Enhancing Epilepsy Surgery Decision Making with a Seizure Matching System
Abstract number :
3.156
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2023
Submission ID :
781
Source :
www.aesnet.org
Presentation date :
12/4/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: John Thomas, PhD – Montreal Neurological Institute, McGill University
Department of Neurology, Duke University
Chifaou Abdallah, MD – Montreal Neurological Institute and Hospital, McGill University; Kassem Jaber, BEng – Montreal Neurological Institute and Hospital, McGill University; Olivier Aron, MD, PhD – Department of Neurology, University Hospital of Nancy, Lorraine University, Nancy, France; Irena Doležalová, MD, PhD – First Department of Neurology, St. Anne's University Hospital, Faculty of Medicine, Masaryk University, Brno, Czech Republic; Vadym Gnatkovsky, MD, PhD – Department of Epileptology, University Hospital Bonn, Bonn, Germany; Daniel Mansilla, MD – Montreal Neurological Institute and Hospital, McGill University; Päivi Nevalainen, MD, PhD – Department of Clinical Neurophysiology, HUS Diagnostic Center, University of Helsinki and Helsinki University Hospital, Finland; Raluca Pana, MD – Montreal Neurological Institute and Hospital, McGill University; Stephan Schuele, MD – Department of Neurology, Northwestern University, Chicago, Illinois; Jaysingh Singh, MD – Department of Neurology, The Ohio State University Wexner Medical Center, Columbus, Ohio, USA; Ana Suller-Marti, MD – Department of Clinical Neurological Sciences, Schulich School of Medicine and Dentistry, Western University; Alexandra Urban, MD – University of Pittsburgh Comprehensive Epilepsy Center, Pittsburgh, United States; Jeffery Hall, MD – Montreal Neurological Institute and Hospital, McGill University; François Dubeau, MD – Montreal Neurological Institute and Hospital, McGill University; Jean Gotman, PhD – Montreal Neurological Institute and Hospital, McGill University; Birgit Frauscher, MD, PD – Montreal Neurological Institute and Hospital, McGill University
Rationale: Precise localization of the epileptic focus is crucial for achieving seizure-freedom in patients with drug-resistant focal epilepsy [1]. Stereotactic-EEG (SEEG) has gained popularity around the world [2], but the number of SEEG patients of each center remains too low to use past cases to directly optimize surgical treatment. Large Open SEEG datasets are now available [3], and an automated system may have the potential to aid epileptologists to review SEEG databases rapidly and identify similar patient cases. We assess the expert interrater agreement in determining if SEEG seizures from different patients are similar and the features contributing to this similarity. We hypothesize that there exists a moderate agreement for seizure similarity across epileptologists and that a generalizable automatic seizure matching system can therefore be developed.
Methods: Three hundred twenty electroclinical seizures from ninety five consecutive patients undergoing SEEG investigation were included. In phase 1, the seizure onset, end, and onset pattern [4] were marked by the consensus of two neurophysiologists specialized in SEEG. To compare pairs of seizures, we used a graphical user interface that integrated multimodal data including implantation scheme, SEEG signals, and expert annotations; the comparison resulted in a 4-point similarity score (very similar, somewhat similar, low similarity, or not related). Next, we defined six clinically relevant features of seizure similarity: onset region, propagation region, duration, spread, propagation speed, and onset pattern. In phase 2, eight experts scored seizure similarity by being shown pairs of randomly selected seizures. Using these scores, we computed the interrater agreement, determined the features correlated with the expert scores, and evaluated the seizure matching system based on these features by conducting a leave-one-expert-out evaluation.
Results: The interrater agreement was fair (4-level classification: percentage agreement = 75±5.1%, Gwet’s beyond-chance agreement kappa = 0.46±0.12; binary classification of similar vs. not related: percentage agreement = 72.1±7.6%, kappa = 0.48±0.14; Fig. 1). The onset region was identified as the most correlated feature with the mean expert labels (Spearman’s rho=0.76, p< 0.001, Fig. 2). We achieved an area under curve of 0.83±0.06 for the binary
Neurophysiology