Harvard Electroencephalography Database: A comprehensive clinical electroencephalographic resource from four Boston...

Area of Science:

• Neuroscience
• Medical Informatics
• Artificial Intelligence

Background:

• Epilepsy research requires large, standardized datasets for advanced analysis.
• Existing electroencephalography (EEG) data resources are often fragmented or lack comprehensive clinical linkage.
• Privacy-preserving data sharing is crucial for advancing clinical neuroscience.

Purpose of the Study:

• To introduce the Harvard Electroencephalography Database (HEEDB), a novel, large-scale resource for epilepsy and clinical neuroscience research.
• To provide a deidentified and standardized EEG dataset that supports artificial intelligence (AI)-driven and reproducible research.
• To facilitate multimodal analysis by linking EEG data with extensive clinical information.

Main Methods:

• Aggregated over 280,000 EEG recordings from 108,000+ patients across four Harvard-affiliated hospitals.
• Harmonized data using the Brain Imaging Data Structure (BIDS) and hosted on the Brain Data Science Platform.
• Linked EEG data with deidentified clinical notes, ICD-10 codes, medications, and EEG reports, adhering to HIPAA Safe Harbor standards.

Main Results:

• The database encompasses routine, epilepsy monitoring unit, and intensive care unit EEGs for all age groups.
• 73% of EEGs are linked to deidentified clinical reports, with 96% of these matched to recordings.
• Findings were extracted using expert curation, regular expressions, and medical natural language processing (NLP) models, enabling multimodal analysis with diagnoses, medications, and hospital course data.

Conclusions:

• HEEDB addresses a significant need for accessible EEG data in epilepsy research.
• The database enables large-scale, privacy-compliant, and clinically relevant analyses.
• It accelerates the development of diagnostic tools, enhances machine learning training datasets, and promotes data sharing in line with FAIR and NIH policies.