Key points are not available for this paper at this time.
BACKGROUND: Epileptic seizures can lead to severe outcomes including sudden unexpected death in epilepsy (SUDEP). Clinical standard for seizure diagnosis and detection requires electroencephalography and video monitoring, which is yet considered not suitable for home use, especially during nighttime sleep in a low-light condition. We proposed a deep learning (DL)-based approach to automatically detect nocturnal major seizures using a wearable armband that can potentially help reduce SUDEP risk through timely caregiver intervention. METHODS: In this prospective cohort study, 68 patients with major seizures were monitored for up to three months using a wearable armband (NightWatch®) capturing tri-axial accelerometry (ACM) and photoplethysmography (PPG) signals. A two-step approach was designed: (1) a pre-screening step using threshold-based algorithms to identify suspected seizure events (ACM standard deviation >0.4 or heart rate increase >10%), and (2) a DL model (CNN-LSTM with attention mechanism) to recognize true seizures. Model performance was evaluated via a 10-fold cross-validation, reporting sensitivity (SEN), false alarm rate (FAR), and area under the ROC curve (AUC). RESULTS: In 788 overnight recordings (6304 hours), a total of 1846 severe seizures were identified. The pre-screening step achieved 0.940 sensitivity in pre-identifying or 'preserving' seizures, reducing data volume by 81% (from 6304 to 1201 hours). The DL model demonstrated a mean accuracy of 0.793 95% CI: 0.745-0.841, a mean sensitivity of 0.762 95% CI: 0.704-0.821, a mean positive predictive value of 0.334 95% CI: 0.229-0.356 and a mean false alarm rate of 0.165/hour 95% CI: 0.097-0.234. These results exceeded those of single (signal) modality detection methods. CONCLUSION: Our two-step approach enables accurate, long-term detection of severe nocturnal seizures in home settings. The wearable system provides a practical solution for continuous monitoring and real-time alerts, thus potentially reducing SUDEP risk and improving patient safety, fulfilling an urgent unmet need in epilepsy care. Furthermore, by enabling long-term home monitoring, this system may help assess the relationship between seizure events and lifestyle-related triggers such as sleep deprivation, stress, physical exertion, or alcohol consumption, thereby supporting the development of personalized preventive strategies.
Dong et al. (Mon,) studied this question.