Concepts of Connectivity
Mara Cercignani1

1Brighton & Sussex Medical School, United Kingdom

Synopsis

This paper will introduce the concept of connectivity and how it can be measured using MRI data. Potential pitfalls will also be reviewed

Target Audience

Scientists and Clinicians willing to apply concepts and measures of brain connectivity to the study of the healthy and diseased brain. The Audience will be expected to be familiar with basic concepts of diffusion MRI.

Outcomes and Objectives

This talk aims at clarifying the differing meanings of the term “connectivity”, and how to measure it. It will also described some of the basic concepts of graph theory, and what are the main pitfalls and limitations associated with MRI-based measures of connectivity.

Background

Connectivity: In recent years the concept of connectivity has become increasingly central to Neuroscience and related fields, with the effort of mapping brain circuits and structures often being the focus of big research initiatives. When applied to the brain, the term connectivity refers to several different and interrelated aspects of brain organization [1]. It can be defined as a pattern of anatomical links ("anatomical/structural connectivity"), of statistical dependencies ("functional connectivity") or of causal interactions ("effective connectivity") between distinct units within a nervous system. Brain connectivity can be described at several levels of scale: the microscale, with individual neurons linked by individual synaptic connections; the mesoscale, with networks connecting neuronal populations, and the macroscale level, characterized by fiber pathways connecting brain regions [2]. During this talk we will focus on the macroscale and on structural connectivity although some concepts can be applied to other types of connectivity.

The importance of connectivity in the clinical context: It is becoming increasingly recognized that many behavioral manifestations of neurological and psychiatric diseases are not solely the result of abnormality in one isolated region but represent alterations in brain networks and connectivity [3]. Thus, the improved characterization of brain networks can have an enormous relevance in discovering the basis of common disorders of the brain, response to recovery from brain injury, individual differences, heritability, normal development and aging.

How to measure connectivity in humans

Although other approaches are possible, the term structural connectivity has relied predominantly on diffusion MRI (dMRI) techniques, such as Diffusion Tensor Imaging (DTI), which are able to characterize the main white matter tracts of the brain in vivo. Thanks to its ability to infer tissue orientation and thus to enable the reconstruction of the main white matter tracts with tractography [4], dMRI provides a unique insight into physical connections between anatomically segregated areas of the brain. In order to develop appropriate tools to measure connectivity one should first define precisely what a quantification of structural connectivity should express. Depending on the context, “decreased/increased connectivity” is meant to indicate differing concepts. It may indicate a reduced/increased number of connections, thinner/thicker connections, or less/more efficient connections. It is not always clear which of this aspects of connectivity is the most relevant. Since the early days of diffusion, researchers have attempted to interpret changes to indices derived from DTI, such as fractional anisotropy (FA), as measures of connectivity. However, it should be remembered that FA is affected by many factors that do not reflect any of the properties of connections that we contemplate under the connectivity umbrella. As tractography reconstructs streamlines which are believed to reflect existing white matter pathways, it is an ideal candidate to provide information about physical connections in the brain. Early attempts to build quantitative indices of connectivity from tractography include anatomical connectivity mapping [5] and track density imaging [6]. More recently, however, a popular approach is the one at the intersection between neuroscience and the rising field of network science and graph theory. In order to apply these concepts to the study of the brain, it is possible to use a topological representation where the entities of a network are modelled as nodes and their interactions as edges, reducing the problem to the analysis of a mathematical graph.

How to build a graph

In order to build a graph from a set of dMRI data, one first has to define the nodes. An obvious choice as a node would be the smallest measurable unit, i.e. a voxel in the case of MRI. Although offering such a unique definition, this method has limits. In the case of MRI, each voxel can contain up to 100000 neurons, and it’s not possible to assess if such neurons form a meaningful unit or not [7]. A common node definition method relies instead on the parcellation of the brain gray matter by means of atlases and then the estimation of structural or functional connectivity between each pair of parcellated regions [8]. As for nodes, also defining edges is not univocal. Counting the number of the tracts (NOS) between each possible pair of nodes, (i.e., the corresponding grey matter regions) results in building the weighted connectivity matrix and representing the structural network [9]. The number of reconstructed streamlines, however, is affected by several factors that do not reflect connectivity (e.g., the shape/geometry of the white matter tracts, orientational dispersion, noise, and distance); alternative approaches, while still relying on tractography to reconstruct the connections, use other indices to weight the edges of the graphs. Such indices can be other quantities derived from dMRI (E.g. FA), or be derived from other modalities, believe for example, to reflect the degree of myelination of the specified connections. The rationale is that myelin enables fast signal transmission, and therefore has a role in supporting connectivity.

Concepts from graph theory

Any complex system of any nature (biological, technological, social) can be represented as a network, composed of its basic elements and their interactions. From a mathematical point of view, such representation is equivalent to a graph G, defined as a pair of sets G=(V,E), where V is a set of elements called nodes or vertices, and E is a set of pairs of different nodes called edges or links [10]. An edge (i,j) connects two nodes. The graph is directed if the set of edges is ordered, and undirected if such set is unordered, reflecting the presence or absence of directionality in the modelled relationships. Moreover, if there is a weight wi,j associated with each edge (i,j) the graph is defined as weighted, and it models further properties of the edges themselves. In the opposite case, the graph is unweighted or binary, representing the presence or absence of connections. This structure can be conveniently defined by means of a matrix representation [11], the so-called adjacency matrix. A number of graph measures that can be used to characterize the network. Broadly they can be classified as measures of either integration or segregation.

Limitations and Pitfalls

Extracting meaningful data from dMRI can be challenging [12] and the available options for acquisition parameters, data preprocessing, diffusion model to fit and tractography algorithm are so numerous to make an informed choice difficult. Data quality is ultimately the most important factor in determining the output, and caution should be exercised when interpreting graphs derived from sub-optimal datasets. Tractography is known to be prone to be both false negatives and false positives [13]. When weighting the connections using an MRI proxy of tissue microstructure, the potential errors and limitations associated with the specific MRI modality must also be considered. In this case the main danger in the overinterpretation of such indices in terms of underlying biology. A different problem is related to the use of network modelling itself on brain structure and function. A broad set of studies has applied graph theory concepts to brain analysis limiting to comparison of conventional metrics used in collateral fields [14]. Despite finding significant and potentially useful results, the interpretation issue has not received the right attention and such studies did forget about the context.

Acknowledgements

No acknowledgement found.

References

1. Horwitz, B. The elusive concept of brain connectivity. Neuroimage 2003: 19, 466-470.

2. Sporns, O., Honey, C. J., and Kotter, R. Identification and classification of hubs in brain networks. PLoS One 2007: 2, e1049.

3. Fox, M. D., Halko, M. A., Eldaief, M. C., and Pascual-Leone, A. Measuring and manipulating brain connectivity with resting state functional connectivity magnetic resonance imaging (fcMRI) and transcranial magnetic stimulation (TMS). Neuroimage 2012: 62, 2232-2243.

4. Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aldroubi A: In vivo fiber tractography using DT-MRI data. Magnet Reson Med 2000, 44(4):625-632.

5. Bozzali M, Parker GJM, Serra L, Embleton K, Gili T, Perri R, Caltagirone C, Cercignani M. Anatomical connectivity mapping: a new tool to assess brain disconnection in Alzheimer's disease. Neuroimage 2011; 54: 2045-2051.

6. Calamante F, Tournier JD, Jackson GD, Connelly A. Track-density imaging (TDI): super-resolution white matter imaging using whole-brain track-density mapping. Neuroimage 2010; 53: 1233-1243.

7. Van den Heuvel M, Hulshoffpol HE, Pol HEH: Exploring the brain network: A review on resting-state fMRI functional connectivity. Eur Neuropsychopharm 2010, 20(8):519-534

8. Fornito A, Bullmore E, Zalesky A: Fundamnetals of brain network analysis. Amsterdam: Elsevier; 2016.

9. Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen V, Sporns O: Mapping the structural core of human cerebral cortex. Plos Biol 2008, 6(7):1479-1493.

10. Borner K, Sanyal S, Vespignani A: Network science. Annu Rev Inform Sci 2007, 41:537-607.

11. Costa LD, Rodrigues FA, Travieso G, Boas PRV: Characterization of complex networks: A survey of measurements. Adv Phys 2007, 56(1):167-242.

12. Jones DK, Cercignani M: Twenty-five Pitfalls in the Analysis of Diffusion MRI Data. Nmr Biomed 2010, 23(7):803-820.

13. Johansen-Berg H, Behrens TEJ: Just pretty pictures? What diffusion tractography can add in clinical neuroscience. Curr Opin Neurol 2006, 19(4):379-385.

14. Raj A: Graph Models of Brain Diseases. I S Biomed Imaging 2015:1552-1555.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)