Covariation Informed Graph Slepians for Motor Imagery Decoding

Area of Science:

• Signal Processing
• Neuroscience
• Machine Learning

Background:

• Non-invasive electroencephalography (EEG) data is complex and often irregularly sampled.
• Graph signal processing (GSP) offers tools for analyzing such data.
• Motor imagery (MI) decoding is crucial for brain-computer interfaces (BCIs).

Purpose of the Study:

• To exploit Graph Slepian functions for robust decoding of motor imagery (MI) brain activity.
• To introduce a data-driven, subject-specific design for Graph Slepian functions using contrastive learning.
• To enhance spatial filtering for improved MI decoding accuracy.

Main Methods:

• Utilized Graph Slepian functions, building upon the graph Fourier transform (GFT).
• Developed a contrastive learning pipeline for subject-specific Graph Slepian function design.
• Integrated these functions as spatial filters in a motor imagery decoding scheme using a support vector machine (SVM).

Main Results:

• The proposed method demonstrated superior performance against popular alternatives in MI decoding on two public datasets.
• Graph Slepian functions enhanced the information from multichannel EEG signals, relating to the participant's intention.
• The technique showed computational efficiency due to simple matrix operations.

Conclusions:

• The data-driven design of Graph Slepian functions provides effective spatial filtering for MI decoding.
• This approach offers a robust and computationally efficient solution for brain-computer interfaces.
• The method holds potential for real-time applications in BCI systems.