Pilot Analysis of a Reduced-Channel EEG-Based Seizure Alerting Algorithm for Use During Sleep
Abstract number :
1.159
Submission category :
2. Translational Research / 2B. Devices, Technologies, Stem Cells
Year :
2025
Submission ID :
706
Source :
www.aesnet.org
Presentation date :
12/6/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Mitchell Frankel, PhD – Epitel, Inc.
Zoë Tosi, PhD – Epitel, Inc.
Avi Kazen, MS – Epitel, Inc.
Tyler Newton, PhD – Epitel, Inc.
Vamshi Muvvala, MS – Epitel, Inc.
Christopher Phillips, MSHS – Epitel, Inc.
Mark lehmkuhle, PhD – Epitel Inc.
Rationale: The risk of seizure-related impairment, including physical injury, asphyxiation, hypoxemia, or even SUDEP, is a constant concern for patients and their caregivers. This risk is especially worrisome during sleep when caregivers may not be alert or present to provide rescue support when a seizure happens. While there are FDA-cleared devices for tonic-clonic seizure alerting and general-use products for motion alerting, there is no EEG-based system that can alert to a wider variety of seizures. As an extension to Epitel’s FDA-cleared, wireless, wearable EEG monitoring system (REMI), an algorithm has been developed that can detect and alert a patient and their caregiver to an electrographically-visible seizure that occurs during sleep within the first 5 minutes.
Methods: Pilot EEG records for 18 participants, aged 6 to 61 years, were selected from previously collected data from patients undergoing seizure monitoring. Participants wore REMI sensors at F7/8 and Tp9/10 electrode locations alongside standard-of-care wired EEG. EEG records were trimmed to only include data from 10p to 8a to encapsulate normal sleep hours, with total durations ranging from 7.6 to 60.0 hours (1-6 nights). A ground-truth data set was determined by expert epileptologist review of the wired EEG, which included 18 electrographic seizures lasting longer than 10s for 10 of the records and comprised tonic-clonic, focal non-evolving, and atypical absence semilogy. The sleep alerting algorithm analyzes 4-channel REMI EEG in 30s packets and follows a traditional machine-learning pipeline of data processing, feature extraction, binary classification, and post-processing to determine if an electrographic seizure is actively occurring. The algorithm is optimized for high sensitivity while still minimizing false alarms because it is critical that potential seizures lead to an alarm to the patient and caregiver. Algorithm performance on the pilot data was evaluated by determining the Positive Percent Agreement (PPA) and False Alarm Rate (FAR: false positives per hour).
Results: The algorithm performed with an event-level PPA of 88.9%, detecting 16 of the 18 known events, with 100% PPA on 9 of 10 records. The event-level FAR was 0.090 FP/hr (0.88 FP/night) over 509 hours of EEG data (52 nights), with zero false positives detected in half (9) of the records. Tonic-clonic seizures are often considered the most dangerous and the algorithm was able to detect all six tonic-clonic events across six participant records. The two missed events were focal non-evolving with review showing localized electrographic activity visible on just a single EEG channel.
Conclusions: This pilot analysis demonstrated that the algorithm can detect electrographically-visible seizures occurring during sleep early in their evolution with high sensitivity and low false alarm rates. Combining this algorithm with a patient-friendly EEG system such as REMI may provide an accurate alerting platform that can enhance patient safety and reduce caregiver anxiety during sleep.
Funding: NIH 1U44 NS121562
Translational Research