Structure and location preserving topological representation with applications on CT segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Topology

Background:

• Traditional contour trees lack geometric terrain structure crucial for CT image interpretation.
• Accurate segmentation of adjacent objects with similar intensities remains a challenge in CT analysis.

Purpose of the Study:

• To develop a new topological representation for CT images that captures both nesting and spatial relationships.
• To evaluate the application of this representation as a constraint for target object segmentation.

Main Methods:

• Constructed a novel tree structure using signed distance transformation of binary CT images.
• Combined intensity-based contour trees with a gradient-based topology tree.
• Applied the topological representation as a constraint in segmentation tasks.

Main Results:

• The proposed tree structure effectively retained nesting and spatial relationships of tissues and objects in CT images.
• Segmentation accuracy was evaluated using ten non-small cell lung tumor CT studies.
• The representation proved useful in separating adjacent objects with similar intensity distributions.

Conclusions:

• The novel topological representation enhances CT image interpretation by incorporating spatial and nesting information.
• This method offers a viable approach for improving segmentation accuracy, especially in complex cases.