Advancement in Brain-Age Estimation: The Largest Study on Brain Age Models Across the Human Lifespan

Writter: Ruiyang Ge (PhD, University of British Columbia)

---

The ENIGMA Lifespan Working Group is excited to share their latest work in estimating the biological age of the brain using structural neuroimaging data: "Brain-age prediction: Systematic evaluation of site effects, and sample age range and size". This method, known as brain-age estimation, has proven to be a valuable tool in understanding brain development and aging, offering insights into various biologically and behaviorally meaningful measures.

Recent research has highlighted the need for robust and publicly accessible brain-age models pre-trained on large samples of healthy individuals. To meet this demand, the team has previously introduced a developmental brain-age model. Now, they are excited to announce the expansion of this work with the development, empirical validation, and dissemination of a comprehensive brain-age model covering most of the human lifespan.

Their new approach involved a thorough examination of various data site harmonization strategies, age ranges, and sample sizes to determine their impacts on brain-age prediction models. The team used brain morphometric data from a discovery sample of over 35,000 healthy individuals, aged 5 to 90 years (54% female). The best-performing model was selected based on its accuracy and reliability. Key findings from their research include:

• Higher accuracy in age prediction from morphometry data was achieved without applying site harmonization, highlighting the need for more careful consideration of cross-site variability and the development of new approaches for site harmonization in this application.

• Dividing the discovery sample into two age groups (5–40 and 40–90 years) resulted in a better balance between model accuracy and explained age variance.

• Model accuracy for brain-age prediction plateaued with sample sizes exceeding 1600 participants.

The pre-trained brain-age predictive models were tested for cross-dataset generalizability in an independent sample of 2101 healthy individuals (age range: 8–80 years; 55.35% female) and for longitudinal consistency in a further sample of 377 healthy individuals (age range: 9–25 years; 49.87% female).

These findings have been incorporated into CentileBrain (https://centilebrain.org/#/brainAGE2), an open-science, web-based platform for individualized neuroimaging metrics. The article is now published as an open-access paper in Human Brain Mapping.