Abstracts

Propagating Source Reconstructions from Stereo-EEG Recordings Predict Postsurgical Seizure Freedom

Abstract number : 1.171
Submission category : 3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year : 2023
Submission ID : 140
Source : www.aesnet.org
Presentation date : 12/2/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: Brandon Thio, BSE – Duke University

Nishant Sinha, PhD – University of Pennsylvania; Kathryn Davis, MD – University of Pennsylvania; Saurabh Sinha, MD PhD – University of Pennsylvania; Warren Grill, PhD – Duke University

Rationale:

This study aims to determine the effectiveness of a novel stereo-EEG propagating source reconstruction method to delineate the epileptogenic zone in drug-resistant focal epilepsy.



Methods:

We developed a propagating source reconstruction algorithm called TEmporally Dependent Iterative Expansion (TEDIE) that reconstructs propagating neural sources. TEDIE iteratively optimizes the number, location, and size of neural sources to minimize the differences between the reconstructed and recorded stereo-EEG signals and makes use of temporal information to refine the reconstruction. The output of TEDIE is a movie of epileptiform activity projected onto patient-specific brain anatomy. We used TEDIE to reconstruct 194 seizures from 45 patients at Duke University Hospital and the Hospital of the University of Pennsylvania. We compared the distance between the seizure onset reconstructions and the nearest resected electrode contact, the patient specific consistency of seizure reconstructions, and, for patients treated with mesial temporal lobe ablations, the proportion of seizure onset reconstructions with anterior temporal lobe sources for seizure-free (Engel 1) and not seizure-free patients (Engel 2-4).



Results: TEDIE identified seizure onset zones that were closer to the resected electrode contacts for seizure-free patients compared to not seizure-free patients (median distances = 11.0 mm vs 18.6 mm; p = 0.04). However, a clinically useful biomarker needs to predict the outcome of surgery without surgery data. Therefore, we investigated the consistency of seizure reconstructions for each patient with multiple seizure recordings to determine whether a patient would be a good candidate for surgery. Seizures were more consistently colocalized for seizure-free patients compared to not seizure-free patients (median distance = 9.4 mm vs 17.7 mm; p = 0.003), and a classifier trained to differentiate between seizure-free and not seizure-free patients had an AUROC of 0.75. When only considering extratemporal lobe cases (n = 16), classification performance improved to AUROC = 0.85, while for temporal lobe cases (n = 26), performance was worse (AUROC = 0.65). We also investigated the proportion of seizure onsets with anterior temporal lobe sources, for patients who received a mesial temporal lobe ablation to determine whether we could identify patients who are good ablation candidates. Patients that were seizure-free after an ablation had a smaller proportion of their seizures with anterior temporal lobe sources compared to not seizure-free patients (median % seizures with anterior temporal lobe sources for seizure-free vs not seizure-free patients = 20% vs 40%; p = 0.04).

Conclusions: The consistency and location of seizure onset zone reconstructions from TEDIE can be used as biomarkers to predict surgical outcomes for epilepsy.

Funding:

This work was supported by Duke MEDx, Duke CTSA Grant UL1 TR002553, and NIH F31NS124094.



Neurophysiology