Bayesian Inference for Brain Activity from Multi-Resolution Functional Magnetic Resonance Imaging

• Activity Number: 221 - Contributed Poster Presentations: Section on Statistics in Imaging
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Andrew Whiteman* and Jian Kang and Timothy D Johnson
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Bayesian; fMRI; Gaussian Process; Computation; Hamiltonian

Abstract

As a presurgical tool, neurosurgeons use functional magnetic resonance imaging (fMRI) to map out functionally relevant brain regions. This application requires a high degree of spatial accuracy, but the fMRI signal-to-noise ratio (SNR) decreases as spatial resolution increases. In practice, both high and standard resolution fMRI may be used, and it is of interest to make more accurate inference on brain activity by combining data with different resolutions. We develop a new Bayesian model to leverage spatial precision from high-resolution fMRI and higher SNR from standard-resolution fMRI. We assume the observed fMRI data measure the mean intensity of brain activity at different locations with different levels of noise. We assign a Gaussian process prior to the mean intensity function and develop an efficient posterior computation algorithm. The algorithm works analogously to Riemann manifold Hamiltonian Monte Carlo with auxiliary parameters. We show in simulation our method makes inference on the mean intensity more accurately than alternatives that use either the high or standard resolution fMRI alone, and develop an analysis of de-identified data from presurgical patients.