Robust Graph Analysis of Brain Connectivity

Vision of the project

A broad range of neurological diseases and disorders have been linked with pathological alterations in the connectivity of the brain. However, robust methods for identifying and characterising abnormalities in connectivity, their evolution over time and response to treatments, are lacking; and addressing this omission represents a pressing clinical need. To establish a clear baseline, we will characterise the typical “normal” connectivity network, and model statistically its variability in a healthy population. We will use magnetic resonance imaging (MRI) and electroencephalography (EEG) to obtain in vivo connectivity information. This work will provide major novel tools for graph analysis of brain connectivity, thereby driving connectivity network methods towards reliable and routine clinical application.



Functional connectomes are abstract representations of how one region of the brain interacts with another and hold promise for understanding human brain activity across a range of cognitive, developmental and pathological states. We construct functional connectomes from simultaneous acquisitions of fMRI and EEG data during rest. fMRI measures the haemodynamic brain activity and thus it is bound to a low temporal resolution, with spatial resolution in the scale of millimetres. EEG measures brain electrical activity from sensors attached on the scalp. It provides high temporal resolution but it is the result of mixing signals from multiple brain sources. By utilising simultaneous EEG-fMRI we have the opportunity to describe brain network dynamics more accurately. This requires understanding the link, between these multi-modal measurements, that underlines their common neurophysiological basis.

Towards this end we searched for electrophysiological ‘signatures’ in distinct frequency bands that most resemble fMRI connectomes. Examining the EEG signal in these frequency bands is important because they are related to fundamental brain rhythms associated with characteristic sensory, motor or cognitive events. Our work is significant because we have developed a state-of-the art approach based on machine learning and statistics to compare functional connectomes derived from different methods and extract the link between them. Based on this inference approach we can predict a subject-specific fMRI connectome from the corresponding EEG connectome and vice-versa. A quantitative measure of distance between connectomes is adapted to evaluate how close the prediction is to the actual measurement. This provides us with a principled way to search for models that encode better the relationship between the two modalitites. We are the first to investigate the relationship between synchronous fMRI and EEG connectomes in a whole-brain, using source space analysis. Source space analysis allows mapping the signal from the EEG sensors to the brain sources, which facilitates a direct interpretation of the results and it is important in medical applications. Towards this end we also exploit non-parametric statistics to provide a control of false positive connections highlighted as the most related across-modalities.

We showed that connectomes derived in low frequency EEG bands most closely resemble rs-fMRI connectomes and that there are signatures of rs-fMRI dynamics across EEG frequencies. This is consistent with the concept of cross-spectral coupling, which is considered a complex communication mechanism between neuronal populations. We also showed that the performance of predicting fMRI from EEG connectomes is considerably better than predicting EEG from fMRI across all frequency bands. This finding has important implications. Although, the spatial resolution of mapping the EEG signal is in the scale of 1-2cm, most fMRI network analysis studies involve averaging the hemodynamic signal within larger regions. Our results indicate that in this spatial resolution the information carried in the EEG signal is richer than the averaged hemodynamic activity. This implies that scalp EEG can be used to provide similar information to resting state fMRI based cortical connectomes at substantially reduced cost while providing much greater dynamic information content.


  • C.S. Parker, F. Deligianni, M.J. Cardoso, P. Daga, M. Modat, M. Dayan, C.A. Clark and S. Ourseling, Consensus between pipelines in structural brain networks, PLoS ONE, 9(10), 2014.

  • F. Deligianni, M. Centeno, D.W. Carmichael and J.D. Clayden, Relating resting-state fMRI and EEG whole-brain connectomes across frequency bands, Frontiers in Neuroscience, 8(258), 2014.
    This article was at the top ten most viewed in Frontiers Neuroscience articles.

  • F. Deligianni, D.W. Carmichael, C.A. Clark and J.D. Clayden, NODDI and Tensor-based Microstructural Indices as Predictors of Functional Connectivity, ISMRM British Chapter, 2015. (ORAL PRESENTATION)

  • F. Deligianni, D.W. Carmichael, C.A. Clark and J.D. Clayden, A prediction framework of functional from structural connectomes reveals relationships between NODDI and tensor-based micro-structural indices, Symposium on Big Data Initiatives for Connectomics Research, International conference on Brain Informatics and Health, 2015. (ORAL PRESENTATION)

  • F. Deligianni, C.A. Clark and J.D. Clayden, Prediction of functional from structural connectomes across micro-structural indices, ISMRM British Chapter, 2014. (ORAL PRESENTATION)

  • F. Deligianni, M. Centeno, D.W. Carmichael and J.D. Clayden, Relating resting-state fMRI and EEG brain connectivity across frequency bands, ISMRM, 2014. (ORAL PRESENTATION) (OHBM14 pdf)

  • F. Deligianni, C.A. Clark and J.D. Clayden, Evaluating structural brain networks based on their performance in predicting functional connectivity, ISMRM, 2014.

  • C.S. Parker, F. Deligianni, M.J. Cardoso, P. Daga, M. Modat, C.A. Clark, S. Ourselin, J.D. Clayden, Consensus between pipelines in whole brain structural connectivity networks, ISMRM, 2014. (ORAL PRESENTATION)

  • F. Deligianni, C.A. Clark, and J.D. Clayden, A Framework to Compare Tractography Algorithms Based on their Performance in Predicting Functional Networks, MICCAI-MBIA, 2013. (ORAL PRESENTATION) pdf

  • F. Deligianni, M. Centeno, D.W. Carmichael and J.D. Clayden, Quantitative Agreement between fMRI and EEG Brain Connectivity Matrices in Different Frequency Bands, BaCI, 2013.

Source Code

GraphPredictiveModels Github Repository