Authors :
Presenting Author: Chifaou Abdallah, MD – McGill University
John Thomas, PhD – Duke University
Olivier Aron, MD – CHU Nancy
Tamir Avigdor, MSc – Analytical Neurophysiology Laboratory, Montreal Neurological Institute and Hospital, McGill University, Montreal, QC, Canada
Kassem Jaber, MSc – Duke University
Irena Dolezalova, doc – 1 St Department of Neurology, Faculty of Medicine, Masaryk University, Brno, Czech Republic
Daniel Mansilla, MD – McGill University
Päivi Nevalainen, MD – Epilepsia Helsinki, Department of Clinical Neurophysiology, HUS
Prachi Parikh, MD – Duke University
Jaysingh Singh, MD – The Ohio State University Wexner Medical Center
Sándor Beniczky, MD – Dianalund and Aarhus University Hospital
Philippe Kahane, MD – CHU Grenoble
Lorella Minotti, MD – CHU Grenoble
Stephan Chabardes, MD – CHU Grenoble
Sophie COLNAT-COULBOIS, MD, PhD – IMoPA Neuroscience, CNRS UMR 7365
Louis MAILLARD, MD, PhD – IMoPA Neuroscience, CNRS UMR 7365
Jeffery Hall, MD, FRCS(C) – Montreal Neurological Institute-Hospital
Francois Dubeau, MD – McGil University
Jean Gotman, PhD – Montreal Neurological Hospital and Institute
Christophe Grova, PhD – Concordia University
Birgit Frauscher, MD, PhD – Department of Neurology, Duke University School of Medicine, Durham, NC, USA
Rationale: Up to 40% of patients continue to experience disabling seizures after stereo-electroencephalography (SEEG) guided surgery. There is a need for identifying reliable features predictive for epilepsy surgery which can be easily implemented in every epilepsy surgical center worldwide. Previous studies are limited by small sample sizes, lack of external validation, and the use of computational approaches which are difficult to implement in clinical practice. We aimed to: 1) identify visually optimal SEEG features with the highest predictive performance for surgical outcome; 2) validate their performance using external datasets; and 3) evaluate their visual analysis reliability among several epileptologists trained at different centers and blinded to the surgical outcome.
Methods: Using a multicentric SEEG dataset of patients who underwent SEEG investigation as part of their presurgical evaluation for focal drug-resistant epilepsy, we included consecutive patients who had preoperative, post-SEEG implantation and postoperative imaging data. All had: i) at least one spontaneous typical electro-clinical seizure recorded during SEEG recordings, ii) cortical stimulation data, and iii) at least one-year postoperative follow-up. From this derivation cohort, we visually extracted 10 SEEG features from interictal, preictal, ictal, postictal periods, and cortical stimulation data (Figure1). We examined their performance at the single feature level and after considering their spatial co-occurrence in classifying seizure-free (Engel Ia) versus non-seizure-free (Engel Ib-IV) surgical outcome patients in terms of the
area-under-the-receiver-operating-characteristic-curve values (AUC). We used the resected channels and surgical outcome as the gold-standard. The performance of the identified optimal feature was evaluated in two external centers (including an open dataset) by computing the balanced accuracy (BA) after applying a frozen threshold from the derivation cohort. Lastly, we asked six experts to visually extract the optimal predictive feature and assessed the inter-rater reliability (IRR) using percentage agreement (PA standard derivation+/-SD) and beyond-chance agreement (Gwet kappa κ+/-SD).
Results: The derivation cohort comprised 100 consecutive patients (40% temporal lobe epilepsy; median age, 30.80 years; (interquartile range, IQR, 22.49-38 years); 53% women; 42% Engel Ia). Spatial co-occurrence of gamma-spikes and preictal spikes was found as the optimal surgical predictive feature (AUC=0.82, Figure 2). Applying a frozen threshold from the derivation cohort, external validation in the two datasets resulted in similar performances, with a BA of 69.2% and 73.2%. Expert IRR for gamma-spikes (PA 96%+/-2%; κ 0.63+/-0.16) and preictal spikes (PA 92%+/-2%; κ 0.65+/-0.18) were substantial.
Conclusions: Spatial co-occurrence of gamma-spikes and preictal spikes, two features easily and reliably identifiable in every epilepsy center, could reduce the burden of SEEG analysis by reducing the time of analysis and could improve surgical outcome by guiding resection margins.
Funding: CIHR, PJT-175056, Doctoral-CIHR.