Optimal graph search segmentation using arc-weighted graph for simultaneous surface detection of bladder and...

Area of Science:

• Medical image analysis
• Computer vision
• Computational geometry

Background:

• Accurate segmentation of interacting anatomical structures is crucial for medical imaging analysis.
• Existing methods often struggle with partially interacting objects and preserving geometric relationships.
• Incorporating both edge and shape information is challenging yet vital for robust segmentation.

Purpose of the Study:

• To develop a novel, globally optimal method for simultaneous surface segmentation of multiple mutually interacting objects.
• To enforce hard constraints in interacting regions and soft shape priors for improved accuracy.
• To apply the method to challenging cases like bladder and prostate segmentation in CT images.

Main Methods:

• A 3-D graph-theoretic approach utilizing an arc-weighted graph representation.
• Solving the segmentation problem via a maximum flow formulation.
• Incorporating hard surface interaction constraints and soft shape smoothness priors into an energy functional.

Main Results:

• Simultaneous segmentation of multiple partially interacting objects with globally optimal solutions.
• Preservation of geometric relationships in interacting surface regions.
• Successful application to bladder and prostate surface detection in CT images, yielding encouraging results despite low object saliency.

Conclusions:

• The proposed arc-weighted graph method offers a robust and accurate approach for simultaneous segmentation of interacting objects.
• This is the first method to use arc-weighted graphs for this specific segmentation problem, achieving global optimality efficiently.
• The technique shows promise for clinical applications, particularly in segmenting challenging structures in volumetric medical data.