Optimal multi-object segmentation with novel gradient vector flow based shape priors.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Computational Anatomy

Background:

• Shape priors enhance medical image segmentation accuracy and robustness.
• Traditional mesh-based shape models suffer from self-intersection and folding issues, requiring complex mitigation algorithms.

Purpose of the Study:

• To propose a novel shape prior representation directly in voxel grid space.
• To extend shape priors for simultaneous segmentation of multiple interacting objects with separation constraints.

Main Methods:

• Developed a shape prior based on gradient vector flows of pre-segmentation within the voxel grid.
• Formulated multi-object segmentation as a Markov Random Field problem.
• Solved the optimization problem using a single minimum s-t cut on a constructed graph.

Main Results:

• Validated the algorithm on brain tissue segmentation (MRI) and bladder/prostate segmentation (CT).
• Demonstrated superior or competitive performance compared to state-of-the-art methods.

Conclusions:

• The proposed voxel-based shape prior effectively addresses limitations of mesh-based methods.
• The algorithm offers a flexible and efficient approach for multi-object segmentation in medical imaging.