Expert Agreement on Identification and Characterization of Electrographic Seizures
Abstract number :
2.471
Submission category :
3. Neurophysiology / 3A. Video EEG Epilepsy-Monitoring
Year :
2023
Submission ID :
1359
Source :
www.aesnet.org
Presentation date :
12/3/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: James Castellano, MD, PhD – University of Pittsburgh
Rishabh Jain, MS – University of Pittsburgh; Wesley Kerr, MD, PhD – University of Pittsburgh; Laura Szklarski, BS – Persyst Development Corporation; Arun Antony, MD – Hackensack Meridian Medical Group; Thandar Aung, MD – University of Pittsburgh; Maria Baldwin, MD – University of Pittsburgh; Niravkumar Barot, MD, MPH – Beth Israel Deaconess Medical Center; Fawad Bilal, MD – Indiana University; Kady Colabrese, BS – Persyst Development Corporation; Joanna Fong-Isariyawongse, MD – University of Pittsburgh; Gena Ghearing, MD – University of Iowa; Simon Glynn, MD – University of Michigan; Lazarus Mayoglou, DO – University of Pittsburgh; Temenuzhka Mihayloca, MD – University of Michigan; Vijayalakshmi Rajasekaran, MD – University of Pittsburgh; Dragos Sabau, MD – Indiana University; Mahmoud Salhab, MD – University of Tennessee; Olga Selioutski, MD – University of Mississippi; Olha Taraschenko, MD – University of Nebraska; Alexandra Urban, MD – University of Pittsburgh; Ahmed Yassin, MD – Jordan University of Science and Technology; Andrew Zillgitt, MD – Beaumont Health; Mark Scheuer, MD – Persyst Development Corporation; Anto Bagic, MD, PhD – University of Pittsburgh
Rationale: Electrographic seizure identification and characterization, such as classification (generalized vs focal), localization, and onset time, are fundamentals of epilepsy management. Building on prior literature, we sought to evaluate expert agreement on identification and characterization of electrographic seizures. This research is of particular importance given the burgeoning application of machine learning and artificial intelligence in electroencephalography (EEG) analysis.
Methods: Scalp EEG data was curated from 120 de-identified prolonged EEG files previously independently marked by three expert readers. Four-hundred and eleven expert-marked seizure events were clipped into 350 files, ranging in duration from 241 to 3320 seconds, including 90-180 seconds of randomized asymmetric padding at the beginning and end of the events. Nineteen (19) board-certified epileptologists of varying experience levels interpreted the individual clips and annotated not only presence of seizures but also various seizure characteristics, such as onset time, duration, and focality. Additionally, as a subcohort (n=3) had evaluated the original prolonged EEG files, we evaluated intra-rater agreement across similar seizure characteristics. We used mixed effects generalized linear and logistic regression to evaluate associations between those seizure characteristics and reader’s experience, level of agreement in seizure, accounting for intra-reader and intra-seizure variability.
Results: We found a non-linear relationship between percentage of total marked seizures and numbers of readers: 50% of seizures were only called by 6/19 (32%) of readers and only 25% of seizures were called by at least 16/19 (85%) of readers In the three readers who interpreted the records twice, intra-reader agreement was weak to moderate (Cohen’s Kappa of 45%, 55%, 67%). More experienced readers did not have a higher sensitivity to specificty balance (favoring calling when consensus was clear), relative to the average reader. For readers who called each seizure, there was better agreement across readers when evaluating seizure classification (generalized versus focal-onset) than when evaluating regional focality of focal seizures (temporal versus non-temporal onset).
Conclusions: We found overall low levels of agreement both across expert EEG reading physicians and within individual readers on electrographic seizure identification and characterization. This lack of consensus highlights the need for further work evaluating the role of specific ictal EEG variables as objective electrographic predictors of treatment outcomes.
Funding: None
Neurophysiology