Optimal graph based segmentation using flow lines with application to airway wall segmentation.

Area of Science:

• Medical imaging
• Computer vision
• Computational anatomy

Background:

• Multi-dimensional, multi-surface segmentation is crucial for medical image analysis.
• Conventional graph-based methods struggle with high-curvature or complex surfaces.
• Refining initial surfaces using graph columns is a common approach.

Purpose of the Study:

• To introduce a novel optimal graph construction method for complex surface segmentation.
• To improve the accuracy and robustness of segmentation for challenging anatomical structures.
• To apply and validate the method for human airway wall segmentation in computed tomography (CT) images.

Main Methods:

• Developed a novel graph construction method using non-intersecting flow lines to create specialized columns.
• Applied the method to segment human airway walls in multi-dimensional CT image data.
• Validated the method by comparing segmentation results with manual annotations and other graph-based techniques.

Main Results:

• The proposed method significantly reduces contour distances and increases the area of overlap compared to existing graph-based methods.
• Segmentation of human airway walls in 649 cross-sectional images from 15 subjects demonstrated superior performance.
• Airway abnormality measurements from 480 scan pairs in a lung cancer screening trial showed high reproducibility and significant correlation with lung function.

Conclusions:

• The novel graph construction method effectively handles complex and high-curvature surfaces in medical image segmentation.
• The method offers improved accuracy and robustness for segmenting human airway walls in CT images.
• The approach shows potential for reliable quantitative analysis of airway abnormalities in clinical settings, including lung cancer screening.