A Review on Signal Processing Approaches to Reduce Calibration Time in EEG-Based Brain-Computer Interface

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) offer direct device communication.
• EEG signal challenges include noise, artifacts, and non-stationarity, leading to high inter-subject/session variability.
• This variability necessitates lengthy calibration for subject-specific classifiers.

Purpose of the Study:

• To review and analyze signal processing approaches for reducing BCI calibration time.
• To explore the effectiveness of transfer learning (TL) and semi-supervised learning (SSL) in EEG-based BCIs.
• To investigate the combined application of TL and SSL for improved BCI performance.

Main Methods:

• Review of existing signal processing techniques, focusing on TL and SSL.
• Explanation of cross-subject, cross-session, cross-task, and cross-device TL concepts.
• Description of SSL utilizing both labeled and unlabeled target subject data.
• Examination of combined TL and SSL strategies.

Main Results:

• TL and SSL effectively utilize available samples to achieve good classification performance.
• Cross-subject TL transfers labeled data from source subjects to aid target subjects.
• SSL leverages unlabeled data from the target subject, complementing TL.
• Combined TL and SSL approaches show synergistic benefits.

Conclusions:

• TL and SSL are promising methods for reducing BCI calibration time and improving classification accuracy.
• These techniques mitigate issues arising from EEG signal variability.
• Future research directions include further optimization and integration of these learning paradigms.