Hierarchical Spherical Deformation for Shape Correspondence.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Computer Vision

Background:

• Establishing accurate shape correspondence is crucial for group-wise anatomical analysis.
• Existing methods often suffer from template selection bias and registration distortion.
• Landmark-based methods are labor-intensive and may not capture complex anatomical variations.

Purpose of the Study:

• To develop a novel, landmark-free, group-wise registration framework for shape correspondence.
• To address and minimize both template selection bias and registration distortion simultaneously.
• To achieve accurate and efficient non-rigid deformation for anatomical surfaces.

Main Methods:

• Proposed a novel spherical deformation technique extending spatial-varying Euler angles for non-rigid local deformation.
• Employed spherical harmonics interpolation for displacements to solve rigid and non-rigid deformations concurrently.
• Implemented a group-wise registration approach eliminating the need for a specific template.

Main Results:

• Achieved high accuracy in shape correspondence, demonstrated by improved cortical surface parcellation.
• Significantly reduced registration distortion in surface area and edge length compared to existing methods.
• Demonstrated fast registration times, completing analysis in approximately 3 minutes per subject.

Conclusions:

• The proposed spherical deformation method offers an accurate, efficient, and robust solution for landmark-free, group-wise shape correspondence.
• This framework effectively mitigates template bias and registration distortion, advancing computational anatomy.
• The method shows significant potential for applications in neuroimaging and other fields requiring precise anatomical analysis.