Statistical Analysis on Brain Surfaces

TL;DR

This paper reviews statistical analysis frameworks for data on brain surfaces, emphasizing geometric methods and random field theory, illustrated with hippocampus data to study effects of family income on child brain development.

Contribution

It provides a comprehensive overview of geometric and statistical methods for surface data analysis, highlighting recent developments and applications in neuroimaging.

Findings

Surface data smoothing improves signal detection on cortical manifolds

Random field theory enables statistical inference on brain surface data

Case study shows potential link between income and hippocampal growth

Abstract

In this paper, we review widely used statistical analysis frameworks for data defined along cortical and subcortical surfaces that have been developed in last two decades. The cerebral cortex has the topology of a 2D highly convoluted sheet. For data obtained along curved non-Euclidean surfaces, traditional statistical analysis and smoothing techniques based on the Euclidean metric structure are inefficient. To increase the signal-to-noise ratio (SNR) and to boost the sensitivity of the analysis, it is necessary to smooth out noisy surface data. However, this requires smoothing data on curved cortical manifolds and assigning smoothing weights based on the geodesic distance along the surface. Thus, many cortical surface data analysis frameworks are differential geometric in nature. The smoothed surface data is then treated as smooth random fields and statistical inferences can be performed within Keith Worsley's random field theory. The methods described in this paper are illustrated with the hippocampus surface data set. Using this case study, we will determine if there is an effect of family income on the growth of hippocampus in children in detail. There are a total of 124 children and 82 of them have repeat magnetic resonance images (MRI) two years later.