Abstracts

Rationale: The current approach for patients and caregivers to track long-term (over days and months) seizure frequency is by maintaining seizure diaries. However, it has been shown that self-reporting of seizure incidence is severely inaccurate. Seizure detection via autonomic signatures such as cardiac or motor signals that are altered by seizures may be an alternative for long-term seizure monitoring. In this work, a multi-modal seizure detection system has been developed using cardiac (ECG) and tri-axial accelerometer (ACM) signals that are utilized independently in order to detect seizures that manifest either via onset of ictal tachycardia or convulsive movement and then send a notification to the system user. Methods: The ECG and ACM signals are acquired using a chest-worn sensor that allows for reliable acquisition of ECG signals and convulsive movements. The ECG signal is used to estimate the ratio of background and foreground heart rate. A user-programmable detection setting determines the threshold for cardiac based seizure detection. The ACM signal extracts movement features that are stereotypical of seizure movements, especially during night-time. These features are combined with patient`s postural information in order to produce a detection system that can detect seizure-related movements during sleep. An additional independent user-programmable threshold accounts for variations in movement intensity due to restrictions posed by blankets, bed-types, and/or the presence of caregivers sleeping next to the patient. Performance of the cardiac based seizure detection algorithm (sensitivity, false positive rate, and detection latency) was retrospectively evaluated using 581 hours of ECG data collected at multiple epilepsy monitoring units; the seizures primarily comprised of complex partial (focal dyscognitive) with or without secondary generalization. Additionally, performance of the accelerometer based seizure detection algorithm was evaluated using 540 hours of ACM data from pediatric patients with hypermotor seizures. Results: Overall performance of the seizure detection system was reasonable for night-time monitoring of seizures (>80% mean sensitivity and a mean of 2 false positives/night). The cardiac based seizure detection algorithm achieved a mean sensitivity of 84.4% with a false positive rate of 2.17/night. Seizure detected by the cardiac based seizure detection algorithm typically consisted of complex partial seizures with or without secondarily generalization. The performance of the accelerometer-based seizure detection algorithm achieved a sensitivity of 97.02% for detecting hypermotor seizures with a false positive rate of 2.1/night. Conclusions: The performance of real-time, prospective multi-modal seizure detection using ECG and ACM signals appear to have high sensitivity and false positive rate for night-time monitoring of seizures. Wearable devices that chronically monitor cardiac or motor signals associated with seizures may significantly improve the overall quality of life of patients and caregivers.