A central concept to study the integrity of functional networks is functional connectivity. In clinical neuroscience, this concept denotes statistical associations between physiologic recordings of different brain areas (Aertsen et al., 1989). In a comprehensive review, Varela and colleagues stated that the extent to which different brain areas are functionally connected depends on the level of synchronous temporal activity, irrespective of signal amplitude (Varela et al., 2001), also called “synchronization.” Functional connectivity can be studied in a task-related paradigm or in a so-called resting-state condition. During a resting-state condition, the subject is in a nonactive, awake state and task related activity is absent. This condition allows the detection of intrinsic activity of the brain (Greicius, 2008; van den Heuvel & Hulshoff Pol, 2010; Deco et al., 2011).
In the study of epilepsy, neurophysiologic techniques, such as (intracranial) electroencephalography (EEG) and magnetoencephalography (MEG), are widely used to localize epileptiform activity and to provide information on how brain areas are functionally connected. Over the last decades, different methods have been described to determine functional connectivity between brain areas (Pereda et al., 2005; Lemieux et al., 2011; Stefan & Lopes da Silva, 2013). Initially, connectivity studies focused on linear correlations between two signals as a function of the frequency during seizure propagation (Brazier, 1972; Gotman, 1983). Later, complex, nonlinear correlations were introduced to investigate functional coupling between different brain areas (Pijn et al., 1990; Bartolomei et al., 2001). Examples of nonlinear correlation measures currently used in epilepsy studies include the nonlinear correlation coefficient (Wendling et al., 2001); synchronization likelihood, that detects both linear and nonlinear interdependencies between two (time) signals (Stam & Van Dijk, 2002); phase lag index, that overcomes volume conduction as a confounding effect (Stam et al., 2007); granger causality, a method that denotes causality between interacting signals (Bressler & Seth, 2011) and partial directed coherence, an effective connectivity measure that is able to distinguish both direct and indirect causality flows (Baccala & Sameshima, 2001). All of these methods have their unique advantages and deal with specific limitations of neurophysiologic recordings (Pereda et al., 2005; Stam & van Straaten, 2012). Wendling and colleagues performed a modeling study that compared different connectivity measures, and concluded that the ideal method eventually depends on the studied model (Wendling et al., 2009). Furthermore, functional connectivity in neurophysiologic studies is usually analyzed in separate frequency bands: delta band (0–4 Hz), theta band (4–8 Hz), alpha band (8–13 Hz), beta band (13–30 Hz), and gamma band (>30 Hz), as each frequency band is associated with distinct networks and functions (Basar et al., 2001).
Using spontaneous low frequency fluctuations in the blood oxygenation level–dependent (BOLD) signal, functional MRI (fMRI) enables functional connectivity investigations with a higher spatial but lower temporal resolution compared to neurophysiologic recordings. Because BOLD signal changes relate to underlying neural activity, fMRI provides only an indirect measure of functional connectivity. The various methods used in this paradigm include a seed or region-of-interest (ROI) –based approach and the independent component analysis (ICA). The ROI-based approach is used to determine temporal correlations from a selected region (or multiple regions) with all other brain areas. Although this method is relatively easy to apply, it requires an a priori, investigator driven, definition of seed regions and has some statistical limitations (Fox & Raichle, 2007). In contrast to the ROI-based approach, the ICA is a data-driven method, which decomposes the BOLD signal into temporally correlated and spatially independent brain areas (Beckmann et al., 2005).