Abstracts

Rationale: Detection of generalized tonic-clonic seizures (GTCS) at home and during activities of daily living could facilitate earlier intervention. However, to date there are no FDA-cleared devices that provide an accurate means for GTCS detection during activities of daily living. This study aims to validate an EMG-based GTCS detection algorithm to be used later in a seizure detection system.Methods: We (a) compared a software algorithm analysis of electromyography (EMG) signals continuously acquired in an Epilepsy Monitoring Unit (EMU) to video-EEG (vEEG) recordings, and (b) optimized settings for the software algorithm using baseline measurements of maximum voluntary muscle contraction (MVC). Inpatient scalp vEEG monitoring was performed along with surface EMG recordings of unilateral biceps and triceps brachii. The EMG data was analyzed separately by a seizure detection software algorithm (Brain Sentinel, Texas) and correlated with the determination of seizure onset and characterization by board-certified clinical neurophysiologists. The algorithm continuously compares recorded surface EMG signals to a baseline sample of muscle activity; GTCSs are signaled with sustained activation of multiple frequency bands (30-40 Hz, 130-240 Hz, 300-400 Hz) of EMG activity. The algorithm was expected to detect GTCSs within 10 seconds of arm movement during the GTCSs, while minimizing false positive detections.Results: In this interim analysis, we included 29 patients enrolled into the study while in the EMU (University Hospital, San Antonio, TX) for routine vEEG monitoring related to seizures with a history of GTC seizures. The mean age was 38.5 (range 14-64) years (55% Male). EMG recordings averaged 42.4 hours per patient, providing over 1228 hours of surface EMG data. Of 191 seizures recorded by both vEEG and EMG in 29 subjects, 11 had a total of 22 GTCSs captured by EMG recording and confirmed by vEEG. The EMG algorithm detected all 22 GTCSs within 30 seconds of arm motor activation. There were no false positive detections. Other seizures, including 84 myoclonic, 34 tonic, 12 absence, 37 focal seizures with impairment and 3 focal seizures without impairment, were recorded by both vEEG and EMG, but none triggered a GTCS alarm using the seizure detection algorithm. No alarms were triggered during any activities of daily living or other interictal motor activities.Conclusions: This interim view of a continuing study demonstrates the feasibility of accurately detecting GTCSs with a non-invasive, arm-worn device analyzing EMG signals. As it pertains to GTCSs, the sensitivity and positive predictive value of the seizure detection algorithm appears to be superior to other devices currently under investigation or currently commercialized. This device is being developed to analyze the EMG signal in real time.