BayesfMRI: User-friendly spatial Bayesian modeling for task fMRI

1325 

Abstract Submission 

Damon Pham¹, David Bolin², Martin Lindquist³, Thomas Nichols⁴, Amanda Mejia¹

¹Indiana University, Bloomington, IN, ²King Abdullah University of Science and Technology, Thuwal, Saudi Arabia, ³Johns Hopkins University, Baltimore, MD, ⁴University of Oxford, Oxford, United Kingdom

Damon Pham  
Indiana University
Bloomington, IN

David Bolin, PhD  
King Abdullah University of Science and Technology
Thuwal, Saudi Arabia

Martin A. Lindquist  
Johns Hopkins University
Baltimore, MD

Thomas Nichols  
University of Oxford
Oxford, United Kingdom

Amanda Mejia  
Indiana University
Bloomington, IN

Spatial Bayesian models are a powerful way to account for spatial dependencies in fMRI analysis [3]. While massive univariate modeling treats each voxel or vertex as a separate entity, spatial Bayesian models place a multivariate prior distribution on the underlying maps of activation, which encodes the spatial dependence and implicitly smoothes the activation estimates. This avoids ad-hoc data smoothing, which induces spatially dependent noise and can lead to false positive clusters [2]. In addition, spatial Bayesian models can use the joint posterior distribution across brain locations to identify areas of activation. This dramatically increases power to detect activations and facilitates the use of meaningful minimum effect sizes, even in individual-level analysis [5]. The open-source BayesfMRI R package provides a user-friendly interface for spatial Bayesian models for task fMRI analysis.

Importantly, the spatial Bayesian models implemented in BayesfMRI are surface-based and subcortical parcel-constrained. These grayordinates-based models leverage spatial dependencies in a neurobiologically appropriate way and avoid the mixing of signals known to occur with whole-brain volumetric smoothing or spatial modeling [1].

Fig. 1 illustrates the main functions of the BayesfMRI package. The BayesGLM function fits a spatial Bayesian model for subject-level task fMRI analysis. Wrapper functions provide direct compatibility with CIFTI, GIFTI and NIFTI-format data. The id_activations function identifies areas of significant activation above a specified minimum effect size, using the joint posterior distribution across locations. This substantially increases power to detect activations while controlling the family-wise error rate (FWER). Second-level analysis relies on the BayesGLM2 function, which uses group-average spatial Bayesian modeling [5]. Alternatively, the function act_prevalence produces group prevalence maps based on the subject-level activation maps. The plot function can be used to produce user-friendly visualizations of all results.

BayesfMRI also includes fast prewhitening using spatially variable high-order AR modeling to effectively mitigate autocorrelation [4]. Planned future functionality includes Bayesian estimation and inference of hemodynamic response function (HRF) shape (e.g. height, width, onset) and data-driven individualized HRF estimation.

Fig. 2 provides a demo of BayesfMRI, with pseudo-code and sample results based on an analysis of the HCP working memory task (n=20). First- and second-level spatial Bayesian models are illustrated. Example visualizations based on the real data are shown on the right. These show smooth estimates and robust areas of activation at both the individual and group level. As an alternative to group averages, group prevalence maps show the proportion of subjects exhibiting activation at each location. This shows similar patterns as the group-average activations but provide greater nuance and extent. These robust prevalence maps are possible due to high power at the first level provided by spatial Bayesian modeling.

BayesfMRI is a user-friendly, open-source R package that facilitates spatial Bayesian modeling for task fMRI analysis. The spatial models are based on grayordinates data to leverage spatial dependencies along the cortical surface and within subcortical structures, which avoids blurring anatomically distinct areas. BayesfMRI implements single- and multi-session/longitudinal subject-level analysis, second-level group analysis, prewhitening, and powerful joint posterior inference to identify areas of activation above a specified minimum effect size. Convenient visualizations are also provided. BayesfMRI is available through both CRAN and GitHub.

Activation (eg. BOLD task-fMRI) ¹

Bayesian Modeling ²

Data analysis

FUNCTIONAL MRI

Modeling

Open-Source Software

Statistical Methods

Workflows