Automated airway segmentation by learning graphical structure

TL;DR

This paper introduces a novel airway segmentation method combining CNN and GNN to leverage graph structures, significantly improving accuracy over traditional CNN-only approaches in chest CT scan analysis.

Contribution

The paper presents an innovative model integrating graph neural networks with CNNs for airway segmentation, enhancing performance by considering neighborhood graph structures.

Findings

Improved airway detection in chest CT scans.

CNN-GNN model outperforms CNN-only methods.

Model applicable to various datasets.

Abstract

In this research project, we put forward an advanced method for airway segmentation based on the existent convolutional neural network (CNN) and graph neural network (GNN). The method is originated from the vessel segmentation, but we ameliorate it and enable the novel model to perform better for datasets from computed tomography (CT) scans. Current methods for airway segmentation are considering the regular grid only. No matter what the detailed model is, including the 3-dimensional CNN or 2-dimensional CNN in three directions, the overall graph structures are not taken into consideration. In our model, with the neighbourhoods of airway taken into account, the graph structure is incorporated and the segmentation of airways are improved compared with the traditional CNN methods. We perform experiments on the chest CT scans, where the ground truth segmentation labels are produced manually. The proposed model shows that compared with the CNN-only method, the combination of CNN and GNN has a better performance in that the bronchi in the chest CT scans can be detected in most cases. In addition, the model we propose has a wide extension since the architecture is also utilitarian in fulfilling similar aims in other datasets. Hence, the state-of-the-art model is of great significance and highly applicable in our daily lives. Keywords: Airway segmentation, Convolutional neural network, Graph neural network