NeuroImage · Sep 18, 2025

Developing and validating an elderly brain template: A comprehensive comparison with MNI152 for age-specific neuroimaging analyses

By Ota K, Nakazato Y, Oyama G

BACKGROUND: The Montreal Neurological Institute 152 (MNI152) brain template, constructed from young adult brains, may not accurately represent older age-specific morphological changes. Accordingly, we developed and validated the new Elderly template. METHODS: MRI scans from 90 OASIS-1 participants, matching Japanese census demographics, were used to construct the Elderly template. Spatial normalization accuracy was compared with that of the MNI152 in the IXI dataset. Following UK Biobank-based intracranial volume quality control (±2 SD; 1232-1850 mL), 282 of 313 scans from individuals aged 20-80+ years were included. Whole-brain similarity was assessed with cross-correlation (CC), mean-squared error (MSE), and 3D structural similarity index measure (3D-SSIM). Dice coefficients were computed for white matter (WM), gray matter (GM), cerebrospinal fluid (CSF), and seven subcortical regions. Generalized linear models were used to test the Age × Template interactions (β₃). RESULTS: Significant Age × Template interactions were observed for CC and MSE (p < 0.001); 3D-SSIM showed a positive but non-significant trend. Dice analyses mirrored this pattern: WM and GM showed minor differences between templates, and the Dice coefficient was parallel. CSF showed a sharp difference at the age ≥60 years. The largest interaction effects (≈1-4 % gain) occurred in the caudate, thalamus, hippocampus, and amygdala, whereas the brainstem, pallidum, and putamen showed minimal differences between templates. CONCLUSIONS: The Elderly template more accurately reflects older age-specific morphological changes and enhances spatial normalization accuracy, compared with the MNI152 template. This improvement suggests advancements in age-specific analyses and neurodegenerative disease research, enabling clinical applications.

doi.org/10.1016/j.neuroimage.2025.121473Open accessPubMed 40975146AI & data