A visual working memory dataset collection with bootstrap Independent Component Analysis for comparison of...

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Electroencephalography (EEG) is crucial for studying brain activity.
• Independent Component Analysis (ICA) is widely used for source separation in EEG data.
• The quality of ICA decompositions can be affected by data preprocessing and sample size.

Purpose of the Study:

• To present a comprehensive dataset of EEG recordings and associated ICA decompositions.
• To provide a benchmark for evaluating different EEG data preprocessing pipelines and ICA algorithms.
• To investigate the impact of dimension reduction techniques like Principal Component Analysis (PCA) on ICA quality.

Main Methods:

• Acquired 71-channel EEG data from 14 subjects performing a visual working memory task.
• Generated 150 Extended Infomax ICA decompositions using bootstrap resampling.
• Included ICA decompositions after PCA with varying variance retention (85%, 95%, 99%).
• Clustered independent components (ICs) within subjects and calculated a quality index (QIc).

Main Results:

• The dataset includes bootstrap ICA decompositions serving as benchmarks.
• Quality index (QIc) measures the stability of ICs to data resampling.
• PCA-based dimension reduction prior to ICA can reduce decomposition quality.
• Equivalent dipole positions are provided for compact brain sources.

Conclusions:

• The presented dataset facilitates the evaluation of EEG preprocessing and ICA methods.
• Understanding the influence of data characteristics on ICA is vital for accurate source localization.
• This resource aids researchers in optimizing their EEG analysis workflows.