SPHARM-Reg: Unsupervised Cortical Surface Registration Using Spherical Harmonics.

Area of Science:

• Neuroimaging
• Computational Anatomy
• Medical Image Analysis

Background:

• Cortical shape analysis is crucial for understanding brain structure and function.
• Existing registration methods often introduce biases due to warp distortion.
• Preserving rotational information during smoothing is a significant challenge.

Purpose of the Study:

• To introduce SPHARM-Reg, a novel learning-based spherical registration method.
• To reduce warp distortion and improve accuracy in cortical shape correspondence.
• To address limitations in joint alignment and rotation-preserving smoothing.

Main Methods:

• Developed a diffeomorphic framework integrating spherical harmonic decomposition of the velocity field.
• Implemented a novel velocity encoding scheme using spherical functions for rotation-preserving smoothing.
• Optimized harmonic components for joint rigid and non-rigid alignment adjustments.

Main Results:

• SPHARM-Reg significantly reduced warp distortion compared to existing methods.
• Achieved high registration accuracy on healthy adult datasets.
• Demonstrated that warp distortion extent impacts statistical significance in clinical analysis.

Conclusions:

• SPHARM-Reg offers a robust solution for accurate cortical shape registration.
• The method effectively minimizes bias in downstream shape analyses.
• Reduced warp distortion leads to more reliable statistical findings in neuroimaging studies.