Reference-free brain template construction with population symmetric registration.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Population registration normalizes images into a common space for clinical and research studies.
• Existing methods often use the arithmetic mean as a reference, leading to overly smooth templates and bias.
• Reducing bias in determining the common space is crucial for accurate population registration.

Purpose of the Study:

• To develop an efficient, symmetric, and reference-free population registration strategy for brain template construction.
• To address the limitations of smooth references and reference dependency in conventional population registration methods.
• To introduce a new metric, average bias, for evaluating template unbiasedness.

Main Methods:

• Proposed a symmetric population registration strategy defining a symmetric population center.
• Translated population registration into a series of pairwise registration problems for easier optimization.
• Constructed brain templates by approximating population intensity and gradient information.
• Introduced the average bias metric for unbiasedness evaluation.

Main Results:

• The proposed method demonstrated reduced bias compared to conventional approaches in synthetic data experiments.
• Reference-free validation confirmed the elimination of reference dependency-related bias.
• A symmetric brain template was successfully constructed from 20 MRI T1 volumes, showing improved clarity and reduced bias compared to DARTEL.

Conclusions:

• The developed symmetric population registration strategy offers an efficient and unbiased approach to brain template construction.
• The reference-free nature of the algorithm mitigates bias associated with reference selection.
• The method yields high-quality brain templates with clear anatomical details, suitable for neuroimaging research.