Message Passing and Edge Attention in Heterogeneous Graph Neural Networks for Disease Diagnosis.

Saved in:

Bibliographic Details
Title:	Message Passing and Edge Attention in Heterogeneous Graph Neural Networks for Disease Diagnosis.
Authors:	Xiaodong Zhu¹ 18085879473@163.com, Dan Yang² asyangdan@163.com, Yang Liu³ liuyang_lnas@163.com
Source:	Engineering Letters. Oct2025, Vol. 33 Issue 10, p3946-3956. 11p.
Subjects:	Message passing (Computer science), Diagnosis, Graph neural networks, Electronic health records, Medical records
Abstract:	Message passing, as the core mechanism of Heterogeneous graph neural networks, can efficiently capture the latent relationships between nodes in the disease diagnosis task. However, the heterogeneity and complexity of medical data make it challenging for conventional message passing to accurately distinguish key information from noise, leading to semantic confusion, over-smoothing, and gradient vanishing issues. To address these challenges, we propose MPEA4DD, a heterogeneous graph neural network that integrates a custom message passing process combining dynamic edge attention with a state selection mechanism. We construct a medical heterogeneous graph from EMRs in MIMIC-III and MIMIC-IV, encode nodes and edges into a unified feature space, and apply our message-passing module, in which an edge attention network dynamically adjusts edge weights and a state-selection network assigns each node one of three interaction states (aggregation, dissemination, or integration) to mitigate the interference of irrelevant neighboring information. Furthermore, to mitigate gradient vanishing and excessive smoothing in deep architectures, we combine contextual semantic fusion with residual connections on GATv2 attention aggregation, achieving effective global information integration and stable gradient propagation in deep layers. Thus, evaluations on both MIMIC-III and MIMIC-IV demonstrate that MPEA4DD significantly outperforms other baseline models in disease diagnosis accuracy. [ABSTRACT FROM AUTHOR]
	Copyright of Engineering Letters is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Message passing, as the core mechanism of Heterogeneous graph neural networks, can efficiently capture the latent relationships between nodes in the disease diagnosis task. However, the heterogeneity and complexity of medical data make it challenging for conventional message passing to accurately distinguish key information from noise, leading to semantic confusion, over-smoothing, and gradient vanishing issues. To address these challenges, we propose MPEA4DD, a heterogeneous graph neural network that integrates a custom message passing process combining dynamic edge attention with a state selection mechanism. We construct a medical heterogeneous graph from EMRs in MIMIC-III and MIMIC-IV, encode nodes and edges into a unified feature space, and apply our message-passing module, in which an edge attention network dynamically adjusts edge weights and a state-selection network assigns each node one of three interaction states (aggregation, dissemination, or integration) to mitigate the interference of irrelevant neighboring information. Furthermore, to mitigate gradient vanishing and excessive smoothing in deep architectures, we combine contextual semantic fusion with residual connections on GATv2 attention aggregation, achieving effective global information integration and stable gradient propagation in deep layers. Thus, evaluations on both MIMIC-III and MIMIC-IV demonstrate that MPEA4DD significantly outperforms other baseline models in disease diagnosis accuracy. [ABSTRACT FROM AUTHOR]
ISSN:	1816093X