Patient-independent automated pediatric seizure monitoring based on expert-labeled electrographic ictal data as core reference knowledge.

Background: Automated seizure detection using scalp electroencephalography (EEG) is essential to the efficient monitoring of seizures in patients with epilepsy. However, patient-independent seizure detection remains challenging, primarily because of the inherent intersubject variability in EEG characteristics. Purpose: We proposed a patient-independent seizure detection approach based on 1,604 single-channel electrographic focal-onset ictal EEG segments verified by epileptologists in patients with focal epilepsy. Methods: We constructed deep learning models trained on these segments and applied them to individual EEG channels to identify seizure occurrences. To evaluate patient-independent detection performance, we conducted internal validation using the 2 datasets employed for segment acquisition, followed by external validation with an independent unseen dataset obtained from a tertiary medical institution. Results: In the internal validation using a leave-one-patient-out scenario, overall sensitivity, false alarm rate, and latency were 80.1%-100%, 0.64-1.13/hr, and 8.0-21.4 seconds, respectively. In the external validation using the final seizure detection model, the corresponding values were 100%, 0.38-0.67/hr, and 10.0-10.2 seconds, respectively, according to detection length. This technique outperformed those of previous patient-independent studies that employed relatively simple deep learning tasks. Conclusion: A curated set of expert-labeled ictal EEG segments served as the core reference knowledge for the proposed seizure detector in recognizing seizure occurrence in unseen patients. Because the proposed approach analyzes individual channels in parallel, it may be clinically applicable to continuous seizure monitoring, particularly in wearable seizure detection systems with a limited number of channels.