A multicenter, video-EEG-based validation of a multimodal wearable device for focal seizure detection in adults: The...

Currently available wearable devices for detecting focal seizures primarily target major motor seizures or involve semi-invasive subscalp implants. There is a pressing need for accurate, non-invasive methods to detect diverse focal seizures for long-term, out-of-hospital monitoring.

In this multicenter study (SeizeIT2), people with refractory focal epilepsy undergoing long-term video-electroencephalography (EEG) monitoring were simultaneously recorded with the Sensor Dot, a multimodal wearable device using EEG and electrocardiography (ECG). Wearable recordings were first processed with an offline detection algorithm, after which the algorithm-labeled segments underwent blinded review by a human expert, and results were compared with ground truth video-EEG. Post-hoc analyses were performed to evaluate how specific seizure characteristics, defined by the gold-standard recordings, influenced detection sensitivity.

We recruited 192 adult participants, documenting a total of 616 focal seizures. The mean duration of Sensor Dot monitoring was 5 days. The seizure detection algorithm achieved an overall sensitivity of 0.73, precision of 0.004 and F1 score of 0.01. Following human review, a precision of 0.83 was achieved, albeit with a sensitivity of 0.31, resulting in an F1 score of 0.45. Seizure detection was influenced by the seizure type (focal impaired awareness seizure (FIAS) or focal-to-bilateral tonic-clonic seizures (FBTCS)) and the presence of a distinct ictal pattern on behind-the-ear EEG, constituting mostly temporal lobe seizures. In a post-hoc subgroup analysis of seizures with a clear ictal EEG pattern and ictal tachycardia (as determined by the ground truth), a sensitivity of 0.74 was observed, whereas a sensitivity of 0.60 was achieved when there was visible electrographic seizure activity without tachycardia.

The overall performance of the Sensor Dot in focal seizure detection was limited in an unselected group of focal seizure types. Our findings suggest that the proposed approach, combining algorithm-based detection with human review, provides detection performance comparable with diary self-reporting, in terms of sensitivity, but with higher precision (0.83 vs. 0.60). Our findings suggest that this multimodal approach may be particularly beneficial for the subset of patients whose seizures are characterized by distinct ictal EEG patterns and tachycardia-features frequently seen in temporal lobe epilepsy. However, the prospective utility of these markers for patient selection requires further validation.

Wearable devices may help improve seizure detection in people with epilepsy, but few non-invasive options exist for detecting focal seizures. We studied a behind-the-ear wearable device (Sensor Dot) using automated analysis combined with expert review. Overall, this approach resulted in detection of seizures with low sensitivity but high precision. Post-hoc subgroup analysis suggested that for seizures characterized by a distinct ictal EEG pattern and tachycardia-features often seen in temporal lobe epilepsy-this multimodal approach achieved a sensitivity of 0.74. Further prospective validation is required to confirm if this performance can be replicated through pre-monitoring patient selection.