Performance metrics for online seizure prediction.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Online seizure prediction from intracranial electroencephalography (iEEG) signals is an active research area.
• Existing studies often focus on prediction algorithms and performance metrics.
• System parameters, such as prediction period and horizon, are frequently set without rigorous justification.

Purpose of the Study:

• To investigate the impact of system parameter specification on online seizure prediction performance.
• To emphasize the importance of data-driven characterization of lead seizures.
• To promote standardized and meaningful comparisons of seizure prediction algorithms.

Main Methods:

• Analysis of system parameters including prediction period and prediction horizon.
• Data-driven characterization of lead seizures.
• Evaluation using both synthetic and real-life iEEG datasets.

Main Results:

• Prediction performance is significantly influenced by the choice of system parameters.
• Ad hoc parameter setting can lead to misleading conclusions about algorithm efficacy.
• Proper specification is essential for reliable online seizure prediction.

Conclusions:

• Meaningful comparison of seizure prediction methods necessitates careful consideration and specification of system parameters.
• Standardized parameter settings are crucial for advancing the field of iEEG-based seizure prediction.
• Future research should prioritize the systematic evaluation of parameter effects on prediction performance.