A Robust and Efficient Curve Skeletonization Algorithm for Tree-Like Objects Using Minimum Cost Paths

Area of Science:

• Computer Vision
• Medical Imaging Analysis
• Geometric Modeling

Background:

• Conventional curve skeletonization methods often generate spurious branches, impacting accuracy.
• Existing algorithms struggle with boundary irregularities and digital artifacts in 3D fuzzy objects.

Purpose of the Study:

• To develop a robust and efficient 3D curve skeletonization algorithm for elongated fuzzy objects.
• To overcome limitations of existing methods by eliminating spurious branches without post-processing.

Main Methods:

• A minimum cost path approach is employed, starting from a root voxel and iteratively expanding the skeleton.
• A novel local significance factor, based on fuzzy distance transform, guides the pathfinding to object centerlines.
• Algorithm termination occurs when the object volume is filled or meaningful branches can no longer be generated.

Main Results:

• The new algorithm demonstrates superior accuracy and robustness in detecting true and false skeletal branches compared to existing methods.
• Evaluated using computer-generated phantoms and in vivo CT imaging of human airways, the method showed improved performance.
• Significant reduction in computation complexity achieved by detecting multiple branches per iteration.

Conclusions:

• The proposed minimum cost path algorithm offers a robust and efficient solution for 3D curve skeletonization of fuzzy objects.
• It effectively mitigates spurious branches and reduces computational load, outperforming conventional techniques.
• This advancement has implications for accurate analysis and modeling of complex biological structures.