Abstracts

Rationale: Bipolar re-referencing, in which each electrode is subtracted from its consecutive neighboring electrode, is often used to address signal quality and refine spatial localization for intracranial and scalp EEG. However, it is unknown how re-referencing at precise inter-electrode distances influences each individual frequency of the resulting signal. We leveraged human intracranial interictal recordings with uniquely large numbers of electrodes to understand how neural power spectra and spatiotemporal details of interictal spike detection are influenced as a function of Euclidean distance between bipolar pairs.

Methods: Intracranial recordings for drug-resistant epilepsy were performed using combinations of high-density subdural grid, depth, and strip electrodes (n=2556, 498, and 224) which had manufactured linear inter-electrode distances of 4, 5, and 10 mm. Recordings for each patient participant (n=11) were re-referenced using all possible electrode pairs (n=339,795 grid, n=11,377 depth, n=2,602 strip) spanning distances from 2 to 80 mm. A multi-taper power spectrum from 2-200 Hz was generated for 1-second windows across artifact-free sections of data. We used cluster-based permutation tests (CBPTs) to compare spectral composition features after converting distance to 2 mm bins to create a novel map of the percent change in power (natural log) for each frequency at each distance. We also determined how bipolar pair distances influence task-related spectral power changes in the superior temporal gyrus (STG) during passive listening. Lastly, high- and low-density (4 mm vs 8 mm) re-referencing schemes were compared for interictal spike localization using a linelength transform detector and CBPTs.

Results: At bipolar pair distances below 8 mm, power was decreased below 30 Hz (p< 0.001) and non-linearly augmented at higher frequencies. For distances above 8 mm all frequencies were non-linearly enhanced. These dependencies were remarkably consistent across grids, strips, and depths. Task-related increases in STG high gamma (50-200 Hz) activity were enhanced at 4 to 45 mm bipolar distances yet paradoxically decreased above 45 mm (p< 0.05). The detection of epileptiform discharges (duration and number of involved channels) was similar between 4 and 8 mm bipolar re-referencing (p >0.05).

Conclusions: Re-referencing at bipolar distances below 8 mm preferentially attenuates specific low-frequency signals and augments higher frequencies, while larger distances augment all broadband (2-200 Hz) signals in a non-linear pattern, for which we generate a distance-frequency map to aid researchers. Task-related activity is captured at most, but not all, bipolar distances, and standard clinical bipolar distances largely capture the spatiotemporal extent of spikes. These findings offer a blueprint for optimizing the SNR of spectral intracranial signals through strategic re-referencing distance selection, with strong implications for epilepsy surgery, human neurophysiology, and improving biomarker detection for closed-loop neuromodulation.

Funding: This work was supported by NINDS grant K23NS110920 (J.K.K.).