fMRI group analysis based on outputs from Matrix-Variate Dynamic Linear Models

TL;DR

This paper presents a detailed methodology for fMRI group analysis using Matrix-Variate Dynamic Linear Models, introducing algorithms for posterior distribution computation and real-time activation assessment, demonstrated on real data.

Contribution

It extends previous work by providing detailed procedures and an additional algorithm for online group-level voxel activation analysis in fMRI studies.

Findings

Algorithms successfully applied to real fMRI data

Effective estimation of group activation patterns

New online trajectory algorithm for voxel activation

Abstract

In this work, we describe in more detail how to perform fMRI group analysis using inputs from modeling fMRI signal using Matrix-Variate Dynamic Linear Models (MDLM) at the individual level. After computing a posterior distribution for the average group activation, the three algorithms (FEST, FSTS, and FFBS) proposed from the previous work by Jim\'enez et al. [2019] can be easily implemented. We also propose an additional algorithm, which we call AG-algorithm, to draw on-line trajectories of the state parameter and therefore assess voxel activation at the group level. The performance of our method is illustrated through one practical example using real fMRI data from a "voice-localizer" experiment.