Video-based automatic seizure detection in pharmacoresistant epilepsy: A prospective exploratory study

The objective of this study was to evaluate the diagnostic yield and clinical utility of an automated AI video-based seizure detection device, Nelli®, (SDD) in pharmacoresistant epilepsy patients. The SDD captures and automatically classifies nocturnal motor behavior suggestive of epileptic seizures or non-epileptic motor behavior of potential clinical value.

Patients with focal epilepsy and pharmacoresistance referred for inpatient long-term video-EEG monitoring were prospectively recruited. Participants were monitored in their home at night with the SDD for a median of 15.5 nights. Captured video recordings were analyzed by clinical experts and each SDD-registration session was classified as diagnostic or not. Clinical utility for each participant was assessed from pre-specified utility measures. The outcome measures were compared between major focal motor and subtle focal motor seizures.

One SDD-registration session in each of the 20 participants was performed and analyzed. Video recordings were captured in 18 sessions. Diagnostic yield was found in 11 registration sessions (55.0 %) and clinical utility in 8 registration sessions (40.0 %). No significant difference was found between the AI-algorithm classification and clinical experts' consensus assessment of captured video recordings as epileptic or not. Positive predictive value was 81.8 % for registration sessions containing video recordings classified as epileptic seizures. The diagnostic yield and clinical utility were significantly higher among major focal motor seizures (81.8 % and 63.6 %) compared to subtle focal motor seizures.

The SDD is useful to evaluate patients with pharmacoresistant epilepsy and major focal motor seizures (hyperkinetic, tonic, clonic, focal to bilateral tonic-clonic seizures); it may facilitate the diagnostic process in patients referred for long-term inpatient video-EEG evaluation and beneficially change anti-seizure treatments. The SDD provided accurate classification of major focal motor seizures as epileptic, or non-epileptic, and may serve as a useful diagnostic tool to distinguish epileptic and non-epileptic episodic events with a prominent motor component.