Prediction of chronological age from hierarchical brain volumes using a random forest regression can provide a personalized lifetime metric of aging. (P6.078)

Objective: To develop a reliable prediction of age using imaging data and demographic information in order to provide a personalized lifetime metric of aging.

Background: Understanding how brain volumes change over time provides insight into development, aging, and disease. Age-related changes have been studied in detail for specific age ranges, e.g. during early life for neurodevelopmental disorders and late life for Alzheimer’s Disease, but a normative description of volume change over the human lifetime has been less well described. Recent advances in the sharing and processing of large data sets have allowed access to many brain images of normal, healthy volunteers in order to perform a large-scale cross-sectional analysis across the lifetime.

Design/Methods: Using an established hierarchy of brain structure, we seek to understand how groups of brain regions can predict age at different developmental time points. To capture nonlinear changes, we propose random forest regression prediction of age based on multi-variate, hierarchical brain volumes alongside categorical demographic variables of gender and handedness. We derived a predictive age model using imaging from 4575 healthy patients from 60 sites and spanning ages from 4 to 94.

Results: Preliminary results show age-prediction within a margin of ± 5 years during ages 4–30 and a margin of ± 10 years during ages 30–90.

Conclusions: Volume-based age prediction provides a personalized analysis of brain morphology regardless of chronological age. This new metric can be compared to an age-matched population to describe individual aging or it can be compared to similar morphology populations to find patterns in disease or the effect of treatment. These results may also provide insight into how developmental networks grow together and their relative importance at different stages in life.

Disclosure: Dr. Bermudez has nothing to disclose. Dr. Huo has received research support from Incyte. Dr. Plassard has nothing to disclose. Dr. Aboud has nothing to disclose. Dr. Cutting has received personal compensation for activities with Houghton Mifflin Harcourt as a consultant. Dr. Bennett has received research support from Incyte.