Improving session-to-session transfer performance of motor imagery-based BCI using Adaptive Extreme Learning Machine

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Session-to-session non-stationarity in electroencephalograph (EEG) data poses a significant challenge for brain-computer interface (BCI) system performance.
• Adaptive methods offer a promising approach to mitigate the impact of this non-stationarity.
• Existing non-adaptive classifiers may struggle with the temporal variability inherent in EEG signals.

Purpose of the Study:

• To propose and evaluate an adaptive extreme learning machine (AELM) for enhancing EEG-based BCI systems.
• To address the challenge of session-to-session non-stationarity in EEG data.
• To compare the performance of the AELM against non-adaptive classifiers.

Main Methods:

• An adaptive extreme learning machine (AELM) was developed to update initial classifiers using EEG data from evaluation sessions.
• Common Spatial Pattern (CSP) algorithm was employed for extracting discriminative features from motor imagery EEG data.
• The AELM's effectiveness was validated on motor imagery data from 12 healthy subjects across calibration and evaluation sessions.

Main Results:

• The proposed AELM demonstrated significantly superior performance compared to non-adaptive Extreme Learning Machine (ELM) and Support Vector Machine (SVM) classifiers (p=0.03).
• Accumulating evaluation session data for classifier adaptation led to significant performance improvements (p=0.001).
• The AELM effectively addressed EEG signal non-stationarity in an online BCI context.

Conclusions:

• The adaptive extreme learning machine (AELM) is an effective method for overcoming session-to-session non-stationarity in EEG data for BCI applications.
• Adaptive strategies, particularly the accumulation of new data, are crucial for maintaining and improving BCI performance over time.
• The proposed AELM offers a robust solution for developing more reliable and accurate online BCI systems.