Cardiac-based detection of seizures in children with epilepsy

Abstract

Introduction: We evaluated a multi-parametric approach to seizure detection using cardiac and activity features to detect a wide range of seizures across different people using the same model.

Methods: Electrocardiogram (ECG) and accelerometer data were collected from a chest-worn sensor from 62 children aged 2-17 years undergoing video-electroencephalogram monitoring for clinical care. ECG data from 5 adults aged 31-48 years who experienced focal seizures were also analyzed from the PhysioNet database. A detection algorithm was developed based on a combination of multiple heart rhythm and motion parameters.

Results: Excluding patients with multiple seizures per hour and myoclonic jerks, 25 seizures were captured from 18 children. Using cardiac parameters only, 11/12 generalized seizures with clonic or tonic activity were detected as well as 7/13 focal seizures without generalization. Separately, cardiac parameters were evaluated using electrocardiogram data from 10 complex partial seizures in the PhysioNet database of which 7 were detected. False alarms averaged one per day. Movement-based parameters did not identify any seizures missed by cardiac parameters, but did improve detection time for 4 of the generalized seizures.

Conclusion: Our data suggest that cardiac measures can detect seizures with bilateral motor features with high sensitivity, while detection of focal seizures depends on seizure duration and localization and may require customization of parameter thresholds.

Keywords: Algorithms; Caregiver alert; Epilepsy monitoring; Seizure detection; Wearables.