IEEG.org EEG Dataset for Seizure Prediction Model Validation

87 seizures from 16 adult patients were used for external validation of a hybrid physics-informed graph neural network model. The dataset contains processed EEG features for predicting seizures, focusing on detecting chimera states. It was published by Masoud Amiri on figshare in April 2026.

Use Cases

• Validate seizure prediction model performance on 16 adult patients from IEEG.org, distinct from the pediatric training cohort.
• Analyze chimera state detection accuracy metrics (84.7%) derived from EEG phase locking values.
• Assess cross-patient generalization performance (79.8%) using the provided validation results.
• Examine false positive rate per hour (0.48) as a key clinical metric for model evaluation.

Strengths

• Contains 87 clinically annotated seizures for validation.
• Provides external validation results from a distinct patient cohort (16 adults) compared to the training data.

Limitations

• Dataset is very small at 5.5 KB, likely containing only summary validation metrics, not raw EEG signals.
• Limited to validation results for a specific model (HP-GNN), not a primary data source for new model training.

Provenance

Source

IEEG.org, an open intracranial EEG data repository.

Collection Method

Processed EEG data used for external validation of a published seizure prediction model.

Freshness

Last updated in April 2026.