As above, so below? Towards understanding inverse models in BCI

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCI) translate brain activity into computer commands.
• Electroencephalography (EEG)-based BCIs face challenges with spatially localized cortical activity mixed in surface signals.

Purpose of the Study:

• Investigate if reconstructing brain activity sources in the cortical volume enhances BCI accuracy.
• Compare physiology-driven source reconstruction with data-driven machine learning approaches.

Main Methods:

• Framework: Common linear dictionary for both approaches.
• Analysis: Contrasting parameter estimation in source reconstruction and machine learning.
• Consideration: Impact of source reconstruction on information loss, feature selection, and EEG nonstationarity.

Main Results:

• Approaches differ primarily in parameter estimation.
• Physiological source reconstruction can improve BCI accuracy, especially when machine learning is absent or suboptimal.
• Challenges like information loss and nonstationarity may persist in reconstructed data, necessitating data-driven techniques.

Conclusions:

• Source reconstruction and machine learning for EEG data representation are related.
• Understanding these relationships aids in applying, comparing, and improving BCI techniques.
• Combined approaches may offer the most robust solutions for EEG-based BCI accuracy.