Optimal surface segmentation using flow lines to quantify airway abnormalities in chronic obstructive pulmonary...

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Accurate segmentation of multi-dimensional and multi-surface structures is crucial for medical image analysis.
• Existing graph-based methods struggle with high-curvature surfaces, leading to self-intersecting solutions.
• Human airway segmentation is vital for diagnosing and monitoring lung diseases.

Purpose of the Study:

• To introduce a novel graph construction method for improved segmentation of complex surfaces.
• To address limitations of conventional graph columns for high-curvature structures.
• To apply and validate the method for human airway segmentation in 3D CT images.

Main Methods:

• Developed a graph construction method using non-intersecting flow lines to derive graph columns.
• Applied the method to segment human airway walls in three-dimensional computed tomography (CT) images.
• Validated accuracy using phantom measurements and compared with a previous graph-based method on 2D cross-sections.

Main Results:

• Achieved sub-voxel accuracy in estimating inner and outer airway radii from phantom data.
• Demonstrated significantly better performance than a prior method, with 89.3% average overlap and 0.096mm average surface distance.
• Medical experts preferred the new method in 68.5% of visual evaluations.
• Airway measurements showed significant correlation with lung function and high reproducibility (R(2) 0.96-0.73) in lung cancer screening data.

Conclusions:

• The proposed graph construction method effectively segments complex, high-curvature surfaces like human airway walls.
• The method provides accurate, reproducible measurements relevant for clinical applications, including lung cancer screening.
• This approach represents a significant advancement in medical image segmentation for anatomical structures.