Statistical analysis tool for time/frequency state spaces.

The electroencephalogram, or EEG, consists of fluctuations in brain electrical activity; it can be measured whilst people are doing cognitive tasks (e.g. seeing objects, reading, listening). Recordings consist in time-series and are collected via multiple electrodes at the surface of the skull. Mathematical operations allow decomposing those recordings into more easy interpretable signals to examine, for instance, different frequencies (e.g. slow and fast fluctuations of the signal). There is currently no tool allowing the statistical assessment of all the EEG dimensions simultaneously (all electrodes, all time points and all frequencies together). Our project aims at providing a freely distributed toolbox allowing such analyses, therefore providing a new way to investigate a lot of research data and potentially generating important new knowledge about the brain.

In recent years, new ways to investigate EEG recordings have emerged. In particular, Event Related Spectral Perturbations (ERSP) and Inter-Trial phase Coherence (ITC) have become the standard to characterize the time-frequency decompositions of signals from electrodes or components (e.g. from independent component analysis). Yet, no adequate statistical tools are available to investigate all the dimensions of ERSP and ITC at once: all electrodes/components, time frames, frequencies, amplitude and phase. Our project will provide a freely distributed toolbox allowing such comprehensive analyses. The toolbox will be based on mass-univariate robust statistics and validated multiple comparison correction procedures. This tool will improve significantly data analysis standards. Having such tool will also expand the spectrum of possible experiments, which are currently often limited to binary designs (e.g. condition A vs. condition B), simply because there is no good way to analyse more complex experiments. In turn, this will lead to an increase in our understanding of biological and psychological phenomena. The toolbox will be distributed along with datasets used to validate our methods so that user can familiarise themselves with the new techniques and other research groups can use them to compare methods.

Our new tool will benefit public and private sector researchers using magneto- and electro-encephalography data analysis. We anticipate five main outcomes from our work:
[1] Researchers will not have to limit their analyses to a-priori interesting features. Instead, they will be able to exploit the full data space.
[2] Our tool will effectively control the negative consequences of exploring a full data space: false positive will be minimized without sacrificing power.
[3] Researchers will have the possibility to analyse more complex designs that cannot be managed by current toolboxes. This should enable researchers to use more sophisticated experimental designs and ask new questions about brain activity.
[4] The free availability of datasets from our own simulations will be useful for validation of methods by other research groups or for people to familiarise themselves with the new techniques.
[5] Overall, our work will contribute to the on-going paradigm shift in MEEG: away from simple group analyses of few regions of interests, and towards higher standards, including comprehensive analyses of single-subject and group data.