Alzheimer's Imaging Consortium.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Automated face recognition advances necessitate de-facing software for brain MRIs.
• Current de-facing tools are validated on limited ethnoracial groups, primarily non-Hispanic White participants.
• Performance disparities in de-facing software across diverse populations remain understudied.

Purpose of the Study:

• To evaluate the performance of de-facing software across different ethnoracial groups.
• To compare the impact of de-facing on brain measurements (gray matter volume and cortical thickness) across non-Hispanic White, African American, and Mexican American individuals.

Main Methods:

• Utilized T1-weighted MRI data from 305 participants in the Health and Aging Brain Study: Health Disparities (HABS-HD) cohort.
• Matched participants across three ethnoracial groups (non-Hispanic White, African American, Mexican American) by age, sex, and cognitive status.
• Applied mri_reface software for de-facing and FreeSurfer for brain measurements, comparing differences between de-faced and original images across groups.

Main Results:

• Cortical thickness differences due to de-facing varied significantly between non-Hispanic White and Mexican American, and between African American and Mexican American groups.
• Gray matter volume differences were significant between African American and Mexican American groups.
• Observed differences were generally small, with minimal impact on most brain regions analyzed.

Conclusions:

• De-facing software exhibits ethnoracial performance variations in neuroimaging studies.
• While statistically significant, the observed differences in brain measurements are unlikely to affect current analyses.
• Future work will focus on optimizing de-facing algorithms to ensure equitable performance across all populations.