Brain maturation: Predicting individual BrainAGE in children and adolescents using structural MRI

Abstract

Background

Neural development during human childhood and adolescence involves highly coordinated and sequenced events, characterized by both progressive and regressive processes. Despite a multitude of results demonstrating the age-dependent development of gray matter, white matter, and total brain volume, a reference curve allowing prediction of structural brain maturation is still lacking but would be clinically valuable. For the first time, the present study provides a validated reference curve for structural brain maturation during childhood and adolescence, based on structural MRI data.

Methods and findings

By employing kernel regression methods, a novel but well-validated BrainAGE framework uses the complex multidimensional maturation pattern across the whole brain to estimate an individual's brain age. The BrainAGE framework was applied to a large human sample (n = 394) of healthy children and adolescents, whose image data had been acquired during the NIH MRI study of normal brain development. Using this approach, we were able to predict individual brain maturation with a clinically meaningful accuracy: the correlation between predicted brain age and chronological age resulted in r = 0.93. The mean absolute error was only 1.1 years. Moreover, the predicted brain age reliably differentiated between all age groups (i.e., preschool childhood, late childhood, early adolescence, middle adolescence, late adolescence). Applying the framework to preterm-born adolescents resulted in a significantly lower estimated brain age than chronological age in subjects who were born before the end of the 27th week of gestation, demonstrating the successful clinical application and future potential of this method.

Conclusions

Consequently, in the future this novel BrainAGE approach may prove clinically valuable in detecting both normal and abnormal brain maturation, providing important prognostic information.

Introduction

Human brain development involves highly coordinated and sequenced events characterized by both progressive (e.g., cell growth and myelination) and regressive (e.g., synaptic pruning) processes (Silk and Wood, 2011). Especially with the advent of magnetic resonance imaging (MRI), cross-sectional as well as longitudinal neuroimaging studies contributed to a better understanding of healthy brain maturation. In addition, with the availability of automated computational methods for analyzing MRI data, such as voxel-based morphometry (VBM; Ashburner and Friston, 2000), it has become feasible to quantify and visualize structural brain changes in vivo (May, 2011) in truly healthy children, which effectively was not possible before (Giedd et al., 1996).

Volumetric MRI studies have reliably established the overall pattern of an initial rapid increase in total gray matter (GM) volume, followed by a phase of slower growth and, after reaching a peak in childhood, by a slow but continued reduction. In contrast, total white matter (WM) volume increases rapidly until the age of 10–15 years, with continued gain well beyond adolescence (Giedd et al., 1999, Groeschel et al., 2010, Silk and Wood, 2011). With the growing number of studies that have investigated both normal and abnormal brain changes with age, most major neuropsychiatric disorders are now thought to arise due to deviations from normal brain development during childhood and/or adolescence (Giedd et al., 2009, Paus et al., 2008). For some of these (e.g., childhood-onset schizophrenia or pediatric bipolar disorder), early detection afforded by a brain maturation reference curve would be clinically relevant.

Recently, Dosenbach et al. (2010) emphasized the need of maturation curves for pediatric brain development to aid in the early detection of neurodevelopmental disorders. They reported how brain maturity across development can be predicted with functional connectivity MRI (fcMRI). Using fcMRI data of 238 healthy subjects (aged 7 to 30 years), they predicted a maturation curve that accounted for 55% of the sample variance. Given the clinical potential of such predictions, approaches that lead to accounted variances of even more than 55% would be highly attractive. At the same time, it would be desirable to keep the duration of image acquisition as short as possible due to the known difficulties of motion artifacts in pediatric MRI studies (Wilke et al., 2008, Yuan et al., 2009). Moreover, such approaches will maintain a high impact on clinical diagnosis and intervention only if they are less prone to measurement bias due to individual alertness and vigilance (Van Dijk et al., 2012).

Focusing on these facts, our group has recently developed a new approach based on structural MRI data that enables one to reliably estimate the brain age of any given subject (Franke et al., 2010). By employing kernel regression methods in a large training database, the complex multidimensional aging patterns across the whole brain are detected and aggregated to one single value (i.e., the estimated brain age). Consequently, although using only one MRI scan per subject, the degree of acceleration or deceleration of brain aging can be directly quantified in terms of years allowing a wide range of analyses and predictions on an individual level. In an exemplary analysis with elderly adults, this brain age estimation model showed its potential to provide clinically relevant information by reporting a statistically significant, positive deviation of 10 years between the estimated and chronological ages in patients with Alzheimer's disease, indicating accelerated brain atrophy and underlining the diagnostic potential of such an approach.

In the present study we show the potential of our BrainAGE framework to reliably predict structural maturity levels of brains from healthy children and adolescents ranging between 5 and 18 years based on one structural MR scan per subject. Further, the age estimation framework will be exemplarily applied to a clinical sample of adolescents born preterm. We hypothesize that the group with a very low gestational age (GA < 27 weeks) would have a significantly lower estimated brain age than the group with a higher gestational age (GA > 29 weeks) due to decelerated brain maturation, which is presumed to be caused by the extremely premature birth status.