Validation of a Discrete Electrographic Seizure Detection Algorithm for Extended-duration Wearable EEG
Abstract number :
3.195
Submission category :
2. Translational Research / 2B. Devices, Technologies, Stem Cells
Year :
2024
Submission ID :
756
Source :
www.aesnet.org
Presentation date :
12/9/2024 12:00:00 AM
Published date :
Authors :
Presenting Author: Mitchell Frankel, PhD – Epitel, Inc.
Mark Lehmkuhle, PhD – Epitel, Inc.
Avi Kazen, MS – Epitel, Inc.
Zoë Tosi, PhD – Epitel, Inc.
Tyler Newton, PhD – Epitel, Inc.
Vamshi Muvvala, MS – Epitel, Inc.
Tobias Loddenkemper, MD – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Lillian Voke, BS – UMass Chan Medical School
Michele Jackson, BA – Boston Childrens Hospital
Edeline Jean Baptiste, BS – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Stephanie Dailey, BA – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Latania Reece, BA – Boston Children’s Hospital
Claire Ufongene, MD – Boston Children’s Hospital
Mark Spitz, MD – University of Colorado - Anschutz
Laura Strom, MD – University of Colorado - Anschutz
Meagan Watson, MPH, MBAc – University of Colorado School of Medicine
Mackenzi Moore, MPH – University of Colorado - Anschutz
Trey Jouard, MS – University of Colorado - Anschutz
Lauren McCall, MS – University of Colorado - Anschutz
Kristal Biesecker, BA, R. EEG T., CLTM – University of Colorado - Anschutz
Christopher Mizenko, MS – University of Colorado - Anschutz
Michelle Sandoval, BS – University of Colorado - Anschutz
Daniel Friedman, MD – New York University Grossman School of Medicine, NYU Langone Health
Jeschke Jay, MA – New York University - Langone
Leslee Willes, MS – Willes Consulting Group, Inc.
Meredith Decker, MS – Willes Consulting Group, Inc.
Rationale: Epitel has developed a wireless, wearable EEG monitoring system (REMI) that is US FDA-cleared for use in healthcare and ambulatory environments for up to 30 days. To support review of this extended-duration data, a novel algorithm, diverse across patients, environments, and seizure types, was developed and validated as a clinical decision support system to detect and highlight regions of REMI EEG that are suggestive of self-limiting electrographic seizures.
Methods: Data was collected from U.S. patients in EMUs and ambulatory settings. Patients wore REMI wireless sensors at F7/8 and Tp9/10 electrode locations alongside standard-of-care wired-EEG. The EEG was preprocessed to account for noise, artifacts, and cross-patient differences. The data was then windowed into 2s segments and features were extracted in time, frequency, and complexity domains. A subset of representative patients was held out for validation, and the segmented-data features from the remaining patients were used to train a machine learning classifier. The output, the probability that each data segment contains seizure activity, was then fed to a post-processor trained to merge the segment probabilities into discrete events (i.e., start/stop times). The algorithm was optimized for high-sensitivity because it is more important to ensure potential seizures are highlighted for clinician review than reduce false positives. Ground-truth electrographic seizures (GTES) were determined using consensus review of the wired-EEG from 3 independent Epileptologists. Algorithm performance was evaluated on the validation data by determining the True Positive Rate (Sensitivity) and False Alarm Rate (FAR = False Positives per hour [FP/hr]).
Results: There were 50 patients in the validation set (age range: 6-61, median: 21) with 19 noted as having no abnormal EEG characteristics (7 suspected of non-epileptic seizures). The other 31 patients experienced a total of 87 GTES (range: 1-11, median: 2). Recording durations ranged from 3.4 to 151.9 hrs (median: 45.2hrs). The algorithm event-level Sensitivity was 86% across all GTES with a lower 95% confidence interval bound (CI) of 80% and mean per-patient Sensitivity of 92%. The event-level FAR was 0.16 FP/hr across all detections with an upper 95% CI of 0.22 FP/hr and a mean per-patient FAR of 0.18 FP/hr. For patients with GTES, the event-level FAR was 0.18 FP/hr with an upper 95% CI of 0.26 FP/hr, while event-level FAR was substantially lower for those without (0.11 FP/hr, 95% upper CI: 0.15 FP/hr). As expected, event-level FAR was higher for ambulatory (0.29 FP/hr) than for EMU patients (0.14 FP/hr). Nine patients had no more than one FP (5 w/no GTES), and 4 patients had no FP (2 w/no GTES).
Conclusions: With low FAR and high Sensitivity, this algorithm has the potential to support review of REMI EEG, especially for extended-duration REMI ambulatory recordings in a person’s normal daily life. We anticipate that this algorithm will drastically reduce the time required for Epileptologists to review the extended-duration REMI EEG without sacrificing clinical accuracy.
Funding: NIH 1U44 NS121562
Translational Research