Improving deformable surface meshes through omni-directional displacements and MRFs

Area of Science:

• Medical Image Analysis
• Computer Vision
• Computational Geometry

Background:

• Deformable surface models, often triangular meshes, are crucial for image segmentation.
• Current methods using vertex movement along line segments (surface normals) face limitations with high mesh curvature, leading to inaccurate boundary fitting.
• These limitations can result in non-corresponding intersections or missed contact with the target boundary.

Purpose of the Study:

• To develop an improved deformable surface model approach for image segmentation.
• To overcome the limitations of line-segment-based vertex movement in high-curvature regions.
• To achieve more accurate and globally regularized deformations onto target object boundaries.

Main Methods:

• Proposed an approach where mesh vertices can move within a surrounding sphere, not just along line segments.
• Employed Markov Random Field optimization for globally regularized deformations.
• Validated the method using synthetic data and real-world medical imaging datasets (Cone-Beam CTs of mandibles and low-resolution CTs of coccyxes).

Main Results:

• The sphere-based vertex movement demonstrated potential for overcoming limitations of traditional methods.
• The Markov Random Field optimization ensured globally regularized and accurate deformations.
• Successful application shown on complex anatomical structures like coronoid processes and coccyxes.

Conclusions:

• The proposed sphere-based vertex movement within deformable surface models offers a significant improvement for image segmentation.
• This approach enhances accuracy, particularly in areas with high mesh curvature.
• The method shows promise for robust segmentation of challenging anatomical structures in medical imaging.