EEG-Based User Reaction Time Estimation Using Riemannian Geometry Features.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Riemannian geometry has shown success in brain-computer interface (BCI) classification.
• BCI regression problems are an important application area for BCIs.
• Traditional feature extraction methods for electroencephalogram (EEG) signals may not be optimal for regression tasks.

Purpose of the Study:

• To apply Riemannian geometry to BCI regression problems for the first time.
• To propose a novel feature extraction method for EEG-based BCI regression.
• To evaluate the performance of the proposed method in reaction time estimation.

Main Methods:

• A spatial filter was employed to enhance EEG signal quality and reduce dimensionality.
• Riemannian tangent space features were extracted from preprocessed EEG data.
• The approach was validated on EEG signals from a sustained-attention psychomotor vigilance task.

Main Results:

• The proposed tangent space features significantly improved reaction time estimation.
• Compared to powerband features, tangent space features reduced root mean square error by 4.30%-8.30%.
• Estimation correlation coefficients increased by 6.59%-11.13% with tangent space features.

Conclusions:

• Riemannian tangent space feature extraction is effective for EEG-based BCI regression.
• This method offers superior performance over traditional features for reaction time estimation.
• The findings open new avenues for applying Riemannian geometry in BCI research.