Landmark-based spherical quasi-conformal mapping for hippocampal surface registration

Area of Science:

• Medical imaging analysis
• Computational anatomy
• Neuroscience

Background:

• Alzheimer's disease (AD) is linked to hippocampal structural changes detectable by MRI.
• Accurate analysis of AD-induced hippocampal morphology requires one-to-one surface correspondence.
• Existing landmark-based methods struggle with large deformations and diffeomorphism.

Purpose of the Study:

• To develop a robust landmark-based spherical registration method for hippocampal surfaces.
• To establish precise one-to-one correspondence for comparing morphological changes in AD.
• To overcome limitations of existing registration techniques in handling large deformations.

Main Methods:

• Utilized eigen-graph for intrinsic landmark extraction from hippocampal surfaces.
• Employed barycentric coordinates and mesh optimization for parameterization to a unit sphere.
• Applied local stereographic projection and quasi-conformal mapping with Beltrami coefficients for landmark alignment and distortion control.

Main Results:

• The method was validated on real hippocampus data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Achieved low area distortion indices (ADI) in both AD (0.4362e-4±0.7800e-5) and normal control (NC) groups (0.5671e-4±0.602e-5).
• Attained high classification accuracy (94.2%) for AD vs. NC groups using morphological features from registered surfaces.

Conclusions:

• The proposed algorithm enables precise landmark alignment and bijectivity on hippocampal surfaces.
• It effectively handles large deformations, crucial for analyzing AD-related structural changes.
• This method enhances the potential for accurate morphological analysis in AD research.