The “Hub Disruption Index,” a Reliable Index Sensitive to the Brain Networks Reorganization. A Study of the Contralesional Hemisphere in Stroke

Your doctor should be able to explain how the good side of your brain is going to help you recover and the protocols necessary to accomplish that. But that is a sick joke since your doctor has no clue how to do that.
http://journal.frontiersin.org/article/10.3389/fncom.2016.00084/full?

Maite Termenon^1,2*, Sophie Achard^3,4, Assia Jaillard^5,6,7 and Chantal Delon-Martin^1,2

• ¹Grenoble Institut des Neurosciences, Université Grenoble Alpes, Grenoble, France
• ²Institut National de la Santé et de la Recherche Médicale, U1216, Grenoble, France
• ³GIPSA-Lab, Université Grenoble Alpes, Grenoble, France
• ⁴GIPSA-Lab, Centre National de la Recherche Scientifique, Grenoble, France
• ⁵Centre Hospitalier Universitaire (CHU) de Grenoble, Grenoble, France
• ⁶Pole Recherche, Centre Hospitalier Universitaire (CHU) Grenoble, Grenoble, France
• ⁷IRMaGe, Institut National de la Santé et de la Recherche Médicale US17 Centre National de la Recherche Scientifique UMS 3552, Grenoble, France

Stroke, resulting in focal structural damage, induces changes in brain function at both local and global levels. Following stroke, cerebral networks present structural, and functional reorganization to compensate for the dysfunctioning provoked by the lesion itself and its remote effects. As some recent studies underlined the role of the contralesional hemisphere during recovery, we studied its role in the reorganization of brain function of stroke patients using resting state fMRI and graph theory. We explored this reorganization using the “hub disruption index” (κ), a global index sensitive to the reorganization of nodes within the graph. For a given graph metric, κ of a subject corresponds to the slope of the linear regression model between the mean local network measures of a reference group, and the difference between that reference and the subject under study. In order to translate the use of κ in clinical context, a prerequisite to achieve meaningful results is to investigate the reliability of this index. In a preliminary part, we studied the reliability of κ by computing the intraclass correlation coefficient in a cohort of 100 subjects from the Human Connectome Project. Then, we measured intra-hemispheric κ index in the contralesional hemisphere of 20 subacute stroke patients compared to 20 age-matched healthy controls. Finally, due to the small number of patients, we tested the robustness of our results repeating the experiment 1000 times by bootstrapping on the Human Connectome Project database. Statistical analysis showed a significant reduction of κ for the contralesional hemisphere of right stroke patients compared to healthy controls. Similar results were observed for the right contralesional hemisphere of left stroke patients. We showed that κ, is more reliable than global graph metrics and more sensitive to detect differences between groups of patients as compared to healthy controls. Using new graph metrics as κ allows us to show that stroke induces a network-wide pattern of reorganization in the contralesional hemisphere whatever the side of the lesion. Graph modeling combined with measure of reorganization at the level of large-scale networks can become a useful tool in clinic.

1. Introduction

In numerous neurological conditions, the adult central nervous system retains an impressive capacity to recover and adapt following injury. Such so-called spontaneous recovery occurs after spinal cord injury, traumatic brain injury, and stroke. Therefore, a basic understanding of the mechanisms that underlie spontaneous recovery of function is the initial step in the development of modulatory therapies that may improve recovery rates and endpoints (Nudo, 2013). In acute stroke, it has been shown that initial damage disrupts communication in distributed brain networks. This initial disorganization is followed by a dynamic reorganization at subacute and chronic stage that may determine the level of post-stroke recovery (Carter et al., 2012). Not only disorganization in structural connectivity has been reported and related to outcome of patients (Moulton et al., 2015) but also functional reorganization in the motor network of both ipsilesional and contralesional hemispheres (Loubinoux et al., 2003; Jaillard et al., 2005; Gerloff et al., 2006; Favre et al., 2014) to compensate for the lesion itself and for remote effects (see Grefkes and Fink, 2014 for a review). The role of the contralesional hemisphere in the recovery process after stroke is supported by several studies using task fMRI paradigms (Gerloff et al., 2006; Lotze et al., 2006; Riecker et al., 2010; Rehme et al., 2011; Teki et al., 2013; Grefkes and Fink, 2014) but it has not been studied before as an independent network (without taking into account the interhemispheric connectivity) of the brain. It is thus of clinical interest to study the reorganization of the contralesional hemisphere in stroke patients by means of functional connectivity fMRI at rest.

In the recent years, there has been a great amount of work developing new investigation methods of the brain connectivity based on fMRI. Among those, the graph theoretical approach seems particularly useful in the context of pathology since it underlines the role of key communicating regions (hubs) in the graph. Since there was no graph metric aiming at capturing this type of reorganization after brain damage, the Hub Disruption Index (κ) was introduced in Achard et al. (2012) to capture it. κ index summarizes graph metric changes at the nodal level in a single value. It is thus a global index capturing changes at the nodal level. For a given graph metric, κ is computed as the slope of the linear regression model between the mean nodal metric value of a reference group and the differential nodal metric value between a given subject (patient or control) and that reference (see Figure 1 for a graphical explanation). If the subject's nodal values are close to those of the reference group (Figure 1C), the κ will be close to 0. Contrary, if the subject's nodal values are different from those of the reference group (Figure 1D), with reduced values in nodes with high metric values in the reference group, the κ will be negative. Once the reference group is computed, the κ can be calculated for each control and each patient individually and statistical tests can be applied to compare the differences between groups.

FIGURE 1

Figure 1. Estimation of κ. The nodal network topology (here, node degree) of an individual subject in relation to the normative network topology of the healthy control group (A) for one healthy volunteer and (B) for one stroke patient. To construct the hub disruption index κ for the degree, we subtract the healthy group mean nodal degree from the degree of the corresponding node in an individual subject before plotting this individual difference against the healthy group mean. κ is the slope of the regression line computed on this scatter plot. This transformation means that the data for an individual healthy volunteer (C) will be scattered around a horizontal line (κ~0), whereas the data for a patient in a stroke (D) will be scattered around a negatively sloping line (κ < 0).

According to Bullmore and Sporns (2009), hubs are crucial nodes for an efficient communication in the network and are identified as nodes with high degree or high centrality values. In this paper, we computed κ using metrics that directly relate to hubs: node degree, betweenness centrality and global efficiency; and also in metrics that explore the neighborhood of the node, such as, local efficiency and clustering coefficient.

The aim of this paper is to quantify the impact of the lesion on the brain network reorganization of the contralesional hemisphere in severe stroke patients at subacute stage. For this purpose, κ index is a perfect tool to assess such reorganization by comparing nodal metrics between healthy volunteers and patients. In order to translate the use of κ in clinical context, an essential requirement to achieve meaningful results is to investigate the reliability of this index. For this purpose, we used the intraclass correlation coefficient (ICC), as it was previously assessed in several studies working with brain graphs reliability in rs-fMRI (Schwarz and McGonigle, 2011; Wang et al., 2011; Braun et al., 2012; Guo et al., 2012; Liang et al., 2012; Cao et al., 2014).

This paper is divided into three parts: in the first part, we assessed the reliability of κ, over different graph metrics, by computing the ICC in a cohort of 100 healthy subjects using the database from the Human Connectome Project (HCP)¹. We calculated the ICCs and their p-values, applying bootstrap and permutation techniques to check for the influence of the number of subjects and of the number of edges (cost) in brain graphs. We also explored whether there is a laterality effect by testing the graphs of the intra-hemispheric connectivity from the left and from the right hemispheres in healthy control subjects using the HCP dataset. In the second part of the paper, we used the κ index to study the reorganization that occurs in the contralesional hemisphere of 20 severe subacute stroke patients. Finally, in the third part, we tested the robustness of the results obtained in this clinical study by randomly choosing 20 subjects as “patients” and 20 subjects as “controls” from the HCP database, computing the difference in κ between them and replicating 1000 times this procedure.