Large-scale Probabilistic Functional Modes from resting state fMRI

Samuel J Harrison,Mark W Woolrich,Emma C Robinson,Matthew F Glasser,Christian F Beckmann,Mark Jenkinson,Stephen M Smith

doi:10.1016/j.neuroimage.2015.01.013

Samuel J Harrison, Mark W Woolrich + Show 5 more

Open Access

https://doi.org/10.1016/j.neuroimage.2015.01.013

Copy DOI

Abstract

It is well established that it is possible to observe spontaneous, highly structured, fluctuations in human brain activity from functional magnetic resonance imaging (fMRI) when the subject is ‘at rest’. However, characterising this activity in an interpretable manner is still a very open problem.In this paper, we introduce a method for identifying modes of coherent activity from resting state fMRI (rfMRI) data. Our model characterises a mode as the outer product of a spatial map and a time course, constrained by the nature of both the between-subject variation and the effect of the haemodynamic response function. This is presented as a probabilistic generative model within a variational framework that allows Bayesian inference, even on voxelwise rfMRI data. Furthermore, using this approach it becomes possible to infer distinct extended modes that are correlated with each other in space and time, a property which we believe is neuroscientifically desirable.We assess the performance of our model on both simulated data and high quality rfMRI data from the Human Connectome Project, and contrast its properties with those of both spatial and temporal independent component analysis (ICA). We show that our method is able to stably infer sets of modes with complex spatio-temporal interactions and spatial differences between subjects.

Highlights

Using resting state functional magnetic resonance imaging (fMRI) it is possible to generate enormously rich data sets that capture some of the complexity of the brain's intrinsic dynamics and connectivity
In order to test the efficacy of our method we simulated an fMRI data set containing an embedded ground truth set of modes
It has long been known that with small numbers of time points temporal ICA (tICA) is much less robust than spatial independent component analysis (sICA) (Friston, 1998; McKeown et al, 1998); each of our simulated data sets has over five times as many time points as the data set used in the paper which introduced tICA as a method for identifying modes from resting state fMRI (rfMRI) data (Smith et al, 2012), so we would not expect this to be the limiting factor

Summary

Introduction

Using resting state fMRI it is possible to generate enormously rich data sets that capture some of the complexity of the brain's intrinsic dynamics and connectivity. Generating representations that meaningfully simplify the data, while still capturing these dynamics, is an immensely challenging problem. For all but the simplest analysis techniques it is necessary to work in a lower dimensional space than the hundreds of thousands of voxels in a typical rfMRI data set. This is typically achieved either by extracting parcels from an anatomical atlas, or using highdimensional sICA (Kiviniemi et al, 2009; Smith et al, 2013a). It is well known that “[i]nconsistent or imprecise node definitions can have a major impact on subsequent analyses” (Fornito et al, 2013), which again throws the question of how best to generate meaningful representations of resting state activity into sharp relief

Objectives

Methods

Results

Conclusion