|
Question: Can deep learning models trained on expert-verified electroencephalography segments reliably detect seizures and be applied to single-channel wearable monitoring?
Finding: This technique demonstrated 80.1%–100% sensitivity and low false-alarm rates. On 1,604 expert-labeled segments, the model outperformed previous studies in terms of latency and accuracy across unseen datasets.
Meaning: Expert-labeled data are essential for achieving universal seizure detection. This channel-specific approach enabled efficient and continuous monitoring using wearable devices. |