Comparing re-referencing methods for human intracranial EEG

Stereo-electroencephalography (SEEG) involves the implantation of depth electrodes in the brain to monitor epileptic seizures for surgical feasibility. As the electrodes advance into deeper limbic structures, the macro contacts distributed along the shaft record electrical activity from more lateral sites, while a bundled array of microwires simultaneously captures highly localised signals from the tip of the electrode. In a clinical setting these electrical signals are recorded relative to a reference signal, which may be sub-optimal for subsequent analyses that are unrelated to clinical objectives. Therefore, a secondary re-referencing operation is often undertaken to transform the signal in relation to a specific hypothesis. Despite a recent surge in SEEG studies, comparative studies on the much-understudied issue of re-referencing are sparse. Here, we analyse data collected from 16 human patients during an episodic memory task and comparatively apply a range of re-referencing methods to both macro contact and micro-wire data. We find that the choice of method changes the proportion of trials that survive artefact rejection, low frequency power, and the aperiodic background against which that power is read. In the hippocampal recordings, however, no method differed from any other in the size of a memory effect in power or spike-field coherence. This early transformative decision therefore bears on how the signal is described. We found no evidence that it changes the detectability of the behavioural effects tested here, and no method reversed the direction of a memory effect.