Automated 3D closed surface segmentation: application to vertebral body segmentation in CT images

Area of Science:

• Biomedical Imaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Accurate segmentation of anatomical structures is crucial for quantitative analysis in biomedical imaging.
• Approximately convex, blob-like structures like vertebral bodies present segmentation challenges in low-dose CT scans.
• Existing methods may lack the precision required for applications such as bone mineral density measurement.

Purpose of the Study:

• To introduce and evaluate the Progressive Surface Resolution (PSR) algorithm for automated segmentation of closed surfaces.
• To apply PSR to the segmentation of vertebral body cortical surfaces in low-dose chest CT images.
• To assess the potential of PSR for automated bone mineral density measurement and compression fracture detection.

Main Methods:

• The PSR algorithm utilizes a closed triangular mesh to ensure surface enclosure.
• Surface vertices are constrained along uniformly distributed radial trajectories in 3D space.
• Segmentation involves determining radial trajectory intersections with the target surface, progressively refined using a dynamic attraction map.

Main Results:

• Visual evaluation demonstrated acceptable segmentation for 99.35% of vertebral bodies.
• Quantitative evaluation on 46 vertebral bodies yielded a mean Dice coefficient of 0.939 (SD=0.011).
• The algorithm achieved high precision, with a maximum Dice coefficient of 0.957.

Conclusions:

• The PSR algorithm exhibits promising performance for vertebral body segmentation based on visual and quantitative assessments.
• This novel approach offers uniform angular resolution for segmented surfaces.
• The algorithm's computational complexity and runtime are linearly dependent on the number of mesh vertices.