JOSA: Joint surface-based registration and atlas construction of brain geometry and function

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Surface-based cortical registration is crucial for medical image analysis.
• Current methods rely on geometric features, assuming folding patterns align with brain function, which is often not the case.

Purpose of the Study:

• To introduce JOSA, a novel framework for cortical registration.
• To jointly model geometric and functional mismatches for improved registration.
• To develop an unbiased, population-specific atlas.

Main Methods:

• JOSA employs a semi-supervised training strategy.
• It jointly models the mismatch between brain geometry and function.
• The framework learns a population-specific atlas simultaneously.

Main Results:

• JOSA demonstrates superior registration performance in both geometry and function compared to state-of-the-art methods.
• The framework achieves high accuracy without needing functional data during inference.
• It can be extended to incorporate various auxiliary data types for registration guidance.

Conclusions:

• JOSA offers a new approach to cortical registration by acknowledging the geometry-function mismatch.
• The framework provides a powerful tool for analyzing brain structure and function.
• It opens avenues for future registration methods integrating multimodal brain data.