Show me your brain and I'll tell you how smart you are: Multivariate prediction of general intelligence from patterns of regional gray matter volume

General intelligence has been associated with individual differences in morphological characteristics of the brain. In particular, gray matter volume has been shown to vary in distinct brain regions depending on intelligence... [ view full abstract ]

General intelligence has been associated with individual differences in morphological characteristics of the brain. In particular, gray matter volume has been shown to vary in distinct brain regions depending on intelligence (for meta-analyses, see Jung & Haier, 2007; Basten et al., 2015). However, the majority of previous investigations focused on correlative associations, which maximize explained variance within a given sample, without considering generalizability to independent samples. To demonstrate the predictive value of individual differences in morphometric patterns of gray matter volume for intelligence, we applied voxel-based morphometry (VBM) on structural magnetic resonance imaging (MRI) data from 308 adult participants (Nooner et al., 2012). We controlled for age, sex, and handedness, and rescaled all gray matter volume values with total intracranial volume to study the association between brain structure and intelligence (Wechsler Abbreviated Scale of Intelligence, WASI; Wechsler et al., 1999) beyond simple differences in brain size. Using Support Vector Regression and a nested cross validation scheme we show that significant prediction of individual intelligence test scores in previously unseen subjects is indeed possible (correlation between observed and predicted IQ scores, r = .32, p < .001). However, albeit the prediction is significant on a group aggregated level, for individual subjects the absolute difference between predicted and observed IQ score can be quite large (mean absolute error, MAE = 10 IQ points). In conclusion, our study provides robust support for the association between brain structure and general intelligence, and demonstrates the potential but also the limits of doing brain-based inferences about cognitive ability at the level of the individual person.