Reflecting Topology Consistency and Abnormality via Learnable Attentions for Airway Labeling

TL;DR

The paper tackles the challenge of automatic airway labeling under substantial anatomical variability and disease-induced deformities by introducing topology-aware mechanisms. It presents a two-stage transformer-based framework augmented with Soft Subtree Consistency (SSC) to encode subtree-level topology and Abnormal Branch Saliency (ABS) to detect anomalies, using a soft subtree map $M_t$, descendant mask $M_D$, and anomaly mask $\tilde{M}_a$ to guide feature aggregation. Key contributions include the SSC module that soft-clusters features within anatomically meaningful subtrees and the ABS module that learns prototypes and anomaly scores to separate normal from abnormal branches, with quantitative gains in subtree consistency, abnormal-branch recall, and overall labeling accuracy on the ATL-fibrosis dataset. The approach yields robust labeling performance across deformities, offering practical benefits for preoperative planning and intraoperative navigation in bronchoscopy. Overall, the method advances airway labeling by explicitly modeling topological relations and abnormalities, potentially improving accuracy and safety in clinical workflows.

Abstract

Accurate airway anatomical labeling is crucial for clinicians to identify and navigate complex bronchial structures during bronchoscopy. Automatic airway anatomical labeling is challenging due to significant individual variability and anatomical variations. Previous methods are prone to generate inconsistent predictions, which is harmful for preoperative planning and intraoperative navigation. This paper aims to address these challenges by proposing a novel method that enhances topological consistency and improves the detection of abnormal airway branches. We propose a novel approach incorporating two modules: the Soft Subtree Consistency (SSC) and the Abnormal Branch Saliency (ABS). The SSC module constructs a soft subtree to capture clinically relevant topological relationships, allowing for flexible feature aggregation within and across subtrees. The ABS module facilitates the interaction between node features and prototypes to distinguish abnormal branches, preventing the erroneous aggregation of features between normal and abnormal nodes. Evaluated on a challenging dataset characterized by severe airway distortion and atrophy, our method achieves superior performance compared to state-of-the-art approaches. Specifically, it attains a 91.4% accuracy at the segmental level and an 83.7% accuracy at the subsegmental level, representing a 1.4% increase in subsegmental accuracy and a 3.1% increase in topological consistency. Notably, the method demonstrates reliable performance in cases with disease-induced airway deformities, ensuring consistent and accurate labeling.

Figures (6)

• Figure 1: Overview of bronchial anatomical labeling and its challenges. (a) and (b) illustrate the preoperative planning path for bronchoscopy guided by hierarchical anatomical labeling. (c) and (d) show successful and failed navigation trajectories, respectively. (e) compares node-wise correctness and subtree-wise consistency under correct labeling ($L_C$) and incorrect labeling ($L_I$). The results indicate that anatomical subtree-wise consistency is more sensitive to labeling errors. (f) and (g) demonstrate how individual variability and abnormal branches pose challenges to labeling accuracy and consistency
• Figure 2: Overview of the proposed method: The airway labeling method comprises a Soft Subtree Consistency (SSC) module, which generates a soft subtree probability map, and an Abnormal Branch Saliency (ABS) module, which produces an abnormality score prediction. Both of them are used to inform attention interaction at the subsegmental level
• Figure 3: (a) illustrates the typical pattern of the left upper lobe, featuring a long lingular branch originating from the left main bronchus and bifurcating into LB4 and LB5. (b) shows a variant pattern where LB4 and LB5 directly bifurcates from the left main bronchus. (c) demonstrates a variant airway at the segmental level. (d) describes LB3 and LB4 at the subsegmental level. (e) shows the soft subtree map of the Airway. (f) details the Segment Consistency Mask
• Figure 4: Qualitative results at subsegmental label for deformed airways. Ground truth of branches are in black, while false predictions of branches are in red. TNN fails to generate anatomically meaningful subtrees, while AirwayFormer predicts incorrect labels. Incorporating the SSC module, our method demonstrates improved accuracy and consistency
• Figure 5: Qualitative results at subsegmental label for abnormal branches. Competing methods misclassify abnormal branch by treating it as a separate category. Incorporating the ABS module, our method achieves superior performace