Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions

Abstract To determine the effect of implanting electrodes on electrographic features of nearby and connected brain regions in patients with drug-resistant epilepsy, we analyzed intracranial EEG recordings from 10 patients with drug-resistant epilepsy who underwent implant revision (placement of additional electrodes) during their hospitalization. We performed automated spike detection and measured EEG functional networks. We analyzed the original electrodes that remained in place throughout the full EEG recording, and we measured the change in spike rates and network connectivity in these original electrodes in response to implanting new electrodes. There was no change in overall spike rate pre- to post-implant revision (t(9) = 0.1, p = 0.95). The peri-revision change in the distribution of spike rate and connectivity across electrodes was no greater than chance (Monte Carlo method, spikes: p = 0.40, connectivity: p = 0.42). Electrodes closer to or more functionally connected to the revision site had no greater change in spike rate or connectivity than more distant or less connected electrodes. Changes in electrographic features surrounding electrode implantation are no greater than baseline fluctuations occurring throughout the intracranial recording. These findings argue against an implant effect on spikes or network connectivity in nearby or connected brain regions.

. Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00248

Automated artifact rejection
In each segment, electrodes were excluded from analysis if they had missing or zero voltage EEG signal in more than half of the segment, suggesting disconnection, or if they met any of the following criteria suggesting substantial electrode artifact in that segment (numerical criteria were chosen by visual analysis of example segments): 1) if greater than 1% of the EEG voltages in the electrode exceeded an amplitude threshold of 10,000 uV, 2) if any EEG voltages exceeded ten times the 99th percentile voltage for that electrode, 3) if greater than 50% of the spectral power in the electrode belonged to the 58 to 62 Hz frequency band, or 4) if the standard deviation of the EEG signal in the electrode was greater than ten times that of the median standard deviation across all electrodes. If more than half of electrodes were discarded as containing excessive artifact, then the entire five-minute segment was discarded as containing excessive artifact. Furthermore, to identify further artifact-heavy segments potentially missed by this analysis, three segments preceding and following each contaminated segment were examined, and if more than three of these surrounding six segments were marked as contaminated by artifact, then the middle segment was also assumed to be contaminated and discarded. We did not attempt to replace this discarded data with new segments around the same time, as periods of artifact and electrode disconnection tend to temporally cluster. Averaged across patients, 7.7% segments (range across patients 2.7%-12.4%) were rejected as either apparently disconnected or artifact-heavy by this method. . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00248

Automated spike detection
Each EEG segment underwent automated interictal spike detection using a previously validated detector which is fully described in the original paper (Brown et al., 2007). Briefly, the segment was high-pass (>7 Hz) and low-pass (<40 Hz) filtered (6th order Butterworth filter).
The filter settings were changed from those in the original reference in order to optimize spike detections based on visual analysis of a sample of our data. Peaks in this filtered signal were identified and subjected to the following criteria: 1) an absolute amplitude threshold, 2) an amplitude threshold relative to the surrounding baseline, 3) maximum (220 ms) and minimum (10 samples, which corresponded to 2.5-10 ms depending on sampling rate) duration thresholds, 4) and the requirement of an after-going slow wave. Amplitude thresholds were tuned for each patient based on visual analysis of 10-minute segments of data. The algorithm was modified such that channels were first referenced to a bipolar montage of adjacent electrodes, as this produced more accurate spike detections by visual analysis. Automated spike detections were discarded if they occurred on only one bipolar channel or on more than half of all channels within 50 ms, suggesting a common artifact. For each spike detection, the spike peak was defined as the maximum of the absolute deviation of the filtered signal relative to the baseline. The interictal spike was then defined to occur on the electrode in the bipolar channel pair with the highest amplitude of the filtered signal on common average reference montage (where only electrodes not discarded as artifact-heavy were included in the common average reference). To validate spike detections, a board-certified epileptologist (EC) visually reviewed 50 randomly chosen detections for each patient and measured the detection accuracy, defined as the percentage of example spikes visually determined to be true spikes (Conrad et al., 2020). Patients with a . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00248 detection accuracy of less than 70% or with sparse spikes (fewer than 5 spikes on more than half of all segments) were excluded from further analyses.

Pearson correlation functional network calculation
After rejecting artifact-heavy electrodes as detailed above, each remaining electrode was remontaged to a common average reference, including as reference only those electrodes present throughout the entirety of the recording (rather than including the electrodes added in the implant revision). This modified common average reference was performed so as to avoid contaminating the original electrode signals with the signal from the newly added electrodes. Each channel signal was then subjected to a notch filter (bandstop IIR filter with cutoff frequencies of 59 and 61 Hz) and a bandpass filter (IIR filter, 8th order, with cutoff frequencies of 1 and 70 Hz). The five-minute segment was then divided into non-overlapping consecutive two-second windows.
The Pearson correlation coefficient was then calculated for every pair of electrodes, taking the full broadband (1-70 Hz bandpass filtered as above) signal in the two-second window. This resulted in a symmetric adjacency matrix for each window, of size NE x NE, where NE is the number of electrodes, and each (i, j) element of the matrix represents the Pearson correlation coefficient between electrode i and j. The adjacency matrices were averaged across all twosecond windows to yield one matrix per five-minute segment. The choice of two-second window and subsequent averaging over windows has been used in prior studies to compare spikes and functional networks across the same time scale (Crippa et al., 2011;Deligianni et al., 2014;Godwin et al., 2017;Wang et al., 2020).

Anatomical differences in changes in peri-revision spike rate and connectivity
We measured the relative change in spike rate and node strength across electrodes from the pre-revision to post-revision period. We excluded electrodes deemed to be outside cerebral tissue. We assigned remaining electrodes to one of the following anatomical locations: white matter, mesial temporal, temporal neocortex, and other. We chose these categories because distinguishing between mesial temporal, temporal neocortical, and other seizure onset localizations is a common clinical problem and these seizure onset localizations have corresponding differences in spike locations (Goncharova et al., 2009). We averaged the relative feature changes across all electrodes in each anatomical location. We compared the relative perirevision change in each EEG feature between anatomic locations across all patients using a Skillings-Mack test, which is a non-parametric method to test for differences between conditions across repeated measures (Hollander et al., 2013;Mack & Skillings, 1980;Skillings & Mack, 1981).

Supplementary Figure legend
Conrad, E., Shinohara, R., Gugger, J., Revell, A., Das, S., Stein, J., Marsh, E., Davis, K., . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions." Network Neuroscience. Advance publication. https://doi.org/10.1162/netn_a_00248 Figure S1. The correlation between relative spike rate change and distance from the implant revision site for each patient. Each plot shows results for an individual patient. Each circle represents a single original electrode contact and shows both its peri-revision relative spike rate change (y-axis) and its distance from its nearest added electrode (x-axis). The Spearman rank correlation coefficient and associated p-value are shown. Also shown is the p-value from a Monte Carlo test comparing the Spearman rank correlation coefficient against those calculated from randomly-chosen pseudo-revision times. Red diamonds indicate electrodes with an infinite relative rate increase (those with zero pre-revision spikes and non-zero post-revision spikes detected), ranked as having the highest relative rate increase in the Spearman rank correlation.
All results are shown comparing the pre-and post-revision spike rates for a 24-hour peri-revision surround period (12 hours pre-and 12 hours post). Cases in which the Spearman rank correlation p-value is significant but the Monte Carlo p-value is not imply that the correlation is no larger than that observed at random times, and likely reflect examples of spatial autocorrelation (nearby electrodes experience similar changes in spike rates). Abbreviations: MC = Monte Carlo. Conrad, E., Shinohara, R., Gugger, J., Revell, A., Das, S., Stein, J., Marsh, E., Davis, K., . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions. EEG recording and spike detector information. Abbreviations: D = depth electrodes, S = strip electrodes, G = grid electrodes, PPV = positive predictive value. The peri-revision gap in recording refers to the period of time in which the EEG was either not recording or leads were disconnected surrounding implant revision.  Each row represents a different duration defining the early and late implantation periods. Each . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions.   Carlo tests comparing the specific analysis test statistic against those obtained from selecting random pseudo-revision times. The test for the anatomical analyses are Skillings-Mack tests. NS . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions.   Conrad, E., Shinohara, R., Gugger, J., Revell, A., Das, S., Stein, J., Marsh, E., Davis, K., . Supporting information for "Implanting intracranial electrodes does not affect spikes or network connectivity in nearby or connected brain regions.