Subject-Dependent Artifact Removal for Enhancing Motor Imagery Classifier Performance under Poor Skills

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG)-based Brain-Computer Interface (BCI) development faces challenges due to low Signal-to-Noise Ratio (SNR).
• Non-stationary, nonlinear EEG signals, low spatial resolution, and subject variability hinder feature extraction for motor imagery (MI).
• Individuals with poor motor skills struggle with MI tasks in low SNR environments, impacting BCI efficacy.

Purpose of the Study:

• To propose and evaluate a subject-dependent preprocessing approach for enhancing EEG-based MI classification.
• To address challenges posed by low SNR, signal artifacts, and inter-subject variability in BCI systems.
• To improve BCI performance for subjects with difficulties in motor imagery tasks.

Main Methods:

• Implementation of a subject-dependent preprocessing strategy incorporating Surface Laplacian Filtering and Independent Component Analysis (ICA).
• Application of power- and phase-based functional connectivity measures for extracting discriminant features.
• Mitigation of electrooculography (EOG) and volume-conduction EEG artifacts.

Main Results:

• The proposed Subject-dependent Artifact Removal (SD-AR) approach significantly improves MI classification performance.
• Enhanced performance is particularly notable in subjects exhibiting poorer motor skills.
• Artifacts such as EOG and volume-conduction effects were effectively mitigated, aiding classification.

Conclusions:

• Subject-dependent preprocessing, including artifact removal and functional connectivity analysis, is crucial for robust EEG-based BCIs.
• The SD-AR method offers a viable solution for improving BCI accessibility and performance for a wider range of users.
• This strategy facilitates the use of straightforward linear classifiers by improving feature quality.