Electronic Thesis and Dissertation Repository

Thesis Format



Master of Science


Statistics and Actuarial Sciences


Jorn Diedrichsen


Multi-voxel pattern analysis (MVPA) provides a powerful framework for making statistical inferences on the information present in brain activity patterns as measured by functional magnetic resonance imaging (fMRI). Many recent studies suggest that MVPA performance benefits from taking into account the spatial voxel-to-voxel correlations in the measurement noise. However, estimating these noise correlations is challenging due to the limited data points and large voxel counts. To address this issue, it is common practice to shrink the empirical correlation estimate towards its identity matrix, which biases the estimate towards the incorrect assumption that voxels are independent. We therefore propose an anatomically-informed model of measurement noise in fMRI, which takes into account the distances of voxels in the measurement volume, their distance on the cortical sheet, and the depth at which they sample the cortex. Our model can predict the noise-correlation structure in new participants and datasets. It improves the noise correlation estimate when used as a shrinkage target, thereby also potentially improving statistical inferences in MVPA.

Summary for Lay Audience

Functional Magnetic Resonance Imaging (fMRI) is a non-invasive imaging technique that allows researchers to measure the activity in the human brain. To compare the fine-grained pattern of brain activation, researchers employ multivariate analysis methods to make inferences on the activity observed in groups of voxels, the 3D analog of image pixels in a brain image. In these analyses, they often assume that the measurement error is independent across voxels. However, that is not the case; fMRI data has a strong and reliable correlation structure across voxels. Mounting evidence shows that, based on this incorrect assumption, researchers are likely to arrive at incorrect decisions. Therefore, the goal of this project is to build a model of the spatial noise correlation structure in fMRI data and use it to improve multivariate inference. We construct potential noise correlation models using different aspects of the anatomical information. For example, two voxels that are close to each other are more correlated than two voxels that are far apart. The spatial distance in the volume hence is one of the anatomical factors that we consider. We also consider the distance along the cortical surface (i.e., taking into account the individual brain folding structure) and whether the voxel measured superficial or deep aspects of the brain. Our results show that the use of our anatomically-informed noise correlation model can lead to better inference than assuming the independence of voxels.