Across the brain, many locations tend to exhibit similar functional behavior, whether at rest or while performing a task. These may be neighboring locations along the cortex, within a subcortical structure, or distal locations belonging to the same functional brain network. Spatial Bayesian models formally account for this functional similarity through priors encoding the expected correlation between all locations in the brain. In comparison to massive univatiate approaches, these models have much higher power to detect effects and produce more accurate and reliable function brain measures. We focus on cortical surface-based and subcortical parcel-constrained models to leverage spatial dependencies without blurring across tissue classes or distal cortical areas.

Functional MRI data is subject to high levels of artifacts from a myriad of sources, including subject head motion, scanner instabilities and processing errors. Eliminating or reducing these sources of noise is essential for optimizing accuracy and power in downstream analysis. Data-driven denoising and scrubbing methods detect abnormalities in the data and reduce them while preserving more signal than conventional head motion-based techniques. Our novel projection scrubbing approach utilizes ICA to isolate noise from signal, combined with multivariate outlier detection methods to flag contaminated volumes. We are working to develop robust outlier detection techniques appropriate for fMRI data to maximize sensitivity to artifacts while avoiding false flags.