Decoding Multi-Class Motor Imagery and Motor Execution Tasks Using Riemannian Geometry Algorithms on Large EEG...

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG)-based brain-computer interfaces (BCIs) face challenges with signal noise and nonstationarity.
• Current state-of-the-art methods struggle with classification accuracy on large BCI datasets.
• Riemannian geometry decoding algorithms offer a novel approach to overcome these limitations.

Purpose of the Study:

• To evaluate the performance of a novel Riemannian geometry decoding algorithm implementation.
• To assess algorithm performance on large-scale, multi-subject BCI datasets.
• To compare different adaptation strategies for decoding motor execution and motor imagery signals.

Main Methods:

• Applied several Riemannian geometry decoding algorithms to a large offline dataset.
• Utilized four adaptation strategies: baseline, rebias, supervised, and unsupervised.
• Tested on a dataset of 109 subjects with four-class bilateral/unilateral motor imagery and execution using 64 and 29 electrodes.

Main Results:

• The baseline minimum distance to Riemannian mean strategy yielded the best classification accuracy.
• Achieved mean accuracies of up to 81.5% for motor execution.
• Achieved mean accuracies of up to 76.4% for motor imagery.

Conclusions:

• Riemannian geometry decoding algorithms are effective for classifying EEG trials in BCI.
• The baseline minimum distance to Riemannian mean strategy is particularly effective for large datasets.
• Accurate EEG trial classification is crucial for developing successful BCI applications for device control.