Content Differences: Variations in acquisition equipment, electrode configurations, and experimental paradigms.
Metadata Inconsistencies: Divergent recording standards for channel names, sampling rates, and event labels.
Format Fragmentation: Coexistence of EDF, GDF, MAT, CSV, TXT, and EEGLAB .set formats, making standardized pipelines difficult to reuse.
UnifiedDataset via raw data paths, existing Locator files, or by merging multiple datasets, allowing the tool to enforce batch processing on a standardized interface.UnifiedDataset: A Python class serving as the unified interface for all dataset operations (loading, merging, batch processing, and exporting).
Locator: A Pandas DataFrame-style table that records critical metadata (file path, domain tag, channel config, sampling rate, duration, completeness) for each EEG file. This design allows researchers to review and correct metadata without altering the original source files.
EEG Parser Module: Supports parsing of various standard and non-standard formats. It utilizes both conventional readers and an LLM-boosted engine to handle complex header information.
Correction Module: Provides a spreadsheet-like interface for reviewing and modifying annotations. It supports dataset diagnosis (generating reports) and visualization (magnitude-frequency curves, channel correlations) to ensure data integrity.
Batch Processing Module: Enables customized pipelines for data cleaning (denoising, ICA, quality scoring), channel alignment, resampling, normalization, and epoch extraction.
Large Language Model (LLM) Boost Module: Leverages LLMs (e.g., ChatGPT, DeepSeek) to automatically extract metadata from unstructured text files (e.g., READMEs), significantly improving the flexibility of handling poorly documented datasets.
Source code for over 10 Large EEG Models.
Extensive evaluation and analysis.
Export configurations for over 50 datasets.
GitHub Repository: https://github.com/Baizhige/EEGUnity
Paper DOI: https://doi.org/10.1109/TNSRE.2025.3565158
Comments & Questions (0)
No comments yet
Be the first to comment!