Quantifying multiplexing in the brain using functional connectivity embedding
Poster No:
1768
Submission Type:
Abstract Submission
Authors:
Nicole Eichert1, Saad Jbabdi2
Institutions:
1University of Oxford, Oxford, Oxfordshire, 2Oxford University, Oxford, United Kingdom
First Author:
Co-Author:
Introduction:
The cerebral cortex is composed of over one hundred distinct functional regions [1]. However, these areas are not homogenous as they can contain various sources of functional heterogeneity. These different functional patterns can overlap within the same region. E.g., primary visual cortex contains a retinotopic 2D map, but also ocular dominance columns. Dorsal parietal lobe contains retinotopic maps as well as body maps. We posit that overlapping patterns of function relate to overlapping patterns of connections, and reflect the rich array of functions that a brain area participates in [2]. We refer to this as multiplexing [3]. Overlapping patterns may also interact, via short-range connections, to enable more complex functions.
Despite this, the dominant paradigm in macroscopic brain connectivity represents brain areas as single, discrete, and homogenous nodes, and falls short of capturing within-area diversity. Here we introduce a method for characterising the diversity of within-area functional patterns, a.k.a. multiplexing. We use low-dimensional embedding of functional connectivity to extract within-area overlapping patterns of connections from pairwise regional connections. We quantify multiplexing through assessing the diversity of overlapping connectivity patterns. Finally, we show that these overlapping patterns are associated with task-relevant networks.
Despite this, the dominant paradigm in macroscopic brain connectivity represents brain areas as single, discrete, and homogenous nodes, and falls short of capturing within-area diversity. Here we introduce a method for characterising the diversity of within-area functional patterns, a.k.a. multiplexing. We use low-dimensional embedding of functional connectivity to extract within-area overlapping patterns of connections from pairwise regional connections. We quantify multiplexing through assessing the diversity of overlapping connectivity patterns. Finally, we show that these overlapping patterns are associated with task-relevant networks.
Methods:
To extract overlapping patterns of connectivity from a brain area, we consider its pairwise connections with an extended set of other areas throughout the brain. We implemented an extension of spectral embedding to bipartite graphs to identify the dominant pattern of pairwise connectivity (Fig1A). For each pair of parcels, the connectivity matrix is reduced to a 1-dimensional embedding space (Fig1B). Resting-state scans from 100 HCP subjects were used to build the connectivity matrix C at group level. By changing the target area (amongst a set of 316 uniform parcels), we can build up a matrix summarising the dominant spatial modes within a given region. We quantify the diversity of these modes using hierarchical clustering with a range of cut-off thresholds. Utility of the metric was evaluated by simulating mixtures from multivariate distributions whilst systematically varying the number of non-overlapping clusters (Fig2B). We derived the diversity metric for each parcel and averaged the result across 5 random parcellations, produced by randomly rotating the original parcellation (Fig2C). Finally, we ask whether the spatial modes are associated with brain activity and accessed 300 task-fMRI maps from neurovault [4] (Fig2D,E). Since the spatial modes are related to the connectivity of a brain area with a set of external regions throughout the cortex, we hypothesise that evoked brain activity within a region would reflect this association.
·Fig 1. Connectivity embedding.
Results:
The clustering-based diversity metric robustly scales with the number of true clusters, independent of the parcellation resolution (Fig2B). Across the cortex, highest diversity in spatial topographies was observed in the insula, the medial frontal lobe and the medial temporal lobe. These regions are often considered part of the multiple-demand network (Fig2C). We found significant assoications between spatial modes within a region and their corresponding external regions during task (Fig2E). The association between spatial modes and task activations is robust across the whole cortex.
·Fig 2. Quantifying multiplexing.
Conclusions:
We present an algorithm to define and measure regional spatial topographies across the cortex using connectivity embedding. We developed a framework to detect topographic diversity from resting-state data, which serves as a marker for the multiplexing ability of a brain region. We showed that topographically organised connectivity is abundant in the brain and relevant for task activity and that multiplexing is highest in the multiple-demand network.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Segmentation and Parcellation
Task-Independent and Resting-State Analysis 2
Keywords:
Cognition
Cortex
FUNCTIONAL MRI
Workflows
1|2Indicates the priority used for review
Provide references using author date format
[2] Jbabdi, S., Haak, K. & Behrens, T. Separating functional modes using spectral methods applied to resting-state FMRI. in OHBM 2013 Conference proceedings.
[3] Jbabdi, S., Sotiropoulos, S. N. & Behrens, T. E. The topographic connectome. Curr. Opin. Neurobiol. 23, 207–215 (2013).
[4] Gorgolewski, K. J. et al. NeuroVault.org: a web-based repository for collecting and sharing unthresholded statistical maps of the human brain. Front. Neuroinform. 9, 8 (2015).