Abstracts

Rationale: Stereo-electroencephalographic (SEEG) electrodes are increasingly used to localize seizure foci in patients with drug resistant epilepsy (DRE) (Gonzalez-Martinez, J Neurosurg 2014). Because of the complexity of SEEG data collected from the entire brain volume, visualization tools are needed to effectively interpret the recorded electrical activity and improve their clinical use (Narizzano, BMC Bioinformatics 2017). Here we developed a Python-based software tool to automatically extract SEEG-recorded ictal and interictal epileptiform activity and visualize them in the three-dimensional (3D) anatomical space.

Methods: We analyzed SEEG data from six patients who underwent evaluation for DRE at the OSF Saint Francis Medical Center in Peoria, Illinois. All patients underwent preoperative magnetic resonance imaging (MRI) and post-implant computed tomography (CT). SEEG data were recorded using DIXI SEEG electrodes sampled at 512 Hz. First, we co-registered the MRI to the post-implant CT. We segmented the SEEG sensors from the CT image by applying image erosion and thresholding and grouping the voxels into sensors by treating them as point clouds. We then mapped the identified sensors to SEEG channels based on the naming convention used during implantation. Next, we selected ictal SEEG segments based on epileptologist annotations and sleep interictal segments sufficiently far from seizures. For both ictal and interictal segments, we applied a notch filter to remove powerline noise and a bandpass filter to extract activity in the ripple band (80 - 250 Hz). We then applied a high-frequency oscillation (HFO) detector identify putative HFO rates (events/minute) in the interictal segments (Cai, Front Neuroinform 2021). We then created a whole brain source model using a 5 mm grid and computed dipole activations as weighted summations of ictal action potentials and HFO rates of nearby SEEG sensors where the weights are based on sensor proximity. We then compared the ictal and interictal candidate locations against clinically determined areas of resection (i.e., gold-standard seizure foci).

Results: We illustrate the results of our visualization tool using one patient’s data. Figure 1 shows results of the segmentation algorithm (i.e., the segmented sensors) overlayed on the CT image. Figure 2 shows the visualizations of HFO rates projected onto the whole brain source model (2A) and ictal activity projected on the whole brain source model at 10 ms prior to seizure onset (2B), seizure onset (2C), 10 ms and 20 ms after seizure onset (2D-E). We generally observed a good concordance between the identified ictal and interictal activations and the gold-standard seizure foci.

Conclusions: We developed a software tool to extract ictal and interictal SEEG activity and visualize in the 3D anatomical space. Our preliminary evaluation using SEEG data from six DRE patients suggests that automated visualization of SEEG activity on a 3D model might enhance the identification of seizure foci and subsequent surgical planning.

Funding: Jump ARCHES Foundation