EEG datasets for seizure detection and prediction- A review

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Machine learning (ML) models are developed using electroencephalogram (EEG) datasets from epilepsy patients for seizure detection and prediction.
• Publicly available EEG datasets exhibit diverse formats and structures, lacking standardized guidelines.
• This heterogeneity hinders the generatability, generalizability, and reproducibility of ML-based seizure prediction studies.

Purpose of the Study:

• To compile and compare characteristics of publicly available EEG datasets used for seizure detection and prediction algorithms.
• To investigate the advantages and limitations of various EEG dataset characteristics.
• To provide guidelines for selecting appropriate EEG datasets for developing reproducible and generalizable seizure prediction algorithms.

Main Methods:

• A narrative review approach was employed.
• Publicly available EEG datasets commonly used for seizure detection and prediction were identified and compared.
• The impact of dataset characteristics on ML model performance, technique selection, and preprocessing was analyzed.

Main Results:

• Seventeen unique characteristics differentiating EEG datasets were identified.
• The influence of specific dataset features on study outcomes, ML technique choice, and preprocessing requirements was examined.
• Variations in dataset characteristics significantly affect the performance and applicability of seizure detection and prediction algorithms.

Conclusions:

• The heterogeneity of EEG datasets poses challenges for developing robust seizure detection and prediction algorithms.
• Guidelines are provided for clinicians and scientists to select suitable EEG datasets.
• This review aims to enhance the reproducibility, generalizability, and effectiveness of ML-based epilepsy research.