Airway Segmentation Techniques in Medical Imaging

Airway segmentation in medical imaging is a cornerstone for diagnosis, treatment planning and interventional navigation in respiratory medicine. Traditionally, segmentation of the tracheobronchial tree relied on thresholding and region‐growing methods, which often required manual intervention to resolve under- or over-segmentation and to trace fine distal branches. Recent advances have shifted towards machine learning and deep learning, enabling fully automated extraction of the airway tree from volumetric computed tomography (CT) scans. Key challenges include the high variability in airway calibre and contrast, the presence of noise and artefacts in low-dose protocols, and the computational demands of processing large three-dimensional volumes. Emerging methods combine lightweight neural architectures with data-centric strategies, graph-based optimisation for wall delineation, attention mechanisms to capture multiscale features and centreline extraction to ensure topological accuracy. This progress has fostered more robust, generalisable and resource-efficient pipelines, bringing fully automatic airway segmentation within reach of routine clinical and research workflows worldwide.

Research from Nature Portfolio

Recent studies have demonstrated that a compact three-dimensional convolutional neural network with a U-shaped encoder–decoder backbone can process entire lungs in one pass and yield highly complete airway trees with few false positives. By validating across datasets that include paediatric patients with cystic fibrosis, subjects with chronic obstructive pulmonary disease and public benchmark collections, this approach has shown strong generalisation and sensitivity, ranking among the top performers in challenge evaluations. The low-memory design makes it amenable to standard GPU hardware and supports integration into clinical pipelines without extensive computational resources.

Airway Segmentation Techniques in Medical Imaging publication trend

The graph below shows the total number of articles in airway segmentation techniques in medical imaging across all publications each year (not limited to Nature Index journals).

Technical terms

Convolutional neural network (CNN): A class of deep learning model that applies convolutional filters to volumetric or image data to learn hierarchical feature representations.

U-Net architecture: A symmetric encoder–decoder network with skip connections that preserves spatial resolution and captures contextual information for precise segmentation.

Dice similarity coefficient (DSC): A statistical metric that quantifies the overlap between predicted and ground-truth segmentations, ranging from 0 (no overlap) to 1 (perfect overlap).

Optimal-surface graph-cut: An optimisation technique that segments structures by finding a minimum-cost surface in a graph representation, often used to delineate thin anatomical walls.

Centreline extraction: The computational process of tracing the central path through a tubular structure to ensure connectivity and to support quantitative morphometry.

References

1. Interpolation-split: a data-centric deep learning approach with big interpolated data to boost airway segmentation performance. Journal of Big Data (2024).
2. Automatic airway segmentation from computed tomography using robust and efficient 3-D convolutional neural networks. Scientific Reports (2021).
3. Segmentation of lung airways based on deep learning methods. IET Image Processing (2022).
4. Reproducibility of a combined artificial intelligence and optimal-surface graph-cut method to automate bronchial parameter extraction. European Radiology (2023).