Predicting Pneumonia Progression in Children Using Spatio-Temporal Graph Neural Networks (GNN-PulmNet)

Abstract

Pneumonia remains a leading cause of pediatric hospitalization worldwide, with 
Mycoplasma Pneumoniae Pneumonia (MPP) being a significant concern. Traditional deep learning 
models such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) 
networks have shown effectiveness in pneumonia detection but lack the ability to capture both spatial 
lung-region dependencies and temporal progression trends. In this paper, we propose GNN-PulmNet, a 
novel Spatio-Temporal Graph Neural Network (ST-GCN) framework that models pneumonia 
progression over time. Our model represents the lung as a graph, where nodes correspond to lung 
regions and edges represent anatomical connections. GNN-PulmNet leverages graph convolutional 
networks (GCN) and temporal modeling to predict the evolution of pneumonia severity, outperforming 
existing models in diagnostic accuracy and progression tracking.