Table 2.

Common analysis methods to study functional connectivity in focal epilepsy

A) Model-based methods	Summary	Considerations
coherence	measures the degree of linear dependency of signals between brain regions by examining similar frequency components	requires selection of a specific frequency band, risking overlooking dependencies in other bands; frequency-specific analysis may be more likely to uncover true, not spurious, neural connections
cross-correlation analysis	measures degree of similarity of signals amongst brain regions as a function of the time lag between these signals	does not require frequency band selection; correlation measurements at zero-time lag are more likely to detect artifact or shared source connections
phase synchrony	compares only the phase of neural oscillations between brain regions to estimate cross-talk	time resolved, with better temporal resolution than coherence; only sensitive to signal phase and not amplitude
model-based methods, overall	non-exploratory, often seed-based methods built upon prior hypotheses	less likely to uncover spurious relationships than data-driven methods, but necessitates prior knowledge for specific hypothesis testing
B) Data-driven methods	Summary	Considerations
principle component analysis (PCA)	uses orthogonal transformation to convert possibly correlated neural signals into a set of uncorrelated signals (principal components) in order to uncover network patterns	may reveal novel network patterns and interdependencies; requires selection of regions of interest given otherwise unmanageable amount of data; may be preprocessing step for ICA for further analysis
independent component analysis (ICA)	identifies a linear combination of components (brain region signals) like PCA, but is restricted to those that are statistically independent	can identify statically independent neural sources (unlike PCA), but statistical dependence between neural signals would degrade results; can be used to further interrogate components identified in PCA
clustering analysis	identifies groups of brain regions with similar neural signals (clusters) based on the time courses of those signals	useful for novel identification of neural networks and included brain regions; the number of clusters must be chosen, and this may affect connectivity results
data-driven methods, overall	exploratory, non-seed-based analysis methods free of prior hypotheses	may uncover novel networks given lack of prior knowledge, but more likely to uncover spurious relationships