Boosting Hyperalignment Performance with Age-specific Templates

Abstract

Hyperalignment aligns individual brain activity and functional connectivity patterns to a common, high-dimensional model space, resolving idiosyncrasies in functional-anatomical correspondence and revealing shared information encoded in fine-grained spatial patterns. Given that the brain undergoes significant developmental and functional changes over the lifespan, it is likely that certain features in brain functional organization are more prominent in certain age groups than others. In this study, we examined whether age-specific functional templates, as compared to a canonical template, could enhance alignment accuracy across diverse age groups. We used the Cambridge Centre for Ageing and Neuroscience (Cam-CAN) dataset (18 to 87 yo) to build age-specific templates and tested their performance for analyzing data in young and old brains in both the Cam-CAN dataset and the Dallas Lifespan Brain Study (DLBS) dataset (20 to 90 yo). We found the congruent age-specific template outperforms the incongruent template for various analyses, including inter-subject correlation of hyperaligned connectivity profiles and predicting individualized connectomes using the template. The results are consistent across both datasets. This work enhances our understanding of age-related differences in brain function, highlights the benefits of creating age-specific templates to refine hyperalignment model performance, and may contribute to the development of age-sensitive diagnostic tools and interventions for neurological disorders.