JUST: A joint universal and specific brain templates generation framework

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Universal and specific brain templates are crucial for neuroimaging research.
• Existing generation-based methods excel at learning either universal or specific templates but lack a unified framework, leading to inefficiencies.
• Current methods often fail to leverage distinguishing information from mismatched cohorts, potentially limiting template specificity.

Purpose of the Study:

• To propose a novel brain template generation framework, JUST, that synergizes universal and specific template learning.
• To enhance template specificity by explicitly leveraging template-cohort divergence.
• To introduce an innovative metric, the Template Distinguishability Score (TDS), for quantifying template specificity.

Main Methods:

• Developed the JUST framework, unifying universal and specific template construction via a conditional generator producing deformations.
• Formulated correlations among specific templates and cohorts, employing a contrastive loss to enforce template specificity.
• Introduced the Template Distinguishability Score (TDS) based on cross-cohort registration deformation norms.

Main Results:

• JUST improves computational efficiency by 33.6% compared to baseline methods.
• Achieved optimal performance on 14 out of 18 metrics, including visual quality, centrality, and specificity, across two public brain datasets.
• Validated the framework and TDS metric in a clinical scenario, showing potential for downstream applications.

Conclusions:

• The JUST framework offers an efficient and effective solution for joint universal and specific brain template learning.
• The proposed TDS metric provides a reliable quantification of template specificity.
• JUST and TDS demonstrate significant potential for advancing neuroimaging research and clinical applications.