In Epilepsia
OBJECTIVE : The aim is to report the performance of an electroencephalogram (EEG) seizure-detector algorithm on data obtained with a wearable device (WD) in patients with focal refractory epilepsy and their experience.
METHODS : Patients used a WD, the Sensor Dot (SD), to measure two channels of EEG using dry electrode patches during pre-surgical evaluation and at home for up to eight months. An automated seizure detection algorithm flagged EEG regions with possible seizures, which we reviewed to evaluate the algorithm's diagnostic yield. In addition, we collected data on usability, side effects and patient satisfaction with an electronic seizure diary application (Helpilepsy®).
RESULTS : Sixteen inpatients used the SD for up to five days and had 21 seizures. Sixteen outpatients used the device for up to eight months and reported 101 focal impaired awareness (FIA) seizures during the periods selected for analysis. Focal seizure detection sensitivity based on behind-the-ear EEG was 52% in inpatients and 23% in outpatients. False detections/hour, positive predictive value (PPV) and F1 scores were 7.13, 0.11%, 0.002for inpatients and 7.77, 0.04% and 0.001 for outpatients. Artefacts and low signal quality contributed to poor performance metrics. The seizure detector identified nineteen non-reported seizures during sleep, when the signal quality was better. Regarding patients' experience, the likelihood of using the device at six months was 62%, and side effects were the main reason for dropping out. Finally, daily and monthly questionnaire completion rates were 33% and 65%, respectively..
SIGNIFICANCE : Focal seizure detection sensitivity based on behind-the-ear EEG was 52% in inpatients and 23% in outpatients with high false alarm rates and low PPV and F1 scores. This unobtrusive wearable seizure detection device was well received but had side effects. The current workflow and low performance limit its implementation in clinical practice. We suggest different steps to improve these performance metrics and patient experience.
Macea Jaiver, Bhagubai Miguel, Broux Victoria, De Vos Maarten, Van Paesschen Wim
2023-Jan-22
focal seizure, machine learning, real-world data, seizure detection, wearable devices
