Correlation and heritability in neuroimaging datasets: a spatial decomposition approach with application to an fMRI study of twins.

Advances in modern neuroimaging in combination with behavioral genetics have allowed neuroscientists to investigate how genetic and environmental factors shape human brain structure and function. Estimating the heritability of brain structure and function via twin studies has become one of the major approaches in studying the genetics of the brain. In a classical twin study, heritability is estimated by computing genetic and phenotypic variation based on the similarity of monozygotic and dizygotic twins. However, heritability has traditionally been measured for univariate, scalar traits, and it is challenging to assess the heritability of a spatial process, such as a pattern of neural activity. In this work, we develop a statistical method to estimate phenotypic variance and covariance at each location in a spatial process, which in turn can be used to estimate the heritability of a spatial dataset. The method is based on a dimensionally-reduced model of spatial variation in paired images, in which adjusted least squares estimates can be used to estimate the key model parameters. The advantage of the proposed method compared to conventional methods such as voxelwise or mean-ROI approaches is demonstrated in both a simulation study and a real data study assessing genetic influence on patterns of brain activity in the visual and motor cortices in response to a simple visuomotor task.