[Paper Review] Structure and position-aware graph neural network for airway labeling

Weiyi Xie, Colin Jacobs+2 more|arXiv (Cornell University)|Jan 12, 2022

Chronic Obstructive Pulmonary Disease (COPD) Research4 citations

TL;DR

This paper proposes a Structure and Position-aware Graph Neural Network (SPGNN) for automated airway labeling in CT scans, combining CNN-extracted features with a GNN that encodes both topological structure and positional relationships using anchor-based distance encoding. The method achieves 91.18% average accuracy on 18 segmental airway branches, significantly outperforming baseline CNNs (83.83%) and structure-only GNNs.

ABSTRACT

We present a novel graph-based approach for labeling the anatomical branches of a given airway tree segmentation. The proposed method formulates airway labeling as a branch classification problem in the airway tree graph, where branch features are extracted using convolutional neural networks (CNN) and enriched using graph neural networks. Our graph neural network is structure-aware by having each node aggregate information from its local neighbors and position-aware by encoding node positions in the graph. We evaluated the proposed method on 220 airway trees from subjects with various severity stages of Chronic Obstructive Pulmonary Disease (COPD). The results demonstrate that our approach is computationally efficient and significantly improves branch classification performance than the baseline method. The overall average accuracy of our method reaches 91.18\% for labeling all 18 segmental airway branches, compared to 83.83\% obtained by the standard CNN method. We published our source code at https://github.com/DIAGNijmegen/spgnn. The proposed algorithm is also publicly available at https://grand-challenge.org/algorithms/airway-anatomical-labeling/.

Motivation & Objective

• To address the challenge of accurate anatomical labeling of airway trees in CT scans, especially in the presence of high anatomical variability and segmentation noise.
• To improve upon existing graph-based methods that are only structure-aware and fail to distinguish symmetric nodes with identical neighborhoods.
• To integrate positional information into graph neural networks to better capture topological relationships such as sibling branches and hierarchical ordering.
• To develop a computationally efficient, end-to-end trainable method that generalizes well across diverse COPD severity stages.
• To provide a publicly available, reusable solution for airway labeling applicable to broader medical image analysis tasks.

Proposed method

• The method formulates airway labeling as a node classification problem on a graph where each node represents a branch in the segmented airway tree.
• Convolutional Neural Networks (CNNs) extract visual features from 3D airway segments, which are then used as initial node features in the graph.
• A novel Graph Neural Network (GNN) architecture is proposed that is both structure-aware and position-aware, using topological distance from predefined anchor branches to encode positional information.
• The GNN aggregates messages from neighboring nodes using graph attention mechanisms, enabling adaptive weighting of incoming features based on structural and positional relevance.
• Skip connections are incorporated into the GNN to mitigate over-smoothing in deep architectures, improving training stability and performance.
• The final node representations are used to predict anatomical labels for each airway branch using a classification head.

Experimental results

Research questions

• RQ1 Can integrating positional encoding into a GNN improve node classification performance in airway trees with symmetric structures?
• RQ2 How does the proposed SPGNN compare to standard CNNs and structure-only GNNs in terms of labeling accuracy across diverse COPD phenotypes?
• RQ3 To what extent does incorporating skip connections in the GNN architecture mitigate over-smoothing and improve performance in deep graph networks?
• RQ4 Does the use of anchor-based positional encoding enhance the model's ability to distinguish sibling branches and hierarchical relationships?
• RQ5 Can the proposed method generalize across varying airway tree topologies and segmentation artifacts?

Key findings

• The SPGNN achieved an overall average accuracy of 91.18% on 18 segmental airway branches, significantly outperforming the baseline CNN method, which achieved 83.83%.
• The SPGNN demonstrated superior performance in correctly identifying sibling relationships, such as correctly labeling LB3 and LB5 as siblings, whereas GATS and CNN methods failed to do so.
• The method reduced misclassification of branches like RB1 and LB10, which were often confused due to anatomical variations and convolutional feature ambiguity.
• Graph Attention Networks (GATs) outperformed other GNN architectures on this dataset, and skip connections helped stabilize performance in deeper models.
• The SPGNN achieved state-of-the-art performance with minimal computational overhead, making it suitable for clinical deployment.
• The model generalizes well across different COPD severity stages, as evidenced by consistent performance across training and test splits.