INTRODUCTION

In the last decades, the study of brain functional connectivity (FC) from multivariate neuroimaging data has become an increasingly active field of research providing novel insights into normal brain functions and their disruption in brain pathologies. Network modeling and analysis applied in this field enable the study of functional connections by extracting significant aspects of network organization. In this work we build on the classical functional network construction, based on the association of vertices of a graph with brain regions where weights on the edges represent the connectivity measures among regions, and propose a new modelling paradigm exploiting the concept of 'topic' borrowed from social networks. Basically, tasks of functional magnetic resonance imaging (fMRI) sequences are mapped onto topics of the semantic analysis. On this basis, the regions that are involved in a given task can be rapidly identified, since they exhibit a greater centrality related to that topic1. Moreover, a new measure, called 'transtopic centrality' (TC), is introduced as the generalization of classical centrality to an ensemble of tasks as opposed to the single task. Doing so, TC provides a measure of the engagement of the region across different topics which is an indirect indication of its functional polymorphism.

METHODS

The proposed method involves the usage of labelled graphs, in which a label is added to each vertex to express a quantification of 'depth' relative to a set of topics. Each label is a vector, whose components are the depths over the single topic of the region represented by the vertex (Figure 1). In social network analysis, the depth over a topic is the strength of the signal in a certain activity. The TC of a node is defined as the weighted sum of the semantic closeness centrality (SCC) across all topics 1. In the transposition to the functional imaging network, topics map to tasks and depths to connectivity measures, i.e. the weights of the functional links. Since these links can take both positive and negative values, we decided to split the two cases and separately focus on the positively and negatively determined networks, respectively. The model was tested on one dataset from the Human Connectome Project2 composed of structural and functional scans of eight conditions that are resting state condition, emotion processing, gambling, language processing, motor, relational processing, social cognition (Theory of Mind), and working memory. The image acquisition, protocols and data preprocessing are described in detail in 3; the brain parcellation and the connectivity matrices were obtained as in 4. Each task corresponds to a topic in our model and the TC was measured as the weighted sum of the SCC of each topic. In this first attempt all the topics were equally weighted.

RESULTS

The ranking of the SCC for 15 regions of interest (ROIs) with the highest centrality out of 139 is reported in Table 1. It is worth noting that some brain regions persist within the top fifteen, but at different positions, while there are regions that selectively appear only on certain topics but not in the others. TC highlighted an extensive activation of visual regions in most of the tasks, which is not surprising given the use of visual stimuli in the protocol. The emerging pattern revealed that the common regions between tasks involved also the frontal and parietal lateral cortex, probably associated with the well-established resting state networks (Figures 2 and 3). As can be derived from the above, if a node presents a high SSC in a given task and a high TC value, than that node will also be important/central for all the other topics or tasks.

DISCUSSION

Different brain regions exhibit different centrality values depending on the topic. These preliminary results show how the TC can be an efficient instrument for deriving a functional-based ranking of ROIs. The results suggest that moving from SCC to TC provides a wider perspective view of the communication patterns between brain regions and network organization and, most importantly, allows identifying those ROIs that are prominent across tasks, which has a high potential as predictor of the global functional impairment that could derive from the damaging of the corresponding brain areas.

CONCLUSION

In this work a new metric of network centrality has been introduced enabling the characterization of the role of individual nodes in brain networks across a spectrum of human behaviour. We presented preliminary experimental data, that, by some examples of these connections, show that the proposed model can be effective in enhancing the understanding of the shared brain functioning during different tasks or resting state conditions.

Acknowledgements

No acknowledgement found.