Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RepAir: A Framework for Airway Segmentation and Discontinuity Correction in CT

John M. Oyer, Ali Namvar, Benjamin A. Hoff, Wassim W. Labaki, Ella A. Kazerooni, Charles R. Hatt, Fernando J. Martinez, MeiLan K. Han, Craig J. Galbán, Sundaresh Ram

TL;DR

RepAir addresses the challenge of incomplete airway connectivity in automated CT airway segmentation by coupling a robust 3D nnU-Net segmentation with a topology-aware reconstruction pipeline. It introduces a skeleton-based discontinuity detector and a 1D CNN classifier to validate and reconnect potential airway links, ensuring a single, anatomically consistent airway tree without sacrificing voxel-level accuracy. Evaluations on ATM'22 and AeroPath demonstrate state-of-the-art performance across voxel-level and topology metrics, outperforming Bronchinet and NaviAirway even in diseased lungs. The framework enhances reliability of airway-derived biomarkers and has potential to support longitudinal lung disease monitoring and quantitative analyses.

Abstract

Accurate airway segmentation from chest computed tomography (CT) scans is essential for quantitative lung analysis, yet manual annotation is impractical and many automated U-Net-based methods yield disconnected components that hinder reliable biomarker extraction. We present RepAir, a three-stage framework for robust 3D airway segmentation that combines an nnU-Net-based network with anatomically informed topology correction. The segmentation network produces an initial airway mask, after which a skeleton-based algorithm identifies potential discontinuities and proposes reconnections. A 1D convolutional classifier then determines which candidate links correspond to true anatomical branches versus false or obstructed paths. We evaluate RepAir on two distinct datasets: ATM'22, comprising annotated CT scans from predominantly healthy subjects and AeroPath, encompassing annotated scans with severe airway pathology. Across both datasets, RepAir outperforms existing 3D U-Net-based approaches such as Bronchinet and NaviAirway on both voxel-level and topological metrics, and produces more complete and anatomically consistent airway trees while maintaining high segmentation accuracy.

RepAir: A Framework for Airway Segmentation and Discontinuity Correction in CT

TL;DR

RepAir addresses the challenge of incomplete airway connectivity in automated CT airway segmentation by coupling a robust 3D nnU-Net segmentation with a topology-aware reconstruction pipeline. It introduces a skeleton-based discontinuity detector and a 1D CNN classifier to validate and reconnect potential airway links, ensuring a single, anatomically consistent airway tree without sacrificing voxel-level accuracy. Evaluations on ATM'22 and AeroPath demonstrate state-of-the-art performance across voxel-level and topology metrics, outperforming Bronchinet and NaviAirway even in diseased lungs. The framework enhances reliability of airway-derived biomarkers and has potential to support longitudinal lung disease monitoring and quantitative analyses.

Abstract

Accurate airway segmentation from chest computed tomography (CT) scans is essential for quantitative lung analysis, yet manual annotation is impractical and many automated U-Net-based methods yield disconnected components that hinder reliable biomarker extraction. We present RepAir, a three-stage framework for robust 3D airway segmentation that combines an nnU-Net-based network with anatomically informed topology correction. The segmentation network produces an initial airway mask, after which a skeleton-based algorithm identifies potential discontinuities and proposes reconnections. A 1D convolutional classifier then determines which candidate links correspond to true anatomical branches versus false or obstructed paths. We evaluate RepAir on two distinct datasets: ATM'22, comprising annotated CT scans from predominantly healthy subjects and AeroPath, encompassing annotated scans with severe airway pathology. Across both datasets, RepAir outperforms existing 3D U-Net-based approaches such as Bronchinet and NaviAirway on both voxel-level and topological metrics, and produces more complete and anatomically consistent airway trees while maintaining high segmentation accuracy.

Paper Structure

This paper contains 9 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. RepAir Architecture and Processing Pipeline
  4. Datasets
  5. Implementation and Training Details
  6. Experiments and Results
  7. Runtime Evaluation
  8. Performance Evaluation
  9. Conclusion

Figures (3)

  • Figure 1: Proposed segmentation pipeline. Teal boxes represent CNN layers. In the nnU-Net-based segmentation network, number of channels is shown below each layer. In the discontinuity classification network, input size is shown below each layer, and number of channels is shown to the left of each layer.
  • Figure 2: Inference time per CT scan for each method.
  • Figure 3: A 3D rendered view of the segmentation results on a representative image from the ATM'22 test set. RepAir is compared with Bronchinet and NaviAirway. Ground truth airways missed by each model (false negatives) are shown in red, with the predicted airway segmentation color coded by airway generation.