A new method for analyzing local shape in three-dimensional images based on medial axis transformation

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Geometry

Background:

• Three-dimensional (3-D) image analysis requires robust methods for shape description.
• Medial axis transformation offers a simplified and reversible representation of structures in 3-D images.

Purpose of the Study:

• To develop a new approach for geometrical shape description in 3-D images using medial axis transformation.
• To classify voxels within 3-D images into boundary, branching, regular, and arc points.

Main Methods:

• Utilizing 3-D medial axis transformation for shape representation.
• Classifying voxels based on topological properties within a local region of interest, scaled by local structure thickness.
• Employing the reversibility of the medial axis to label the entire object.

Main Results:

• A novel voxel classification scheme for 3-D shapes based on medial axis properties.
• Successful evaluation on simulated 3-D images.
• Demonstrated application in identifying bone structures from high-resolution tomographic data.

Conclusions:

• The proposed 3-D medial axis transformation method provides an effective way to describe and label geometrical shapes in 3-D images.
• This approach has practical applications in medical image analysis, particularly for bone structure identification.