xu, C Zhang, and C Guestrin. Training deep nets with sublinear memory cost. preprint at. 2016, 2016.
L Ding, W Fei, Y Huang, S Ding, W Dai, C Li, J Zou, et al. Ampa: Adaptive mixed precision allocation for low-bit integer training. ICML, 2024.
B Jacob, S Kligys, B Chen, M Zhu, et al. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018.
G Hinton, O Vinyals, and J Dean. Distilling the knowledge in a neural network. Technical report, arXiv preprint arXiv:1503.02531, 2015.
C Boufaied, T Nguyen, and R de Souza Santos. Practitioner insights on fairness requirements in the ai development life cycle: An interview study. ACM Transactions on Software Engineering and Methodology, 2025.
M Islam, Z Li, and B Glocker. Robustness stress testing in medical image classification. Lecture Notes in Computer Science, 2023.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 The Author(s)
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
