Neonatal EEG Recordings with Expert Seizure Annotations

Helsinki University Hospital collected multi-channel EEG recordings from 79 term neonates admitted to its NICU, with a median recording duration of 74 minutes. Three human experts annotated the EEGs for seizures, with an average of 460 seizure events annotated per expert, resulting in a consensus of seizures in 39 neonates and seizure-free status in 22. This dataset serves as a reference set for analyzing seizure characteristics and developing automated detection methods.

Use Cases

• Developing automated neonatal seizure detection models based on multi-channel EEG signals.
• Analyzing inter-observer agreement in EEG interpretation using annotations from three experts.
• Studying the temporal and spatial characteristics of seizures from different pathologies.
• Evaluating the effects of anti-seizure medication on EEG patterns in neonates.
• Creating a benchmark reference set for validating new EEG analysis techniques.

Strengths

• Contains expert annotations from three independent reviewers, supporting reliability analysis.
• Includes EEG from 79 term neonates, with detailed consensus labels for 39 with seizures and 22 without.
• Provides multi-channel recordings with a median duration of 74 minutes (IQR: 64-96 minutes), offering substantial time-series data.

Limitations

• Key metadata such as specific column names, file formats, total file size, and last update date are unavailable from the sources.
• The dataset scale is limited to a single hospital cohort (Helsinki University Hospital), which may affect generalizability.
• The description is repeated verbatim across multiple platform entries, suggesting potential lack of detailed, varied documentation.

Provenance

Source

Nathan J. Stevenson, University of Helsinki, Neonatal Intensive Care Unit at Helsinki University Hospital.

Collection Method

Multi-channel EEG recordings from human neonates, visually interpreted and annotated by three experts.

Geography

Helsinki, Finland.