Statistical methods for the brain functional connectivity networks, fusion with anatomical connectivity. Towards a new diagnostic and prognostic tool for the evaluation of consciousness disorders. – InfoNetComaBrain

The five methodological objectives detailed in this project are : (1) The
analysis of sensitivity of the connectivity graphs to physiological noise and pre-processing. This preliminary work is a
mandatory step as the method has never been validated with respect to these sources of error. (2) The definition of a
model of multivariate fractal time series in order to analyse the brain data as a whole rather than pairwise. (3) The
derivation of nonlinear, conditional dependence measures allowing to construct connectivity graphs with enhanced
robustness as compared with correlation. (4) The analysis of graph : parameter selection, group comparisons, time
evolution. (5) The handling of multimodality allowing the fusion of anatomical and functional brain networks. From a
clinical point of view, the objectives are : (1) Aid in differential diagnosis. This is based on the quantification and
qualification of networks of propagating information in brain for different states of unconsciousness as recognized by
clinical assessment. The analysis and visualization of the information networks as graphs will promote the necessary
interactions with the neuroscientists and clinicians. (2) Prognostic assessment and identification of pathophysiologic
mechanisms. This will be based on a longitudinal study of a cohort of patients. The features of changes of information
networks prior to recovery will be identified. (3) Awareness mechanism. The findings on information networks in patients
with consciousness disorders along with comparisons with other state of unconsciousness like sleep and anesthesia will
impact the understanding of mechanism of awareness.

As pointed out above, it is a challenging task to relate patterns of activation to a measure of awareness and
the spontaneous fluctuations in brain activity observed at rest seem likely to reflect changes in the case of
disorders of consciousness. The novel approach proposed here consists in characterizing the propagation of
information in the brain at rest (without any stimulation) and will open up new horizons in the understanding of
the mechanism of awareness that have been until now left unexplored. This unconventional approach based
on multivariate fractal models and the fusion of DTI and fMRI connectivity networks will allow individual
analyses which is crucial for the study of patients with consciousness disorders. This approach is promising
new improvements in the understanding of the pathophysiology of brain injury and also in the understanding
of the complex architecture of the brain. Moreover, the results obtained during this project will have direct
implications on the care of patients with consciousness disorders. The four main achievements of this project
will be : (1) Functional and structural characterisation of whole brain networks using fMRI and DTI data (2)
Longitudinal studies of consciousness disorders patients based on novel efficient statistical tests on
multivariate fractal models (3) Highly robust analysis of fMRI data by means of brain connectivity networks
(4) Potential new criteria for diagnosis and prognosis for consciousness disorders.

After 18 monts of work on this project, we are
planning to analyse the longitudinal data
acquired recently on 6 patients, one year after
the accident. The PET data are also available for
each patient, our objectif is to combine these
data to the connectivity graphs in order to relate
metabolism to functional connectivity.

Results will be disseminated through publications, participation to conferences
in the different communities : statistics, signal processing and neuroscience. We will consider participating to
conference dedicated to medical practice in order to get close contact with the medical practitioners. Also, a
website devoted to the project will be created to make freely avalaible some of the data and softwares. A
particular attention will be made to present the results to a large audience, especially to medical practitioners.
The software developments of task 3, 4 and 5 will be included in an existing software brainwaver which is a
package for the open-source, free software R. The generic algorithms that may be developed during the
course of the project (such as implementation of various dependence measures) will be integrated as a
module of this software and made freely avalaible to the wider scientific community. The more specific user
codes that implement the fusion described in task 5 will be retained as an in-house software for use in this
and future projects.

Studies of the mechanisms in the brain that allow complex neuronal activity to arise in a coordinated fashion have produced some of the most spectacular discoveries in neuroscience, and still promise huge potential understanding as novel technological and especially methodological tools are developped.

Neurophysiology has allowed to understand small-scale networks of neurons, and the study of brain lesions has identified local brain
areas specifically associated with a certain function. Non-invasive brain imaging techniques, such as electro- or magneto-encephalography (EEG and MEG) and functional magnetic resonance imaging (fMRI) have brought brain research an incredible amount of a novel type of data, that is, multivariate time series representing local dynamics at each of multiple sites or sources throughout the whole human brain while functioning. These data have already demonstrated that neurophysiological processes typically have long memory or fractal scaling properties at a univariate level of analysis and demonstrate complex network topological organization at a multivariate level of analysis using wavelet correlations. Compared to estimators employed previously, this wavelet correlation is well-behaved for long memory and nonstationary processes. These findings suggest that brain dynamics and networks may have important statistical properties in common with other, substantively diverse, complex systems that are currently the focus of exciting activity in the general field of statistical physics.

This project is focusing on the study of consciousness disorders (a medical state that follows coma) for which
understanding the disconnexion process of the brain is crucial to improve everyday management for these patients. Consciousness disorders is a major concern in public health. Due to the progress in intensive care, more and more patients will survive severe acute brain damage. Disorders of consciousness can be acute and reversible or they can be irreversible and permanent.

In the context of consciousness disorders, the aim of this project is to characterize multivariate neurophysiological datasets to elucidate
the biological basis for information processing and propragation in the human brain. Neuroimagery provides us with time series, associated with either voxels or sensors, which correspond to the information processed by a brain area in time. These data have brought to light the dynamic nature of the brain, which follows complex time patterns, a subject which has been at the center of neuroscience for a century; but so far, the space-time patterns have remained nearly unexplored. This is why the revolution of understanding that they can bring is yet to come, as what they reveal of the circulation of the information in the brain and its dynamics has not been yet successfully analysed.

Participating in this novel understanding, this project will open up new directions to help clinicians for the diagnosis and to give better predictions concerning the possible recovery of these patients. This will also bring new opportunities to understand the spontaneous fluctuations in brain activity observed on resting state healthy volunteers and the nature of consciousness. More generally, after Geschwind's pioneer work in neurology and psychiatry, pathologies could be described in terms of disconnection syndromes. Thus, the approach described here
might be extended to other pathologies such as multiple sclerosis, epilepsy, stroke, Alzheimer disease, schizophrenia.

This project requires a very broad skillset, comprising statistical signal processing, complex network analysis and visualization, and systems neuroscience. Thus, the team is composed of experts in medical applications (S. Krémer), neurosciences (C. Delon-Martin), statistics
(S. Achard and J.-F. Coeurjolly) and signal processing (V. Noblet).