Distance of bipolar re-referencing imparts nonlinear frequency-specific influences on intracranial recording signal measurements.

Abstract

<i>Obective.</i>Bipolar re-referencing (BPRR), in which one electrode's signal is subtracted from a neighboring electrode to produce a differential signal, can improve signal readability and refine localization for intracranial electroencephalography. There is wide variation in manufactured electrode array spacing, yet how BPRR affects specific frequencies at precise inter-electrode distances has not been systematically evaluated.<i>Approach.</i>Intracranial recordings with uniquely large numbers of electrodes were obtained for sixteen patients with drug-resistant epilepsy. We evaluated combinations of high-density subdural grid, depth, and strip electrodes (<i>n</i>= 3,664, 742, and 336) with manufactured linear inter-electrode distances of 4, 5, and 10 mm, respectively. BPRR was performed using all possible electrode pairs (<i>n</i>= 445 305 grid, 16 004 depth, 3278 strip) spanning distances from 2-60 mm. Multi-taper power spectra were generated separately for grid, depth, and strip contacts. Distances were consolidated across patients and anatomical areas for generalizability, and distance-related influences on task-related brain activity and quantitative interictal epileptiform discharge localization were evaluated.<i>Main results.</i>We identified 8 mm as a consistent reversal point for BPRR, below which low-frequency signals (<30 Hz) had consistently decreased power, and higher frequencies had increased power. Larger distances increased all broadband (2-200 Hz) signals. Task-related increases in superior temporal gyrus 50-200 Hz activity were consistently enhanced across 4-40 mm bipolar distances. There were non-significant difference trends between 4 and 8 mm re-referencing on epileptiform discharge detection.<i>Significance.</i>BPRR distance imposed specific transition points for distance and frequency (roughly 8 mm and ∼30 Hz, respectively) that produced differential effects on measurements of signal power. The consistency across brain regions and electrode types (depth, subdural) suggests these influences are physical brain bio-signal properties, potentially related to spatial wavelength of periodic oscillations in lower frequencies in contrast to more aperiodic activity in higher frequencies. A distance-frequency relation map is provided to help optimize neural signal biomarker quality for intracranial applications by guiding strategic re-referencing distance selection.