Does topological data analysis work for EEG-based brain-computer interfaces?

Area of Science:

• Neuroscience
• Data Science
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) enable machine communication via brain activity (e.g., electroencephalography - EEG).
• Topological Data Analysis (TDA) extracts shape-based features from data, showing potential in diverse applications.
• Systematic evaluation of TDA for EEG-based BCIs is limited.

Purpose of the Study:

• To systematically evaluate the efficacy of TDA for EEG-based BCI.
• To investigate if topological features of EEG dynamics differ across mental states.
• To establish a benchmark comparing TDA with established BCI methods.

Main Methods:

• EEG data from three public datasets (motor-imagery and mental workload) were analyzed.
• Topological features, particularly persistence diagrams, were extracted from EEG dynamics.
• Feature vectors were generated and classified using linear and non-linear classifiers.

Main Results:

• TDA demonstrated significantly lower performance in intra-subject classification.
• TDA achieved comparable or superior performance in inter-subject classification.
• Persistence features consistently outperformed other topological features and showed a theoretical and experimental link to spectral power.

Conclusions:

• This study is the first to evaluate TDA for both intra- and inter-subject classification across various BCI datasets.
• TDA, especially persistence, offers a valuable approach for inter-subject BCI classification.
• New insights into the relationship between persistence and classical EEG features were provided.