Decoding of image properties from single-trial visual evoked potentials recorded by ultra-high-density EEG

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Electroencephalography (EEG) offers non-invasive, high temporal resolution for studying neuronal activity.
• Traditional EEG analysis relies on group-level statistics and trial averaging, limiting single-subject/trial insights.
• Advances in high-density EEG enable more granular investigation of brain responses.

Purpose of the Study:

• To combine ultra-high-density (uHD) EEG with cross-validated single-trial decoding for improved generalizability and reproducibility.
• To investigate the impact of spatial resolution and electrode density on decoding performance.
• To decode specific image properties from single-trial visual evoked potentials (VEPs).

Main Methods:

• Utilized a 512-channel uHD EEG system to record occipital lobe activity from participants viewing diverse images.
• Extracted image properties (contrast, hue, luminance, saturation, spatial frequency) and used VEPs for cross-validated regression analysis.
• Spatially subsampled data to assess the effect of electrode density on decoding accuracy.

Main Results:

• Successfully decoded image properties from single-trial VEPs, with contrast, saturation, and spatial frequency showing the highest decoding accuracy.
• Achieved a grand average decoding performance (Pearson's r) of 0.50 between predicted and actual image property scores.
• Demonstrated that increased electrode density significantly improves decoding performance, even below 10 mm inter-electrode distance.

Conclusions:

• Single-trial VEPs contain robust modulations by image properties, sufficient for accurate decoding.
• Ultra-high-density EEG significantly enhances the decoding of visual information compared to standard or subsampled configurations.
• High electrode density is crucial for improving the precision and reliability of brain-computer interfaces and neuroscience research.