Electroencephalography (EEG) is widely used clinically and in research, including AI-driven applications for cognitive state analysis and neurological disorder detection, such as epilepsy. However, automated seizure detection faces challenges, including inconsistent windowing, timestamp misalignment, label-based signal segmentation, and unstructured large-scale EEG data—especially critical in event-driven settings. To address these, we introduce Meta-EEGs, a structured, domain-agnostic EEG representation for temporally labelled tasks. Meta-EEGs provide consistent windowing with precise time alignment, enable event-based segmentation based on annotations or class labels, and organise raw EEG recordings into a simpler, relatively reduced in volume format suitable for AI model input. They also support the creation and management of hierarchical EEG datasets, currently lacking in the field. As case studies, we applied Meta-EEGs to the CHB-MIT and Siena Scalp EEG Databases, generating structured datasets that are publicly available on Figshare and have been downloaded over 2,000 times since 2022. The working code is accessible on GitHub. Main features and applications: Provides consistent window definition, timestamp alignment, signal segmentation, and standardised structuring for large-scale EEG studies. Releases two fully annotated, reduced in volume datasets for automated seizure detection, supporting reproducible and generalisable analysis. Enables AI model development for seizure detection, event classification, patient-specific and cross-patient analysis, and hierarchical EEG tasks without repeated initial preprocessing.
Handa et al. (Mon,) studied this question.