User Identity Protection in EEG-Based Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Information Security

Background:

• Brain-computer interfaces (BCIs) leverage electroencephalogram (EEG) signals for device control.
• EEG signals contain sensitive private information, including user identity.
• Current BCI research prioritizes signal decoding accuracy over privacy.

Purpose of the Study:

• To identify and address the privacy risks associated with user identity leakage in EEG data.
• To develop novel techniques for creating identity-unlearnable EEG data.
• To ensure the primary functionality of BCIs is preserved while enhancing user privacy.

Main Methods:

• Proposing two distinct approaches to transform raw EEG data into an identity-unlearnable format.
• Evaluating the effectiveness of these methods across seven diverse EEG datasets and five BCI paradigms.
• Measuring the reduction in user identification accuracy after applying the proposed transformations.

Main Results:

• EEG-based BCI user identification accuracy decreased significantly, from an average of 70.01% to a maximum of 21.36%.
• The proposed methods successfully removed user-specific identity information from EEG data.
• The core performance of the BCI tasks remained largely unaffected by the privacy-preserving transformations.

Conclusions:

• User identity can be effectively learned from EEG data in BCIs, posing a significant privacy concern.
• The developed methods provide a robust solution for mitigating identity leakage in EEG-based BCIs.
• These findings pave the way for more secure and privacy-conscious BCI applications.