Semiautomated classification of nocturnal seizures using video recordings
Peltola J, Basnyat P, Armand Larsen S, Østerkjaerhuus T, Vinding Merinder T, Terney D, Beniczky S.
Epilepsia. 2022;00:1–7.
https://doi.org/10.1111/epi.17207
Hot Off The Press – This new publication by Peltola et al. 2022 examines the precision of a semi automated classification of nocturnal motor seizures, using a hybrid system (#Nelli®). The system combines #video-based automated #seizure detection with an artificial intelligence based algorithm that automatically classifies seizure type. The identified episodes and the seizure classifications are to be reviewed by clinical experts.
Highlights
• The hybrid system significantly reduced the duration of episodes to be reviewed by clinical experts.
• All tonic-clonic and clonic seizures, as well as the majority of focal motor seizures, were accurately classified by the algorithm.
• An agreement coefficient of 0.33 was identified between the hybrid system and the gold standard long-term video-EEG (#VEEG) monitoring.
Patients (N=40; median age: 15 years) with nocturnal motor seizures were prospectively recruited following gold standard long-term VEEG monitoring. The reduction in workload and the seizure classification of the hybrid system was compared to the gold standard VEEG monitoring.
All seizure-related events were detected by the hybrid system and the duration of episodes to be reviewed by the clinical experts was distinctly reduced from 1874 hours to 259 hours. The algorithm accurately identified 82% of focal motor seizures and 100% of clonic and tonic-clonic seizures. However, seizures with discrete motor symptoms were less accurately detected. The agreement coefficient for seizure classification between the Nelli® system and the gold standard was 0.33 (95% CI: 0.20-0.47).
Despite the difficulty in identifying seizures with subtle motor symptoms, semi automated seizure detection and classification may facilitate the detection of nocturnal major motor seizures by significantly reducing the time required for reviewing video-recordings.