Abstracts

Rationale: Stereo-EEG (SEEG) electrodes are commonly placed to locate the onset of seizures in patients with intractable epilepsy prior to epilepsy surgery. Currently, the wires and recording devices are connected to the SEEG electrodes after their placement, creating room for potential mistakes affecting the label of the SEEG electrodes and in turn, the localization of the seizure onset zone. This study develops an automated algorithm to verify the correctness of these wire connections using electrical stimulation of intracranial electrodes and analyzing simultaneous scalp EEG recordings. Methods: Experiments using both simulated data and real data were conducted. For simulated data, 9 SEEG electrodes were placed on a 3-sphere head model with varying number of scalp EEG electrodes. For real data, cortical electrical stimulation was performed in a patient with 7 SEEG electrodes and 16 scalp EEG electrodes in the standard 10-20 system. The SEEG electrodes stimulated currents using bipolar, biphasic, square wave pulses, ranging 1-10 mA, 5-25 Hz.Our algorithm exploits the fact that scalp EEG electrodes are sensitive to stimulating currents applied to the intracranial electrodes. Intuitively, using scalp EEG signals, the algorithm attempts to identify which SEEG electrode could have generated that particular EEG signal utilizing the EEG forward model.During the stimulation period for each SEEG electrode, several scalp EEG recording samples are extracted at the peak of stimulating pulses. Then, hypothesized noiseless scalp EEG signals are generated using the forward model for each electrode. Finally, the algorithm attempts to identify the correct label by (i) finding the source whose hypothesized noiseless signal is the closest to the real recordings for every data (time) point, forming intermediate labels for each data point (ii) and taking the majority of labels of the scalp EEG recording samples as the final label for each SEEG electrode. Results: Simulation results: The attached figure shows the results of the simulation signals with SNR=10. All nine electrodes were correctly localized. When the labeling accuracy for each scalp EEG recording samples was examined,  the accuracy improved with larger Stereo EEG distance and more scalp EEG electrodes, likely because more EEG electrodes can capture higher spatial resolution that helps differentiate signals from each SEEG electrodes.Real data: For real recordings, five out of seven electrodes were correctly identified without knowing the correct labels of any of the SEEG electrodes as a reference. 79.05% of the extracted EEG recording samples were correctly labeled. The mislabeled electrodes were in close proximity. There is a substantial bias between the hypothesized signals and the real scalp EEG data, perhaps because of forward modeling approximations, which we believe is the reason for the mislabeling. Conclusions: Simultaneous stereo and scalp EEG, along with data from cortical stimulation, helps verify the locations and connections of stereo EEG electrodes, which can prove to be an essential step to avoid electrode mislabeling and ensuing surgical errors. It is important to understand the cause for the bias in predicted EEG signals through higher density scalp recordings, acquiring data from more patients, and more rigorous simulations that account for uncertainties in forward modeling. Funding: None