Data-driven Re-Referencing of Intracranial EEG Based on Independent Component Analysis (ICA)

Intracranial recordings from patients implanted with depth electrodes are a valuable source of information in cognitive neuroscience. They allow for the unique opportunity to record brain activity with a high spatial and temporal resolution. To extract the local signal of interest in stereotactic EEG (S-EEG) data, a common pre-processing choice is to re-reference the data with a bipolar montage. With bipolar reference, each channel is subtracted from its neighbour in order to reduce commonalities between channels and isolate activity that is spatially confined. We here challenge the assumption that bipolar reference can effectively perform this task. We argue that in order to extract local activity, the distribution of the signal source of interest, as well as the distribution of interfering distant signals and noise sources need to be considered. Those can have a variable spatial extent and are modulated by electrode spacing, location and anatomical characteristics. Those factors are not accounted for by a fixed referencing scheme and bipolar reference can therefore not only decrease the signal to noise ratio (SNR) of the data, but also lead to mislocalization of activity and consequently to misinterpretation of results. We promote the perspective of regarding referencing as a spatial filtering operation with fixed coefficients. As an alternative, we propose to use Independent Component Analysis (ICA), to derive filter coefficients that reflect the statistical dependencies of the data at hand. We argue that ICA performs the same task that bipolar referencing pursues, namely undoing the linear superposition of activity and can therefore be used to identify activity that is local. We first describe and demonstrate this procedure on human S-EEG recordings. In a simulation with real data, we then quantitatively show that ICA outperforms the bipolar referencing operation in sensitivity and importantly in specificity when it comes to revealing local time series from the superposition of neighbouring channels.