Modeling EEG-based Motor Imagery with Session to Session Online Adaptation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Subject-specific calibration is crucial for electroencephalography (EEG)-based Brain-Computer Interfaces (BCI) in motor imagery (MI) detection.
• Fixed calibration models use only initial calibration data, while progressive adaptive models incorporate data from ongoing rehabilitation sessions.
• Previous retrospective studies indicated superior performance of progressive adaptive models over fixed models.

Purpose of the Study:

• To deploy and evaluate an online progressive adaptation model within a BCI-based stroke rehabilitation system (nBETTER).
• To compare the performance of online progressive adaptation, light-weight adaptation, and fixed calibration models in a clinical trial setting.
• To confirm the effectiveness of progressive adaptive models for online EEG-based MI detection in stroke patients.

Main Methods:

• A clinical trial involving 11 stroke patients undergoing 18 rehabilitation sessions over 6 weeks using the nBETTER system.
• Retrospective analysis comparing three modeling strategies: fixed calibration, online progressive adaptation, and light-weight adaptation.
• EEG data was used to train and evaluate subject-specific models for MI detection.

Main Results:

• The online progressive adaptation model achieved a mean accuracy of 74.04%, and the light-weight adaptation model achieved 74.53%.
• The fixed calibration model had a lower mean accuracy of 68.17%.
• Statistical analysis (ANOVA) showed a significant difference (p < 0.00001) between the adaptation models and the fixed model, confirming the superiority of adaptive strategies.

Conclusions:

• The study confirms the effectiveness of progressive adaptive models for EEG-based BCI in detecting motor imagery in an online stroke rehabilitation setting.
• Online adaptation strategies significantly improve MI detection performance compared to traditional fixed calibration models.
• The nBETTER system demonstrates the successful implementation of adaptive BCI for enhanced stroke recovery.