Abstracts

Rationale: Resection surgery can eliminate seizures in patients with epilepsy. However, proper resections rely on robust localizations of the epileptogenic zone (EZ), the minimum amount of tissue needed to be removed to eliminate seizures. EZ localization uses stereo-EEG (sEEG) to record signals from regions of interest (ROIs), but the optimal number and positions of electrodes are unknown. Implanting the minimum necessary number of electrodes is essential because each additional electrode increases the risk of hemorrhage, while too few can lead to improper localization and resection. Optimizing electrode configurations to record from ROIs with the minimum number of electrodes may reduce the risks of sEEG and improve EZ localization, increasing rates of seizure freedom and survival.

Methods: We used patient specific neuroimaging to generate an anatomically realistic volume conduction head model. We generated a set of ninety thousand valid electrode trajectories by varying insertion location, angle, and depth, and avoiding sulci. We calculated the leadfield by simulating the voltages at each electrode contact location generated by all cortical sources. Since epileptic sources are patches of synchronously active cortex, we combined cortical sources into patches (~5 cm²) at each cortical location and used the leadfield to calculate the resulting contact voltages. We developed a metric of electrode recording sensitivity to quantify the ability of any configuration to record from a ROI. We then calculated occurrences of collisions between all pairs of electrodes to omit them in our subsequent optimization. We performed an iterative search to build optimal electrode configurations, picking the electrode at each iteration that maximally decreased the unmapped area of the ROI at 1 mV resolution. Finally, we compared the recording sensitivity of our solution configurations to that of the truly implanted set.

Results: We found configurations that were sensitive to sources, or mapped, over 87% of the left temporal lobe, 44% of the left hemisphere, and 38% of the full cortex with at least one electrode using 7, 10, and 15 electrodes respectively and a 1 mV resolution. However, using a 100 µV resolution these configurations mapped >95% of the same ROIs. Additionally, we found that our optimized electrode configuration for the full cortex mapped 10-20% more of the ROI compared to the clinically implanted configuration (Figure 1).

Conclusions: Electrode configurations cannot fully map ROIs using a 1 mV resolution. However, computational algorithms that can interpret signals ≥100 µV may enable near-full cortical mappings with sEEG. Nonetheless, our algorithm identifies electrode placements with higher recording sensitivity than current practice and determines when additional electrodes produce redundant information. Our visualizations of recording sensitivity may also aid in estimating the mapping area of any configuration. These methods may improve the efficacy and reduce the risks of sEEG, increasing the success rates of EZ localization and resection surgery.

Funding: Duke MedX